Alaamer/medium-articles-posts-with-content

0

null

0

25ec8038ee01

2018-02-26

2018-02-26 23:06:35

2018-02-27

2018-02-27 01:36:23

16

false

en

2018-02-27

2018-02-27 10:58:50

8

14de29a40f47

4.565094

4

0

Where we learn that it takes one neuron to solve multilinear regressions and logistic regressions, which I thought were pretty advanced…

3

All the single neurons Where we learn that it takes one neuron to solve multilinear regressions and logistic regressions, which I thought were pretty advanced back in university A neuron in machine learning looks like this if you Google it: Neuron representations, various artists Here is my own representation of a single neuron — a bit more wavy, unpractical when it comes to drawing networks, hopefully not as bad for education purposes. My own representation of a single neuron A neuron has a body, dendrons to the left connecting to that body, synapses that are receptors for input coming to the neuron; to its right, the neuron has an axon, and an axon terminal (or multiple ones) that transfer the neuron’s information to the next one. The axon terminals connect to synapses. Messages at the synapses are numbers, that are amplified or reduced as they pass through the dendrons, in proportion to the dendrons weights, noted w, or b in the case of the bias dendron (the bias dendron is always plugged to an incoming signal equal to 1). If a weight w is equal to 0, then the dendron is as good as dead — it passes no information at all. Signals from each dendron are summed in the neuron’s body, as the weighted input z. The weighted input z then passes to the axon where it is optionally touched by an activation function σ. The activation function can be the Identity function (~the activation function does nothing to the weighted input), the Sigmoid function, the Rectified Linear function, but really, can be any function needed for a given application. What we get after the activation function is a¹, the neuron’s activation state. Sample activation functions according to Wikipedia. Linear, non-linear, monotonous are not, symmetric or not… Activation functions can even be functions of not just the weighted input z, but of other parameters like weighted inputs in other neurons — see for example the softmax activation function. From a place where I can actually use MathJax, here is what I have to say about the forward pass for our single neuron: dot product is a scalar product, by opposition to the Hadamard product (or element-wise product) also used in machine learning Putting single neurons to work So far our neuron takes input and produces an output, but this is it. If the weights and bias are set in to certain values, the output can be useful — otherwise we just have a random number generator. In order to make the neuron useful we need to train it. So we need a trainer for that neuron, that will (i) be able to evaluate the neuron’s performance, and (ii) create a learning procedure so the neuron performs better. We use a cost function to define (i), and the back-propagation algorithm to address (ii). So what now? Well, let’s replace C and σ in the equations above with actual functions and we will start to see how those formulas are just generalizations of classic regression problems. Single neurons solve (multi)linear regressions The goal of linear regressions is to find, given a dependent variable and n explanatory variables, a line, expressed as a function of the explanatory variables such that the distance between the dependent variable and that line is minimal. A simple example of a linear regression is: Here is exactly the same program, but written using the conventions we have used to describe our neuron so far: in which we recognize a single neuron net problem, where the axon activation function σ is the identity function, and C is the squared l2 (or Euclidian) norm. This neuron (Id activation function) with an squared l2 norm solves 3-parameter linear regressions Single neurons solve logistic regressions Similarly, logistic regressions can be expressed as single neuron network problems. Consider the stated goal of a logistic regression: which is equivalent to the single neuron net program: This neuron solves classification problems with 3 input parameters In summary, using a single neuron, logistic vs linear regressions are simply a matter of flipping the activation and cost functions of the neuron: A Python implementation and real regressions solved by a single neuron Here is how the single neurons fared on four different regression exercises: Black and white dots are inputs, blue lines are the neuron’s answers. Left: logistic regression exercises (separate the black and white dots), right: linear regerssions (find the line closest to all points). Noise was introduced to prevent the neuron from getting a perfect answer. I increased the order of the datasets to allow the neuron to find polynomial patterns. The two logistic regressions were solved by a single neuron with 10 input dendrons, a sigmoid activation function and a cross entropy cost function The two linear regressions were solved by single neurons with 5 to 10 input dendrons, a identity activation function and a squared l2 cost function The core code to reproduce those results fits is reproduced below, and you can find the full code on GitHub. Nothing too crazy, and just like that we learnt that single neurons are incredibly powerful. The next question I am asking myself is why the rectified linear activation function seems so prevalent in actual neural networks, if the Identity and the sigmoid functions are so strong. In particular, rectified linear activations cause neurons to “die” (= output 0) if their weighted input is negative — what value could possibly be gained from creating neurons that exhibit that sort of behavior? Nothing useful in the context of single neurons, but possibly something good when it comes to neural networks, as we shall learn. Main piece of code to create the single neurons that produced the charts above. Full code at https://github.com/cedricbellet/1neuron_net.

All the single neurons

16

all-the-single-neurons-14de29a40f47

2018-02-28

2018-02-28 10:57:39

https://medium.com/s/story/all-the-single-neurons-14de29a40f47

false

799

Knowledge by bits

null

Biffures

null

biffures

null

cedricbellet

Machine Learning

machine-learning

Machine Learning

51,320

Cédric Bellet

null

e5d1e0544198

cedricbellet

191

78

20,181,104

null

0

null

0

null

2018-02-01

2018-02-01 11:27:14

2018-02-14

2018-02-14 15:34:29

1

false

en

2018-02-14

2018-02-14 15:50:17

1

14de6e9ec122

0.267925

0

1

0

Click — — — — — — — — → here!

5

Disruptive AI in the Recruitment Blockchain 🤖 Click — — — — — — — — → here!

Disruptive AI in the Recruitment Blockchain 🤖

0

disruptive-ai-in-the-recruitment-blockchain-14de6e9ec122

2018-02-14

2018-02-14 15:50:19

https://medium.com/s/story/disruptive-ai-in-the-recruitment-blockchain-14de6e9ec122

false

18

null

Blockchain

blockchain

Blockchain

265,164

Adriaan van der Heijden

I like tech and more. Blogging about Sourcing & Recruitment and possibly the occasional short story.

e4a8bddfafac

AdvdHeijden

59

189

20,181,104

null

0

null

0

null

2018-03-26

2018-03-26 18:02:13

2018-04-02

2018-04-02 00:45:08

26

false

en

2018-04-02

2018-04-02 17:57:30

11

14e12404fe73

10.978302

2

0

In this article, I will share what I taught about computer science and software engineering job to the 7 to 9th grade students.

5

What middle school students need to know about computer science and software engineering job I got an opportunity to do a computer science workshop at Community Montessori School, Tampa. In this article, I will share what I taught about computer science and software engineering job to the 7 to 9th grade students. How I prepared? I did the workshop in multiple sessions. I believe that if we can get kids excited and inspired, they will learn anything. So for the first session, I designed content to make them curious about computers, and to enlighten them on fascinating things about software engineering job and future of computer science. In the second session I introduced them algorithms and in the third session I taught programming using python. This would help them go through the complete process of figuring out a algorithm for the problem and then writing a program for it. This article only covers the first session of the presentation. You can read about second session and third session in follow up articles. Presentation I used lot of pictures and analogies to make them easily understand the concepts. I found awesome presentation tool gitpitch to create my slides. To make presentation engaging, I asked quizzes. Quizzes also helped me to check their understanding and increased their confidence as there were able to answer it correctly. I used slide fragment feature to show just the questions, waited for student’s response and then revealed the answers. The slides are embedded below. Presentation Slides Transcript Here is the complete transcript of the talk. Are Computers Smart? Computers are really stupid. But it is very fast. Humans are smart, but slow. Computers can tirelessly do repetitive tasks whereas humans will get exhausted . In fact, the hardware inside the computer doesn’t know anything other than 0’s and 1’s. All the videos, music and textual information you see in computers are stored as just 0’s and 1’s. Why the Digital World has only 0 & 1 ? All the hardware is made from wonderful semiconductor material silicon. It can easily change state between stop conducting ( 0 ) and start conducting ( 1 ) electricity through them. Silicon is used to make a tiny device called transistor which stores one bit of 0 or 1 — referred as binary. What’s inside Computers? Computers are made of billions of tiny transistors which can be programmed to flip states between 0 and 1 using electrical signals. Just like billions of neurons fire in our Human brain based on visual, auditory or sensory signal, Transistors inside computers switches states depending on input signals. Why computes is a Genius Invention? It will be super expensive, if we had to build a new computer for every different thing we want to do. Instead, we invented a general purpose computer containing a assembly of transistors that can do different things, depending on which transistors are switched on and off. Quiz: Say True or False Computers can understand only English — false Computers could do anything from adding two numbers to flying an airplane — true What is a Computer? Computer is a electronic machine that can be programmed to do zillions of different tasks. Components of a Computer A computer CPU has two major components, the processor which does all the computational work and the memory which stores all the information. Processor is like a heart in our human body which keeps beating all the time and memory is like our brain that remembers everything. Input devices are used to feed information to computer and output devices help computers to communicate back to us. Eyes and Ears in our body are examples of input devices as they take in visual and auditory information. Our hands and legs are like output devices since they can produce actions based on input. Can you tell some examples Input and Output devices? Can same device be used for both Input and Output? Like our mouth for example it can take food in and as well used to speak information out. Yes! A touch screen is good example of same device used as input and output. That’s whey touch screen devices are very easy to use. What Does GB or MB Mean in Phone or Computers? Bytes or Gigabytes is unit for measuring memory just like pounds for weight, ounces for liquid etc. Quiz: Say True or False Computers can do everything by itself — False. How Do you tell Computers to do something? We need to provide a sequence of instructions telling a computer to what to do which is called as Algorithms Example Algorithm to tell if temperature is Below or Above the Freezing Point. How Transistors Execute Algorithms? Transistors are combined to create logic gates that can perform logical operations like AND, OR, NOT. Every algorithm, no matter how complex, can be reduced to just these three operations: AND, OR, and NOT. Algorithm logic gates example — if a fever can be caused by influenza or malaria, and you should take Tylenol for a fever and a headache. Algorithms can solve complex problems Algorithms can be combined together to solve complex problems. Just like simple mathematical operations are combined to produce complex mathematical formulas. Algorithms fly the airplanes, forecast weather, play games, tell us driving directions and so on. Quiz: Match Profession to Work Answers: Match Profession to Work What is Programming? Programming is the act of taking complicated human ideas and breaking them down into simple algorithms that computers can understand and follow. A Programmer creates algorithms and codes them up in human friendly language like Java or Python. What is Software? Software is collection of programs that are written in language humans can understand and then they are converted into binary that computers can execute them. Software run every device — computers, smart phones, XBox, Billing Machines Quiz: Say True or False Without Software the hardware devices would be idle — true. Why Software Industry is at Rise Today? Six decades into the computer revolution, four decades since the invention of the microprocessor, and two decades into the rise of the modern Internet, all of the technology required to transform industries through software finally works and can be widely delivered at global scale. Every industry is Revolutionized by Software Quiz: Say True or False If you learn programming, you can apply the skills in any domain you choose. — True Why Being A Software Developer is Great? Drives Creativity and Innovation It’s very creative profession as much like a musician or a painter. Programming is fundamentally about creating solutions to problems. Make People’s Life Better You will solve problems to make people’s life better. For eg. It was very frustrating and also expensive to book a taxi cab for a ride. Uber made it easy using software solution. So now using Uber is not only easier, its even cheaper to use Uber than owning a car in some cities like SFO. Impact Millions of People quickly Telephone took 75 years, Internet took 4 years, however Angry Birds took only days! to reach 50 Million users. Collaborative Software engineering is a team sport and you work together to build amazing things. Future Proof As per US-BLS projections, computing will be the safest STEM career options for the foreseeable future. Pays Well Because of high demand, Software engineers are paid well. Software engineers can create lot of value with no cost of raw materials. All you need is your time and a computer. Work from Anywhere With the internet, you can literally work from any where in the world. Quiz: Which one of these a Software Engineer Doesn’t Do? a. Makes people’s life better. b. Solves complex problems. c. Sells computers to customers. d. Works together in a team. Answer: c What Traits should I have to become Software Engineer? Software Engineering is not for every one, you should have some specific traits or mindset to be a good software engineer. Here are some: Love Making things for Other People to Use You Feel happy when people use the product you created Enjoy Solving Puzzles You are passionate about solving complex problems. I love the process of getting from a unknown state of a problem to a final solution. The Ah-ha moment I get while figuring out a solution to a problem is what keeps me working. Like Experimenting and Research You won’t always get it working in first try. You will try something, if it doesn’t work you have to scratch that off and try different approach. So you should love experimenting All the information to solve your problem won’t be in front of you. You search the web, read the documentation, refer the books to come up with a solution. So you should like doing research. You want to make things easier in life Microsoft’s founder Bill Gates said He would always “hire a lazy person to do a difficult job” . Because A lazy person will find an easy way to do it. If you hate doing boring or mundane tasks and can come up with smart solutions to make things easier, you will be welcomed in Software Industry. You Have Interest in Both Arts and Engineering Software engineering utilizes right mix of both Artistic and Engineering skills. Writing a program is an art like writing a poem at the same time it needs applying logical skills to solve the problem. Like in Poem you have to adhere to the constraints of language like grammar, vocabulary, you have to conform to the constraints of programming language in programming. According to Steve Jobs: The greatest innovations come from the intersection of technology and arts. To emphasize this, Steve Jobs during his product launches would conclude with a slide, projected on the screen behind him, of street signs showing the intersection of the Liberal Arts and Technology. You are a curious learner Last but not the least, You need to continuously learn new things and keep up ever changing technology. Quiz: Say True or False A Software engineer can stop learning after college. — False Learning Arts will help in Software engineering job. — True What’s Fascinating about computers Today? Before first industrial revolution, Humans had to do everything manually and solely rely their physical strength. After invention of steam, first industrial revolution helped us automate difficult jobs. Still working with and distributing steam is hard. After invention of Electricity, the second industrial revolution simplified lot of the laborious work. After invention of computers, the third industrial revolution automated lot of the work and made most of the things possible with click of the button. Now we are going through forth industrial revolution in which we are heading towards making machines to learn and creating artificial intelligent systems. What is Machine learning? Machine learning makes computers to learn skills that humans can’t explain. It figures out Algorithm by analyzing large set of sample data. With machine learning, computers write their own programs, so we don’t have to. Difference Between Traditional Programming and Machine learning We can think of machine learning as inverse of programming just like Square root is inverse of Square. Quiz: Say True or False Can Computer learn human skills? — True Applications of Machine Learning Computers have learned to recognize our face, our voice, handwriting , etc. AutoDraw Fast drawing for everyone. AutoDraw pairs machine learning with drawings from talented artists to help you draw stuff…www.autodraw.com AutoDraw uses the magic of machine learning with drawings from talented artists to help everyone create anything visual, fast. ( I demoed AutoDraw to show how I can easily draw hard pictures, It was fun) Robots are Becoming our Assistants iRobot can find it’s way through your home and clean. It is smart enough to sense the surroundings for eg. it doesn’t fall when it finds stairs. Siri , Alexa, Google Now are able to answer our questions just like humans. Computers beat world chess champion IBM’s Chess computer Deep Blue beat world Chess Champion Garry Kasparov in 1997. So we are able to program computers to beat us! AlphaGo Beats World Go Champion Lee Sedol Programming chess is easy because it’s completely based on logic. But Go is very complex game as it need to learn millions of board positions. It won’t be possible without machine learning. The Supercomputer IBM Watson Watson won the first prize against former winners Brad Rutter and Ken Jennings in Jeopardy TV contest Watson can understand human’s natural language and answer questions. Watson has got all the knowledge of encyclopedias, dictionaries, newswire articles, and literary works. Watson can process about million books, per second. How will future look like? Self driving cars will become common on the road. Bots will become part of our daily life. All devices around us will be a smart device like smart phones. You will be helped by Chat Bots and Virtual assistants on the internet. We will able to cure deadly diseases like Cancer using Machine learning. Math Trick to Teach binary numbers I wanted to end the session with a fun game. So I played a mind reading game based on binary numbers. I wrote about the game in my personal blog. Cool Mind reading trick to teach maths to kids In this post, I will share a cool mind reading trick based on math and computer science fundamentals. It has…erajasekar.com Credits Here are links to articles that helped me to prepare this presentation. The Master Algorithm: How the Quest for the Ultimate Learning Machine will Remake our world 5 Reasons Why Software Developer is a Great Career Choice This week I will give a presentation at a local high school on what it is like to work as a programmer. I am…henrikwarne.com Why Software Is Eating the World This week, Hewlett-Packard (where I am on the board) announced that it is exploring jettisoning its struggling PC…a16z.com Computer Science The Department of Computer Science at Calvin College offers programs in computer science, information systems, digital…cs.calvin.edu It Took the Telephone 75 Years, Internet 4 Years, Facebook 3.5 Years To Do What Angry Birds Did in… "Angry Birds Stella Pop" was released in December 2014 by Rovio Entertainment, the maker of the classic Angry Birds…www.techworm.net

What middle school students need to know about computer science and software engineering job

31

what-middle-school-students-need-to-know-about-computer-science-and-software-engineering-job-14e12404fe73

2018-04-08

2018-04-08 16:20:34

https://medium.com/s/story/what-middle-school-students-need-to-know-about-computer-science-and-software-engineering-job-14e12404fe73

false

2,366

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Rajasekar Elango

I’m Rajasekar Elango, Principle Software Engineer at Salesforce.com. https://doculet.net/

17656edbe73

e.rajasekar

104

65

20,181,104

null

0

null

0

5fc6c58d16f6

2018-04-29

2018-04-29 22:50:08

2018-04-29

2018-04-29 23:50:36

1

false

en

2018-05-01

2018-05-01 22:39:55

0

14e1cf3e3cc1

2.4

5

0

Automated trading signal generation supports in dealing with Digital Asset Exchanges in particular, not only because of their particular…

5

Predicting digital asset trends through automated technical analysis Automated trading signal generation supports in dealing with Digital Asset Exchanges in particular, not only because of their particular nature in terms of the difficulty of making gains, but also for the reason that continuing losses are illogical and unjustified to most investors. It is then a necessity to use modern software techniques to determine the relevant method of dealing and trading to avoid losses. When we eliminate the human emotion - which is usually the main reason for achieving “consecutive” losses that erode the capital - and rely on the machine to analyze the data and nomination of assets without any human intervention; consistent performance in profitable trading is the logical consequence. Previous historical data of digital asset prices are used to predict future price trends, and the prediction model usually uses two layers to analyze the data. Technical analysis in the first layer and then a second layer of comprehending based on machine learning. In addition to fund management strategy that makes use of the recommendations made by the model to determine the course of capital invested. It builds a portfolio of entry and exit signals resulting from the model, and concludes how far the forecasting model is relative to the performance of the whole market. Predicting the direction of future highs and lows is a topic that has been widely studied in many fields in trading, finance, statistics and computer science. The fundamental motivation for sure is to make gains, and professional traders usually use basic analysis and technical analysis to analyze markets and make investment decisions. The basic analysis is the traditional approach of studying the fundamentals of companies such as revenues and expenses, market positioning, annual growth percentage, traded asset technical potential, and so on. Technical analysis, on the other hand, only examines historical price fluctuations and variations. Technical analysis experts study historical prices to define price action patterns using data at different time intervals in an attempt to predict future price movements. Thus there is an inherent correlation between price and the traded asset, which can be used to determine the times of entry and exit for each asset. In finance, statistics, computer science, and most traditional models; statistical models and neural network models are used. Being derived from the price data of the forecast, the dominant strategy in computer science uses spectacular algorithms, Neural networks, or a combination of both (advanced neural networks) where different values of technical indicators are compiled and their future results classified according to the most profitable models and fed to the machine enabling it to match those indices and extract them if they are formed again on the same assets. This is definitely the beauty of AI (Artificial Intelligence). The more information becomes available over time, the more it adapts and learns, and the better it performs. The fruitful outcome of this process; is the investment strategy that can and will be used in identifying markets for trading and investing in terms of profitability through simulating a virtual wallet using trade signals generated by the adopted investment strategy. With the introduction of RedCab’s platform in Q3 2019, and with REDC tokens becoming mainstream with massive demand from the network, and governed capped supply from Proof-of-Driving and Proof-of-Marketing token generation algorithms; it’s crucial to automate the trading strategies to balance the bid and ask market depth and appreciate the token exchange rate overtime with consistent and steady growth rate month over month.

Predicting digital asset trends through automated technical analysis

98

predicting-digital-asset-trends-through-automated-technical-analysis-14e1cf3e3cc1

2018-05-13

2018-05-13 16:21:25

https://medium.com/s/story/predicting-digital-asset-trends-through-automated-technical-analysis-14e1cf3e3cc1

false

583

Decentralized Peer to Peer transportation solution on the blockchain.

null

redcabeg

null

RedCab

info@redcab.io

redcab

BLOCKCHAIN,TRANSPORTATION,PEER TO PEER,MARKETING,ICO

redcab_llc

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

amr diab

Chief Trading Officer @ RedCab LLC | Co-founder @ BitPeaks.com

54fe5d1c70bd

3asaleza

4

1

20,181,104

null

0

null

0

null

2017-09-08

2017-09-08 16:32:33

2017-09-10

2017-09-10 01:57:25

1

false

en

2017-09-14

2017-09-14 09:41:10

0

14e1dde83919

0.830189

0

“Never doubt that a small group of thoughtful, committed people can change the world” — Margaret Mead

3

More than a Startup “Never doubt that a small group of thoughtful, committed people can change the world” — Margaret Mead Neunet is that small group of people. We are committed to using breakthrough technologies in order to break the barriers that exist between people. We’re here to shake things up, break rules, construct new things that never were, connect the unconnected, and ascend human progress to new heights. At the end of the day, it is what we do, and what we have done, that matters to the continuation of the verses in this book of Life. In the upcoming weeks you will learn more about what we’re doing, what we’re all about, and how our success means your success and the success of our children and grandchildren. It’s been an interesting journey so far and in the coming days it will prove to be even more inception-like. There’s only one question that we want you to ask yourself: Are you ready? Welcome to Neunet

More than a Startup

0

startups-more-than-a-startup-14e1dde83919

2017-09-14

2017-09-14 09:41:11

https://medium.com/s/story/startups-more-than-a-startup-14e1dde83919

false

167

null

Entrepreneurship

entrepreneurship

Entrepreneurship

226,400

Idea by Design

Transforming the world through technology

20bf296ea44b

ideaxdesign

25

33

20,181,104

null

0

null

0

null

2017-09-11

2017-09-11 22:01:29

2018-02-26

2018-02-26 19:38:28

1

false

en

2018-02-26

2018-02-26 20:52:54

0

14e3472333ed

3.641509

5

0

I am always amazed, and at the same time inspired, by analogies between technological systems and their natural counterparts. To the extent…

5

GANs and Snoozes: On Amazing Parallels between GANs and Brains I am always amazed, and at the same time inspired, by analogies between technological systems and their natural counterparts. To the extent that I start thinking if it is only my imagination or there is actually so many parallels between these two systems. Last time I felt like that was when I started comparing GANs (Generative Adversarial Networks) with what happens in the brain. Apart from artistic applications to generate new works of art using a dataset of examples, GANs can play an important role in artificial intelligence. GANs are able to use unlabelled data as well as labelled data in a semi-supervised manner. For example, a GAN can learn from all the unlabelled pictures (which is pretty cheap these days, compared to labelled pictures) to learn how those pictures generally look like and what are the usual features in the pictures (such as horizontal, vertical, and diagonal edges, etc.) and then learn to distinguish pictures of certain objects using a small dataset of labelled pictures of those objects. This is due to ingenious and special structure of GANs, which consists of a generator network and a discriminator network. Generator network is used to generate new data similar to what is already available, and discriminator is used to distinguish between real data and the data forged by the generator. This would create an arms race between these two networks to the point that both networks will become pretty good at their own craft. Up to this point, all has been done without GAN looking at a single labelled data. But both networks have been able to construct structures about features and concepts only from locality of phenomena in time and space. Spatial and temporal locality of phenomena means that certain things usually happen together, close to each other (in space or time or both). These patterns in time and space are what allow our brains and neural networks such as GANs to predict what might happen next or what the other half of that car we can’t quite see would look like. Neural networks and our brain both capture these patterns at different levels of abstraction at different layers of their network. Recently Deep Neural Networks were able to capture much more abstract concepts such as “catness” or “dogness” of a patch of image. What happens in unsupervise learning of GANs is that some of these abstract concepts form in the “head” of the neural network without assigning any labels to them. Then when labelled data is available, a few datapoints (labelled pictures) would be sufficient to label these consolidated concepts and their combinations. Now, starts the interesting parallel between GANs and the brain. When unsupervised, a GAN periodically switches between two modes: first, seeing real data while the discriminator in the GAN is learning to appreciate it as real data; and second, generating fake data while the generator learns how to better fool the discriminator and discriminator learns to detect this forgery. When unsupervised, the labels for discriminator are “real” and “fake”, which produce the gradient for it. Labels for generator are “discriminator fooled”, and “not fooled” that produce the error gradients, which backpropagates through generator network and trains it. Essentially, when performing unsupervised learning, the GAN works in two main modes of seeing the real world, and dreaming an imagination generated by itself. These two modes of operation are very similar to awake-asleep cycles in animals. At least what we as humans experience during sleep is a world amazingly realistic in many aspects but unrealistic in a few others that are comically off and illogical. Studies have shown that both the levels of certain hormones and the patterns of brainwaves are significantly different in these two states. As if a sleeping brain is generating imaginary sensory data (and possibly higher-level abstract data), learning to distinguish them from reality of wakefulness. By going through these cycles the brain prepares a discriminator with all the conceptual features needed to detect patterns and learn from minimum number of labels, instructions, supervision, and experiments. But also, it creates a generator that is a sufficiently good model of reality that allows brain to predict what may happen next (both next in time and next in space). But what happens if we don’t give enough chance to a GAN to dream? Its performance drops significantly. First, the number of fake data point will decrease compared to the real data points, which creates an unbalanced dataset. This makes discriminator to see more real data than fake, which forces it to simply bet on “real” as its statistical experience shows that most of the data is real. What happens to the generator is even more interesting and dramatic. The generator misses most of its chance to be trained and would almost completely loses its ability to model the reality. These two deficiencies makes a GAN to lose its grip on reality by not being able to tell things apart with the discriminator, and not being able to correctly predict the future or rest of the sensory data. It sees a cute cat but thinks it is an angry dog. Or thinks that if someone keeps walking on the edge of a cliff, one would fly smoothly into the sky. Sound familiar? These symptoms are known in psychology as hallucination, impaired cognition, and psychosis. All of them are well-known symptoms of long-term sleep deprivation.

GANs and Snoozes: On Amazing Parallels between GANs and Brains

54

on-amazing-parallels-between-gans-and-brains-14e3472333ed

2018-03-20

2018-03-20 07:02:18

https://medium.com/s/story/on-amazing-parallels-between-gans-and-brains-14e3472333ed

false

912

null

Machine Learning

machine-learning

Machine Learning

51,320

Hooman Shayani

null

124e4c393cd0

hooshaya

1

12

20,181,104

null

0

null

0

7f48c10a07d

2018-03-01

2018-03-01 16:35:32

2017-12-29

2017-12-29 21:27:51

1

false

en

2018-03-01

2018-03-01 17:08:59

5

14e436a3981a

1.516981

0

null

3

Automatic Insights: How AI and Machine Learning Improve Customer Service Author: Brian Hughes / Source: Entrepreneur Artificial intelligence, or AI, iallows computer systems to automatically recognize and perform certain jobs that formerly would have required human intervention. If you’ve ever loaded a new image into the photos application on your computer and had it instantly recognize the faces of every person there, you’ve seen the power of AI on display. Machine learning, on the other hand, takes things one step farther and allows computer systems to essentially learn and improve from experience — without necessarily being programmed to do so. Using the same example as above, say you load an image into the photos app and tag a photo of yourself and your significant other. When you load another photo featuring the two of you into the app a few weeks later, it will nstantly recognize you and display your names — without you doing anything manually. It will continue to recognize you in additional photos uploaded in the future. This is the power of machine learning on display. So, if computers can perform tasks that used to require human interaction (as with AI) and get more effective at this the longer they perform those tasks (as with machine learning, you’re probably getting a picture of the broader implications of this technology for business. AI + machine learning = automatic insights In an enterprise environment, concepts like AI and machine learning are usually small but significant parts of the more recognizable concept of business analytics. To be successful, a brand needs to know as much as possible about the people it is trying to serve. The good news is that those customers are creating massive amounts of data at all times; in fact, more data has been created in the last two years than all the data accumulated up to that point. Unfortunately, that’s also the bad news — because pulling anything meaningful from this mountain of information becomes an uphill battle, to say the least. However, that’s where AI and machine learning come… Click here to read more

Automatic Insights: How AI and Machine Learning Improve Customer Service

0

automatic-insights-how-ai-and-machine-learning-improve-customer-service-14e436a3981a

2018-03-01

2018-03-01 17:09:00

https://medium.com/s/story/automatic-insights-how-ai-and-machine-learning-improve-customer-service-14e436a3981a

false

349

oneQube’s AI powered audience automation software stack enables brands to build powerful relevant digital audience for their product, content and brand. oneQubes software and team of audience architects develop, and manage highly engaged passionate audiences.

null

OneQube

null

oneQube

Connect@oneqube.com

oneqube

BLOCKCHAIN,MARKETING TECHNOLOGY,SOCIAL MEDIA MARKETING,AUDIENCE DEVELOPMENT,DIGITAL MARKETING

oneQube

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

oneQube

Intelligent audience automation software to develop highly engaged digital audiences, grow brand awareness, drive traffic and transactions.

e7162a4f66e5

oneQube

19

180

20,181,104

null

0

null

0

null

2017-09-02

2017-09-02 19:49:10

2017-09-02

2017-09-02 19:51:20

1

false

en

2017-09-02

2017-09-02 19:51:20

0

14e557d0d81c

0.271698

0

null

3

Distributed AI Blockchain, Collective Machine Learning, Big Data Deep-Mind Computing, Peer-to-peer Neural Network Big Data Computing

Distributed AI Blockchain, Collective Machine Learning, Big Data Deep-Mind Computing, Peer-to-peer…

0

distributed-ai-blockchain-collective-machine-learning-big-data-deep-mind-computing-peer-to-peer-14e557d0d81c

2018-05-02

2018-05-02 06:45:03

https://medium.com/s/story/distributed-ai-blockchain-collective-machine-learning-big-data-deep-mind-computing-peer-to-peer-14e557d0d81c

false

19

null

Neural Networks

neural-networks

Neural Networks

3,870

Neural Club

null

46dc65f5d58

neuralclub

3

7

20,181,104

null

0

null

0

5e5bef33608a

2018-01-05

2018-01-05 13:33:43

2018-01-05

2018-01-05 14:48:46

0

false

en

2018-01-05

2018-01-05 14:49:36

2

14e58d817240

7.4

0

Over the course of the year 2017, big internet companies like Google, Amazon, Facebook and Apple have been scrutinized like never before…

4

Google, Facebook and the Internet: Time Theft and the Threat of Idiocracy Over the course of the year 2017, big internet companies like Google, Amazon, Facebook and Apple have been scrutinized like never before. To some extent, it is quite surprising that more attention is being paid to where these internet giants are paying their corporate taxes than to what really lies behind their colossal revenues. For sure, Google, Amazon, Facebook and Apple are providing us with product and services that carry great advantages. Access to information and entertaining content, communication, and social interactions have never been easier. However, it would be a great mistake to overlook the costs at which these advantages come. There are growing concerns regarding not only our data privacy, but also the impact of new technologies and of the internet on our lives and on our brains. Nicholas Carr’s claim that Google is making us stupid has been a source of heated debates on the benefits and drawbacks of Google, Facebook, and of the internet in general. This essay is my contribution to the “Google trial” that Carr initiated, and which will likely keep on gaining momentum as the capacities of artificial intelligence increase and as the internet of things makes internet even more omnipresent and intrusive in our daily lives. Google, Facebook and others are taking time and cognitive capacities away from us, designing software that make us dependent and making big money from this time theft and triggered dependence. I will argue that the internet is taking time away from us, and that it is also altering how we think and behave in a negative manner, making us “stupid” as Carr puts it. In my argumentation, I will rely on the definitions of the Cambridge dictionary of “smart” and “stupid”, that are respectively the “ability to think and understand” and the act of “showing poor judgement”. Research and essays from Nicholas Carr, Sherry Turkle, Douglas Rushkoff and Katherine Hayles will be discussed. Distraction, Addiction, and Monetization: The Great Time Theft Google, Facebook and the internet are supplying users with an endless stream of information, flooding webpages and interfaces with hyperlinks. Their business activities are about providing users with content, but also about encouraging them to generate additional content, supporting the never-ending process of information and content diffusion. Besides supplying internet users with an endless stream of information, the internet giants are also behind the process of insuring that there is demand for all of this content. This can be seen, for example, through the massive investments that Google are undertaking to bring high-speed, fiber-powered internet connection to Kansas City, or that Facebook are making to connect developing countries to the internet. These efforts are made with the will to increase the fluidity of the products and the number of webpages that can be visited within a defined time span, in the case of Google, and to increase the number of users of their services, in the case of Facebook. But what is worrisome is the addictive aspects of the internet services provided by Google, Facebook and others. In an effort to grow the demand for their services, software engineers design the internet products in order to hook the users and to make them addicted. Sherry Turkle, MIT professor in social studies of science and technology, warns that technology addiction is more and more resembling the food addiction symptoms of obesity (Weller, 2017). Turkle draws a parallel between the way the food industry adds chemical and sugar to processed meals to make people keen on eating them and the way internet corporations design their applications and webpages to maximize the time that is spent on them. Mechanisms exploiting the cognitive properties of the human brain are incorporated into the technological products and services. As a consequence, these products or websites become too powerful for our brains, and users are progressively losing control over how much time they dedicate to these internet services. This is backed by a number of scientific research that discovered that likes and notifications have our brains release dopamine, a chemical that has a rewarding function and which promotes repeated behavior. Knowing that, it is no wonder to learn, in an analysis that was conducted by the research firm dscout, that an average smartphone user touches it more than 2’600 times per day, with the upper decile reaching out to their devices more than 5’400 times a day (Winnick, 2016). In the same research, it was found that about half of these tactile interactions with smartphones were either with Facebook, or with applications developed by Google (Gmail, Chrome or YouTube). This is fair to say that these high numbers are beyond what is required for a natural and controlled used of a device, and shows the compulsive nature of the use of Google and Facebook among others. To illustrate the addiction triggered by the intensive use of the internet and by the likes of Google and Facebook, I would like to share a personal experience. Reflecting on the issues of distraction, inefficient use of time because of meaningless online activities, and dependence to the internet, I committed to a Facebook-free month. I logged out all of the devices that were connected to my profile and installed a sophisticated, three-steps connection process to make any temptation of logging-in as tedious as possible. While I was expecting to have a hard time for the first few days, I expected the difficulty to be of a similar magnitude as when changing habits or routines. When after three weeks, my fingers were still mechanically typing facebook.com in the search bar of my browser without any conscious intention, I became self-aware of the addiction to Facebook I had unconsciously developed. By contrast with the food or the tobacco industry, addiction mechanisms embedded in the software of the popular internet companies are not dissimulated. Such techniques are the subject of courses in elite universities and of best-selling books. The addiction of users to their services is an inherent part of the activities of companies like Google and Facebook, as their business model consists in selling as much of their users’ attention to advertisers as they can. As American columnist and media theorist Douglas Rushkoff remarks, their purpose is to centralize the interactions that were decentralized at the origins of the web, to be able to concentrate these interactions in a single marketplace and to sell the data generated by these interactions. In the new, digital economy, attention is money, and it is in the interest of the monopolistic internet giants to keep us constantly distracted, and to make us addicted to their services to maximize the value they can extract from our data. In the new digital, attention-based economy, time has become a coveted and pricey resource. After the monopolization of free time, technological service providers are now going after time spent working or sleeping. Google, Facebook and the internet in general organize the addiction of users to their services to maximize the attention time they can sell. Consequently, this decreases the amount of time that we can spend on important projects, on learning new skills or concepts, and increases the time that is devoted to meaningless activities. In that sense, it can be affirmed that an intensive use of the internet decreases our ability to think and understand and makes us more likely to show poor judgements. Valuable time, carrying an opportunity cost of new knowledge or skills acquisition, is stolen away from us, sold for advertising money to the highest bidders. Over time, this carries a risk of a degradation in the overall intelligence of populations. The Thinking Process and Information Flood Not only do we have less time to get smarter and do we spend more time on meaningless activities, our whole neural and thinking process is being affected by the presence and our intensive usage of internet in the course of our daily lives. Nicholas Carr argues that he “is not thinking the way he used to think” and that Google has altered his capacity to “deep-read” and thereby to “deep-think”. On parallel, Katherine Hayles, Professor at Duke University, explains that new technologies and connected devices induce a shift in cognitive styles, from what she calls deep attention to hyper attention. She defines deep attention as the capacity to “concentrate on a single subject for a long period of time”, while hyper attention as the ability to “switch focus rapidly between different tasks”. In the light of Katherine Hayles arguments, it should be understood that Nicholas Carr claim is that Google deteriorates the capacity to focus at length on complex readings and thinking activities, and favors a thinking process which is more oriented towards multi-tasking and fragmented attention. Carr is seeing this as problematic, and argues that his concentration capacity is being diminished permanently as a consequence of his recurrent usage of Google and the internet. On the other hand, Kevin Kelly, founding executive director of Wired Magazine, argues that while we might be losing 20 points of IQ when we are off Google, we are gaining 40 IQ points by being on Google virtually all the time. However, as Carr argues, using Google extensively to the extent that it becomes an extension of our memory and of our brains flattens our intelligence towards artificial intelligence. This leads to a Taylorization of the thinking process, which breaks down our reflections into small, discrete steps involving the use of Google search engines. This is a major issue, and Carr’s argument that our intelligence is flattening to artificial intelligence has to be understood in the sense that an intensive use of technology gets our brains used to functioning like machines. The problem with this is that machines are better are being machines than humans are, and the flattening of our intelligence makes us worse at doing things and thinking in ways for which we are better than machines, at least for the moment. To conclude, Google, Facebook, and other big internet corporations are designing their software in a way that takes time away from their users, and exploits the cognitive and neural circuits flaws of the human brain in order to distract their users towards their services and sell their attention to advertisers. The intensive use of the internet triggers a shift in attention style, from deep to hyper attention, with cognitive processes that are more and more taylorized, and that make the human brain functioning more like an algorithm. The risks of flattening of the overall intelligence of populations are real, and this would trigger immense challenges as technology and artificial intelligence is only going to get smarter with time. The solutions will have to be twofold: on the individual level, awareness needs to be raised, through personal experiences, or through applications that measure one’s activity and provide users with a state of play on the actual usage and time spent on their devices. But the problem will only be solved if individual awareness and initiatives are completed by public policies. It was shown in this paper that internet big players such as Google and Facebook are actively and deliberately designing their products in a way that makes the users addicted. With this in mind, there is a fair ground to argue that internet companies whose business model is to distract their users to sell their attention should be strictly regulated in the manner of tobacco companies. The importance of a strong awareness and debates on this issue cannot be understated. It is about regaining the control of our brains and limiting the influence of the internet giants on their users. It is about reflecting on the aspects that make us inherently human.

Google, Facebook and the Internet: Time Theft and the Threat of Idiocracy

0

google-facebook-and-the-internet-time-theft-and-the-threat-of-idiocracy-14e58d817240

2018-05-09

2018-05-09 12:42:32

https://medium.com/s/story/google-facebook-and-the-internet-time-theft-and-the-threat-of-idiocracy-14e58d817240

false

1,961

Latest News, Info and Tutorials on Artificial Intelligence, Machine Learning, Deep Learning, Big Data and what it means for Humanity.

becominghuman.ai

BecomingHumanAI

null

Becoming Human: Artificial Intelligence Magazine

team@chatbotslife.com

becoming-human

ARTIFICIAL INTELLIGENCE,DEEP LEARNING,MACHINE LEARNING,AI,DATA SCIENCE

BecomingHumanAI

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Guillaume Gauthier

null

c7d5250f0e6d

ggm.gauthier

0

1

20,181,104

null

0

null

0

e39f862bc3e2

2018-03-15

2018-03-15 20:38:00

2018-03-26

2018-03-26 16:02:43

8

false

en

2018-07-15

2018-07-15 19:13:46

10

14e5c0295b94

6.797484

174

0

1

Basics, weight initialization pitfalls & best practices

5

Deep Learning Best Practices (1) — Weight Initialization Basics, weight initialization pitfalls & best practices https://pixabay.com/photo-1600668/ Motivation As a beginner at deep learning, one of the things I realized is that there isn’t much online documentation that covers all the deep learning tricks in one place. There are lots of small best practices, ranging from simple tricks like initializing weights, regularization to slightly complex techniques like cyclic learning rates that can make training and debugging neural nets easier and efficient. This inspired me to write this series of blogs where I will cover as many nuances as I can to make implementing deep learning simpler for you. While writing this blog, the assumption is that you have a basic idea of how neural networks are trained. An understanding of weights, biases, hidden layers, activations and activation functions will make the content clearer. I would recommend this course if you wish to build a basic foundation of deep learning. Note — Whenever I refer to layers of a neural network, it implies the layers of a simple neural network, i.e. the fully connected layers. Of course some of the methods I talk about apply to convolutional and recurrent neural networks as well. In this blog I am going to talk about the issues related to initialization of weight matrices and ways to mitigate them. Before that, let’s just cover some basics and notations that we will be using going forward. Basics and Notations Consider an L layer neural network, which has L-1 hidden layers and 1 output layer. The parameters (weights and biases) of the layer l are represented as In addition to weights and biases, during the training process, following intermediate variables are computed Training a neural network consists of 4 steps: Initialize weights and biases. Forward propagation: Using the input X, weights W and biases b, for every layer we compute Z and A. At the final layer, we compute f(A^(L-1)) which could be a sigmoid, softmax or linear function of A^(L-1) and this gives the prediction y_hat. Compute the loss function: This is a function of the actual label y and predicted label y_hat. It captures how far off our predictions are from the actual target. Our objective is to minimize this loss function. Backward Propagation: In this step, we calculate the gradients of the loss function f(y, y_hat) with respect to A, W, and b called dA, dW and db. Using these gradients we update the values of the parameters from the last layer to the first. Repeat steps 2–4 for n iterations/epochs till we feel we have minimized the loss function, without overfitting the train data (more on this later!) Here’s a quick look at steps 2 , 3 and 4 for a network with 2 layers, i.e. one hidden layer. (Note that I haven’t added the bias terms here for simplicity): Forward Propagation Backward Propagation Initializing weights W One of the starting points to take care of while building your network is to initialize your weight matrix correctly. Let us consider 2 scenarios that can cause issues while training the model: 1. Initializing all weights to 0 Let’s just put it out there — this makes your model equivalent to a linear model. When you set all weight to 0, the derivative with respect to loss function is the same for every w in W^l, thus, all the weights have the same values in the subsequent iteration. This makes the hidden units symmetric and continues for all the n iterations you run. Thus setting weights to zero makes your network no better than a linear model. It is important to note that setting biases to 0 will not create any troubles as non zero weights take care of breaking the symmetry and even if bias is 0, the values in every neuron are still different. 2. Initializing weights randomly Initializing weights randomly, following standard normal distribution (np.random.randn(size_l, size_l-1) in Python) while working with a (deep) network can potentially lead to 2 issues — vanishing gradients or exploding gradients. a) Vanishing gradients — In case of deep networks, for any activation function, abs(dW) will get smaller and smaller as we go backwards with every layer during back propagation. The earlier layers are the slowest to train in such a case. The weight update is minor and results in slower convergence. This makes the optimization of the loss function slow. In the worst case, this may completely stop the neural network from training further. More specifically, in case of sigmoid(z) and tanh(z), if your weights are large, then the gradient will be vanishingly small, effectively preventing the weights from changing their value. This is because abs(dW) will increase very slightly or possibly get smaller and smaller every iteration. With RELU(z) vanishing gradients are generally not a problem as the gradient is 0 for negative (and zero) inputs and 1 for positive inputs. b) Exploding gradients — This is the exact opposite of vanishing gradients. Consider you have non-negative and large weights and small activations A (as can be the case for sigmoid(z)). When these weights are multiplied along the layers, they cause a large change in the cost. Thus, the gradients are also going to be large. This means that the changes in W, by W — ⍺ * dW, will be in huge steps, the downward moment will increase. This may result in oscillating around the minima or even overshooting the optimum again and again and the model will never learn! Another impact of exploding gradients is that huge values of the gradients may cause number overflow resulting in incorrect computations or introductions of NaN’s. This might also lead to the loss taking the value NaN. Best Practices 1. Using RELU/ leaky RELU as the activation function, as it is relatively robust to the vanishing/exploding gradient issue (especially for networks that are not too deep). In the case of leaky RELU’s, they never have 0 gradient. Thus they never die and training continues. 2. For deep networks, we can use a heuristic to initialize the weights depending on the non-linear activation function. Here, instead of drawing from standard normal distribution, we are drawing W from normal distribution with variance k/n, where k depends on the activation function. While these heuristics do not completely solve the exploding/vanishing gradients issue, they help mitigate it to a great extent. The most common are: a) For RELU(z) — We multiply the randomly generated values of W by: b) For tanh(z) — The heuristic is called Xavier initialization. It is similar to the previous one, except that k is 1 instead of 2. In TensorFlow W = tf.get_variable('W', [dims], initializer) where initializer = tf.contrib.layers.xavier_initializer() c) Another commonly used heuristic is: These serve as good starting points for initialization and mitigate the chances of exploding or vanishing gradients. They set the weights neither too much bigger that 1, nor too much less than 1. So, the gradients do not vanish or explode too quickly. They help avoid slow convergence, also ensuring that we do not keep oscillating off the minima. There exist other variants of the above, where the main objective again is to minimize the variance of the parameters. 3. Gradient Clipping — This is another way of dealing with the exploding gradient problem. We set a threshold value, and if a chosen function of a gradient is larger than this threshold, we set it to another value. For example, normalize the gradients when the L2 norm exceeds a certain threshold –W = W * threshold / l2_norm(W) if l2_norm(W) > threshold An important point to note is that we have talked about various initializations of W, but not the biases b. This is because the gradients with respect to bias depend only on the linear activation of that layer, and not on the gradients of the deeper layers. Thus there is no diminishing or explosion of gradients for the bias terms. As mentioned earlier, they can be safely initialized to 0. Conclusion In this blog, we’ve covered weight initialization pitfalls and some mitigation techniques. If I have missed any other useful insights related to this topic, I would be happy to learn it from you! In the next blog, I will be talking about regularization methods to reduce overfitting and gradient checking — a trick to make debugging simpler! References https://www.coursera.org/learn/deep-neural-network/lecture/RwqYe/weight-initialization-for-deep-networks Neural networks: training with backpropagation - Jeremy Jordan A Gentle Introduction to Exploding Gradients in Neural Networks by Jason Brownlee Vanishing gradient problem https://www.quora.com/Why-is-it-a-problem-to-have-exploding-gradients-in-a-neural-net-especially-in-an-RNN About Me: Graduated with MS Data Science at USF and undergrad in Computer Science, I have 2 years of experience in building predictive and recommendation algorithms, and deriving business insights for finance and retail clients. I am excited about opportunities for applying my machine learning and deep learning knowledge to real-world problems. Do check out my other blogs here! LinkedIn : https://www.linkedin.com/in/neerja-doshi/

Deep Learning Best Practices (1) — Weight Initialization

969

deep-learning-best-practices-1-weight-initialization-14e5c0295b94

2018-07-15

2018-07-15 19:13:46

https://medium.com/s/story/deep-learning-best-practices-1-weight-initialization-14e5c0295b94

false

1,501

Established in 2016, the Data Institute at USF serves as the umbrella organization for data science research and programming at the University of San Francisco. We offer MS Data Science, BS Data Science and continuing education certificates.

null

usfca.msds

null

USF-Data Science

info@datascience.usfca.edu

usf-msds

DATA SCIENCE,MACHINE LEARNING,DEEP LEARNING,CODING,STATISTICS

datainstitutesf

Machine Learning

machine-learning

Machine Learning

51,320

Neerja Doshi

MS Data Science student at USF, https://www.linkedin.com/in/neerja-doshi/

3dff2ff1664

neerja.doshi

262

72

20,181,104

null

0

null

0

null

2018-05-30

2018-05-30 07:14:59

2018-05-30

2018-05-30 08:51:42

2

false

en

2018-05-30

2018-05-30 08:51:42

2

14e5e510e011

2.711635

4

0

In today’s fast-moving world, most people have come to expect mobile apps to be super smart — adapting swiftly to users’ activity or…

5

Get started with Machine Learning using ML Kit for Firebase Image from IE.edu In today’s fast-moving world, most people have come to expect mobile apps to be super smart — adapting swiftly to users’ activity or delighting them with astonishing smarts. Hence, I think machine learning will become an integral part of mobile development. That’s why I felt elated when Google gave a big announcement at Google I/O ‘18, they introduced ML Kit in beta: a new SDK that brings Google’s machine learning expertise to mobile developers in a powerful, yet easy-to-use package on Firebase. I couldn’t be more excited! Image from Giphy ML-Kit has come to stay so you can now use machine learning in your apps to solve real-world problems.Isn’t this amazing? I am kick-starting a series on ML-Kit and will be sharing a plethora of applications. In this article, I will share everything you need to know about this amazing Sdk. So let’s start!! ML Kit is a mobile SDK that brings Google’s machine learning expertise to Android and iOS apps in an effective yet easy-to-use package. Whether you’re a beginner or expert in machine learning, you can incorporate the functionality you need in just a few lines of code. There’s no need to have vast knowledge of neural networks or model optimization to get started. On the other hand, if you are an expert ML developer, ML Kit provides convenient APIs that enable you use your custom TensorFlow Lite models in your mobile apps. The power of ML-Kit Production-ready for common use cases: Google ML Kit comes with a set of ready-to-use APIs for popular mobile use cases: recognizing text, detecting faces, identifying landmarks, scanning barcodes, and labeling images. Simply pass in data to the ML Kit library and it gives you the information you need. On-device or in the cloud: ML Kit’s selection of APIs run on-device or in the cloud. Google’s on-device APIs can process your data quickly and work even when there’s no network connection. Cloud-based APIs, on the other hand, leverage the power of Google Cloud Platform’s machine learning technology to provide you with even higher level of accuracy. Deploy custom models: For developers with prior knowledge of Machine Learning, you can always bring your own existing TensorFlow Lite models. Simply upload your model to Firebase, and GCP will take care of hosting and serving it to your app. ML Kit serves as an API layer to your custom model, making it simpler to run and use. What’s in the Box ML Kit gives you five ready-to-use (“base”) APIs that address common mobile use cases: Text recognition Face detection Barcode scanning Image labeling Landmark recognition The 3 basic steps in ML-Kit implementation process Integrate the SDK: Quickly include the SDK using Gradle or CocoaPods. Prepare input data: For instance, if you’re using a vision feature, capture an image from the camera and generate the necessary metadata such as image rotation, or prompt the user to select a photo from their gallery. Apply the ML model to your prepared data: By applying the ML model to your data, you generate insights such as the emotional state of detected faces or the objects and concepts that were recognized in the image, depending on the feature you used. Use these insights to power features in your app like photo embellishment, automatic metadata generation, or whatever else you can imagine. Awesome!!! What’s next? In the next article, I will show you the easiest way to incorporate Text recognition in your app project. We are going to build a simple mobile app that will use Text recognition API. Get ready!!!

Get started with Machine Learning using ML Kit for Firebase

10

get-started-with-machine-learning-using-ml-kit-for-firebase-14e5e510e011

2018-05-30

2018-05-30 09:46:33

https://medium.com/s/story/get-started-with-machine-learning-using-ml-kit-for-firebase-14e5e510e011

false

617

null

Machine Learning

machine-learning

Machine Learning

51,320

Temidayo Adefioye

Technical Writer | Google Certified Associate Android Developer | Open Source Advocate | Software Engineer at diylaw.ng | Raising Tech Leaders is my passion!

67bfc10193ff

temidjoy

141

63

20,181,104

null

0

null

0

null

2018-05-17

2018-05-17 19:29:57

2018-05-17

2018-05-17 20:11:30

6

false

en

2018-05-17

2018-05-17 20:13:05

1

14e5ed0fa4f6

3.734906

0

For downloading the code Follow this link: https://github.com/Ayanzadeh93/cs231n

5

CS231n- Implementing the KNN in the Assignment1 For downloading the code Follow this link: https://github.com/Ayanzadeh93/cs231n For plotting 5 examples of each CIFAR-10 class ,we have used subplot function from built-in function of the python library. The code can be seen in KNN.py code which simply plots 5 examples of each class in Fig1. Figure 1: Sample images of CIFAR-10 dataset. There are 5 samples from each 10 classes of this dataset. Implementing Euclidean distance: Two loops: For implementing with two loops, This code calculate the distance between each instance in testset and trainset iteratively. Finally, it creates a matrix of distances in which shows distance between test instance i and train instance j . kNearestNeighbor.py One loop: For implementing with only one loop, we use broadcasting technique. Actually, in this part we use a for loop in order to go through each instance of test data and in each iteration i. subtract one instance of test data from whole train matrix. Then, we use sum function along axis 1 to add al of the subtracted elements. Finally after finishing the for loop we again will have a matrix of distances. kNearestNeighbor.py No loop: For this part, we use matrix multiplication to find a formula in order to calculate the Euclidean distance. After multiplying the test matrix with transpose of training matrix, each element of this new matrix is result of vector multiplication of one instance from train set and one instance form test set. Actually, in the detail, first we should take square 3 from (X t_rain)and(X_t est), afterwards,each of X of test ant train matrices are summed along row, besides we calculating the multiplication of X t rainandX t est and multiply the result of them to negative two. (−2 ∗ (X t rain ∗ X t est) ). At the end with the help of vectorization feature in numpy we can sum up the result of (−2 ∗ (X t rain ∗ X t est) ) with the with the sum of the square of (X t rain)and(x t est) kNearestNeighbor.py Sampling from main dataset: In order to create a new dataset containing 5000 train and 500 test samples. As can be seen in following figures number of samples in different classes are not equal. But number of samples of each class are close and can be said that data is almost balanced. Following figures show histogram of new sampled train and test dataset in Figure2. Elapsed Time: In case of computation time of these three different ways of implementing I used Time package(Tic-Toc) for measuring the time computation of given methods. Computation time for no loop, one loop and two loops implementations was respectively 0.291302s, 59.508242s and 26.03s Not surprisingly no loop implementation is best implementation and one loop has the worst performance.For evaluating the performance of these, we used the computer that has the following information: Intel Core i7–4702MQ 2.2 GHz and 8 GB RAM and this is captured based on this system. Two loops implementation took 26.03 seconds One loop implementation took 59.508242 seconds No loop implementation took 0.291302 seconds visualization of the Euclidean distance matrix : Figure 3 shows the Euclidean distance between the training and the test set. Figure 3: Euclidean distance between the training and the test set predicting the labels: For predicting the labels of the instances in testset according to their distance, firstly, we sort them according to their distances using np.argsort an then I choose k instances with best distances and then I use np.bincount to counts number of occurrences of each label. Finally using argmax I find the closest instance to each query instance. k-fold cross validation In this part, we split our training data and labels into 5 folds and test the k for 1, 3, 5, 8, 10, 12, 15, 20, 50, 100. based on the k-cross validation technique, in each iteration we choose the 4 out of 5 fold as training set and the rest one is chosen as validation set. Following, table1 shows the mean accuracy of for each k of running the algorithm for different k parameters. Figure 4: Impact of changing K parametr in KNN algorithm on average accuracy of classifier. Please note that all of them have done using 5-fold cross validation and these results are average accuracy(calculate the average of folds in each k). Figure 3 also shows this idea that k=10 has led to best results. By using k=10 on the sampled dataset which contains 5000 training and 500 testing samples algorithm could gain 0.2820 average accuracy. At the end you can see the results of accuracy of each k in 5-fold cross validation. Figure 5: Impact of changing K parameter in KNN algorithm on accuracy of classifier in k-fold cross validation.

CS231n- Implementing the KNN in the Assignment1

0

cs231n-implementing-the-knn-in-the-assignment1-14e5ed0fa4f6

2018-05-17

2018-05-17 20:13:07

https://medium.com/s/story/cs231n-implementing-the-knn-in-the-assignment1-14e5ed0fa4f6

false

738

null

Machine Learning

machine-learning

Machine Learning

51,320

Aydin Ayanzadeh

null

8d07b070518d

a.ayanzadeh

2

6

20,181,104

null

0

null

0

6c08165ab9c1

2018-07-27

2018-07-27 16:31:23

2018-07-27

2018-07-27 16:32:37

0

false

en

2018-07-27

2018-07-27 16:32:58

11

14e7784426fb

1.441509

0

The Hot Stocks Outlook uses VantagePoint market forecasts that are up to 86% accurate to demonstrate how traders can improve their timing…

5

VantagePoint Hot Stocks Outlook for the Week of July 27th, 2018 The Hot Stocks Outlook uses VantagePoint market forecasts that are up to 86% accurate to demonstrate how traders can improve their timing and direction. In this week’s video, we analyze forecasts for Google ($GOOGL), Walt Disney ($DIS), Adobe ($ADBE), Berkshire Hathaway ($BRK/B), and Universal Display Corp ($OLED). This Week’s Hot Stocks Outlook Google ($GOOGL) Google ($GOOGL) had a predictive moving average crossover to the upside in early-July indicating a bullish trend. As soon as the blue line crossed above the black line, VantagePoint users knew they should start taking long positions in this market. In 11 trading days, $GOOGL was up 10.08% or $116.78 per share. Walt Disney ($DIS) Walt Disney ($DIS) follows a similar pattern to the upside. The market had a crossover to the upside in early-July when that blue line made the cross above the black line. The neural index also reflected that short-term strength. Since that crossover, the market had a great run and was up almost 6% in 12 trading days or $6.04 per share. Adobe ($ADBE) Adobe ($ADBE) follows this pattern too. Despite whatever trading strategy that traders are following, that blue line crossed above the black line and was a clear indication that an uptrend was beginning. This bullish trend continued and since that crossover 10 trading days ago, $CHKP was up 12.13% or $11.98 per share. Berkshire Hathaway ($BRK/B) Berkshire Hathaway ($BRK/B) follows the same idea and has really had a great run. It’s also no stranger to the Hot Stocks Outlook. We featured this move last week too! That market had a bullish crossover in early-July. Traders knew, with confidence, that they could begin going long in this market when that crossover of the blue line above the black line Since that crossover of the blue line 11 trading days ago, the market was up almost 5% or $8.96 per share. Universal Display Corp ($OLED) Universal Display Corp ($OLED) is basically the same as the others. An uptrend started in early-July indicating to traders that the trend was bullish and to start taking long positions. In 14 trading days, the market was up over 15% or $13.47 per share. Click here to get your VantagePoint Demo >>

VantagePoint Hot Stocks Outlook for the Week of July 27th, 2018

0

vantagepoint-hot-stocks-outlook-for-the-week-of-july-27th-2018-14e7784426fb

2018-07-27

2018-07-27 16:33:06

https://medium.com/s/story/vantagepoint-hot-stocks-outlook-for-the-week-of-july-27th-2018-14e7784426fb

false

382

the cashflow stories that matter. covering finance, wealth accumulation, venture capital, bitcoin, and money, money, money.

keepingstock.net

keepingstock

null

Keeping Stock

stories@amipublications.com

keeping-stock

STOCK MARKET,FINANCE,CASH FLOW,STOCKS,FINANCIAL REGULATION

keepingstock

Trading

trading

Trading

19,801

Vantagepoint ai

Patented software using Artificial Intelligence to help traders predict the market with up to 86% accuracy. Get a free demo: www.vantagepointsoftware.com/

d47649555a99

Vantagepoint

238

140

20,181,104

null

0

Benchmark (points) Mode Cnt Score Error Units PerformanceTest.testStrTreeIndex 15000 thrpt 200 0.736 ± 0.024 ops/s PerformanceTest.testKdTreeIndex 15000 thrpt 200 0.507 ± 0.006 ops/s PerformanceTest.testNaive 15000 thrpt 200 0.578 ± 0.006 ops/s

1

null

2018-03-04

2018-03-04 20:57:44

2018-03-04

2018-03-04 21:42:10

1

false

en

2018-03-04

2018-03-04 22:24:49

5

14e87deaaede

4.792453

2

0

I’ve been between jobs and it was pretty cold outside, so I hacked around with my laptop for a while.

3

Benchmarking DBSCAN performance I’ve been between jobs and it was pretty cold outside, so I hacked around with my laptop for a while. I wanted to test out if and by how much the DBSCAN algorithm can be sped up by indexing the input points, and I wanted to do this in Java. Search problems are often best solved with trees of some kind. I usually shy away from using them as they often make the code more complex and harder to maintain. I wanted to find out if indexing can speed up the DBSCAN algorithm sufficiently to justify the increased code complexity. Introduction to DBSCAN DBSCAN is a powerful clustering algorithm which in most implementations is O(n²). It has some pretty cool properties, such as not requiring the user to know how many clusters it is searching for, as well as not needing those clusters to be a uniform shape. Additionally, not all points need to be assigned to a cluster making it suitable for use with real world noisy data. For each point, it searches for other points within a distance, epsilon. When it finds a neighbour within epsilon, it searches for neighbours of that point. If the cluster at the end of this search is greater than a parameter termed minPts, the cluster is included in the output. You can read all about it and see pretty pictures on Wikipedia The performance problem comes from the neighbourhood searching. The way most implementations (like this one) handle it is simply looping through all of the data for each query point looking for unprocessed nearby points — so for each point we may have to examine each other point, hence the O(n²) complexity. By using some kind of tree structure to index the points, we should be able to improve the runtime to O(n log n) — at least in a universe where theory actually works. Why did you spend your free time doing Java? Java gets a hard time these days, with everyone pushing Go or Kotlin or Scala. I hadn’t written any Java for around a year having been working mostly in Scala and fancied the practice. Java isn’t so bad; I find that Java code is usually more performant than Scala, even if it does take a few minutes more to write. Additionally I hadn’t worked with Java 8 before this, so took the opportunity to get some experience with the new language features. Nitty Java details I have a Maven project with multiple modules. The core module contains the abstract classes which other modules implement, along with the DBSCAN code which pushes the responsibility of searching a neighbourhood onto a subclass of AbstractIndex. Being able to specify a contract and implement an algorithm once for anything adhering to that contract was great, and after spending some time with Go is a feature I would really miss if I moved over to Go full time. There are two modules for different data structures, naive which doesn’t use a tree and geospatial which imports two trees from the jts library. I went with the StrTree and the KdTree — it’s useful to compare different search structures. All of the important code in these modules is unit tested with JUnit. There is another module containing the performance tests, more on them in a bit. I used some features of Java 8 for the first time here, specifically .forEach on collections and mapping with streams. It was pretty annoying having to convert lists to streams and back again to perform a map, Java really isn’t there yet with functional programming, but the IntStream.range was a pretty elegant way to do something n times. All the code is on github: https://github.com/guyneedham/tree-dbscan. Benchmarking with JMH I was recommended the JMH library by a friend who highlighted it as a really great way to benchmark Java code (other languages supported). It took a bit of getting used to, but here’s the high level overview. When you install JMH, it creates a new module in your Java project set up for running benchmarks. The API allows the developer to annotate methods with @Benchmark, which the Runner looks for and will performance test for you. Like any library, JMH takes a bit of getting used to but it’s well worth it. It’s really flexible and is a thorough way to performance test Java code. The final output of a run after a lot of verbose logging is like this: It turns out that the throughput (or thrpt) mode counts how many times the method runs in a given time period. It performs many runs, and outputs a mean with a 99% confidence interval. So, are trees worth it? Tree indexes for DBSCAN are only worth it with large data sets. The score is directly from the output of JMH, and in this case is the number of times per second that the function can be called. Higher is better — the function returns more times per second. Points — that’s the number of data points in the generated data set we’re trying to cluster. The chart shows that for a small number of points, the algorithm performs better without any kind of tree. Though I haven’t tested this, I believe that this is due to the cost of indexing the data. Once we have 5000 points or more, the StrTree structure is faster to search. The KdTree never pays back our investment in it. The difference in performance with an index is pretty small. You’d have to really want the speedup to pay the code complexity cost of using an index. What did I learn? Well it’s important to performance test your assumptions. Just because something should work blazingly fast doesn’t mean that it will in every case. Consider this interview exchange: Me, an interviewer: Your solution is pretty good, but we were wondering why you didn’t use a tree to speed up the search? You, a candidate: I tried that out. The performance took a nose dive. Here are the results. Now at this point, if the interviewer doesn’t believe you or bangs on about theoretical O complexity you must ask yourself if this is really someone you want to work with. When working with large data sets, it’s often the case that a smarter algorithm will save you time and money. With cloud computing it’s all too easy to say let’s just get more servers, or storage is cheap, or something else along the same lines. You might well save a lot of money thinking about your problem a bit more. However as I’ve shown here, you must performance test your improved solution. Java 8 isn’t all that.The streams and the forEach statements reduce a lot of boilerplate code, but the streams are a little annoying and apparently have a pretty high cost. Overall, I really feel that Java really isn’t that bad. Abhorring Java just because it’s Java and not Scala or some new hip language I’ve never heard of and not for any real reason isn’t good enough. I’m not interested into getting into a big Scala vs Java debate, I enjoy them both in different settings.

Benchmarking DBSCAN performance

23

benchmarking-dbscan-performance-14e87deaaede

2018-03-05

2018-03-05 17:39:15

https://medium.com/s/story/benchmarking-dbscan-performance-14e87deaaede

false

1,217

null

Programming

programming

Programming

80,554

Guy Needham

Coding, traveling, cycling.

8053d7c2a089

guy.l.needham

33

1

20,181,104

null

0

null

0

60b40d78a847

2018-08-01

2018-08-01 06:59:56

2018-08-01

2018-08-01 08:53:15

9

false

es

2018-08-27

2018-08-27 11:01:29

4

14e94caee041

3.660377

0

“Lo más difícil de aprender en la vida es qué puente hay que cruzar y qué puente hay que quemar.” - Bertrand Russel

5

APRENDE PYTHON YA! Libro: Los pilares de Python. #4 - La instalación con Anaconda “Lo más difícil de aprender en la vida es qué puente hay que cruzar y qué puente hay que quemar.” - Bertrand Russel Antes de comenzar con la instalación de Anaconda, debemos definir dos conceptos: las librerías o paquetes y los IDE o entornos de desarrollo integrado. Las librerías son programas escritos por la comunidad (personas como tú o como yo, que escriben en el mismo lenguaje de programación) enfocados en ayudarte en ciertas tareas más concretas fuera de las funciones comunes que vienen integradas en python. Por ejemplo, la librería numpy está enfocada en el cálculo vectorial, la librería pandas simplifica el proceso de tratar con bases de datos y tablas o la librería seaborn la cual es útil para visualización de datos). Hay miles de librerías de python. Sin embargo, lógicamente no hay que utilizar todas, hay 20 que son utilizadas en más del 80% de los proyectos. Por ejemplo, en mi trabajo con Inteligencia Artificial en finanzas utilizamos: numpy, pandas, seaborn, Scikit-learn (como librería de machine learning), dask (si la cantidad de datos es alta) y numba (para mejorar el tiempo de ejecución). Los IDE o Entorno de desarrollo integrado: Los IDE no son nada más y nada menos que el lienzo y los pinceles que utiliza un artista para pintar su obra. Los IDE tienen funciones de autocompletado, que nos sugieren la función que debemos utilizar, herramientas de control de errores o debugging y algunos disponen de visualizador de variables. En resumen, son programas que nos ayudan a trabajar. Lo normal es que ahora pienses, ¿cómo puedes saber que librerías tengo que instalar o que IDE utilizar si ni siquiera hemos empezado a programar ? Pues aquí es donde entra en acción Anaconda. Anaconda según Wikipedia: Anaconda es una distribución libre y abierta de los lenguajes Python y R, utilizada en ciencia de datos, y aprendizaje automático. Con nuestras palabras, es un archivo de alrededor 500 MB que nos simplifica la vida al contener las librerías más utilizadas actualmente, avisarnos de sus actualizaciones y traer incorporado un IDE llamado Spyder. Además, Anaconda incluye otro lenguaje de programación llamado R y su respectivo IDE (R Studio). Dicho lenguaje está más orientado al análisis de datos y no es el objetivo del presente libro. Normalmente, si alguien lee cualquier libro o manual de hace unos años, el procedimiento de instalación de python era el siguiente: 1- Descargar la última versión de python compatible con tu sistema operativo desde la página web de www.python.org. 2 - Instalar python. 3 - Instalar las librerías que pensamos que vamos a utilizar. 4 - Descargar el IDE o interfaz de entorno de desarrollo. Anaconda es una de esas herramientas que siguen la filosofía de python de no reinventar la rueda. La instalación de Anaconda es igual de fácil que la instalación de un programa o videojuego. 1- Nos dirigimos a https://www.anaconda.com/download/ 2- Hacemos clic en descargar en función a nuestro sistema operativo (Windows, macOS o Linux) y si es 32 o 63 bits. 3- Continuamos como cualquier instalación: Con estos simples pasos ya tenemos todo listo para empezar a programar. Os veo en el siguiente capítulo donde hablaremos ligeramente de anaconda y el IDE que vamos a utilizar Spyder. — — — — — — — — — — — — — — The End — — — — — — — — — — — — — Si te gusta esta pequeña y gratuita revista puedes ayudar simplemente compartiéndola o suscribiéndote a la revista. Soy Rubén Ruiz, trabajo en la industria financiera en Inteligencia Artificial. Como proyecto personal llevo esta pequeña revista donde experimentamos con Inteligencia Artificial… Puedes también seguirme en: Instagram (Vida personal, es divertido) => @rubenruiz_t Youtube (Canal sobre IA, intento que sea divertido )=> Rubén Ruiz A.I Github (Donde subo mi código, esto ya no es tan divertido) => RubenRuizT

APRENDE PYTHON YA! Libro: Los pilares de Python. #4 - La instalación con Anaconda

0

aprende-python-ya-libro-los-pilares-de-python-4-la-instalación-con-anaconda-14e94caee041

2018-08-27

2018-08-27 11:01:29

https://medium.com/s/story/aprende-python-ya-libro-los-pilares-de-python-4-la-instalación-con-anaconda-14e94caee041

false

652

Experimentos con Inteligencia Artificial.

null

AI experiments en Español

rubenruiz90@gmail.com

ai-experiments-en-español

PYTHON,PROGRAMMING,ARTIFICIAL INTELLIGENCE,R,DEEP LEARNING

null

Python

python

Python

20,142

Ruben Ruiz

null

2db774b0464f

rubenruiz_26771

24

22

20,181,104

null

0

null

0

null

2018-01-04

2018-01-04 22:45:27

2018-01-04

2018-01-04 22:47:38

9

false

en

2018-01-04

2018-01-04 22:47:38

0

14eaea031d72

2.162264

0

Title: Weather Condition Prediction from Image

3

Week 5— Warmth of Image Title: Weather Condition Prediction from Image Team Members: Berk GÜLAY, Samet KALKAN, Mert SÜRÜCÜOĞLU E-mails Respectively: berkgulay.cs@gmail.com , abdulsametkalkan@gmail.com , mertsurucuogluu@gmail.com We have approximately 6500 datas. Each data consists of matrix of image with RGB values and we splitted them into training and validation data with %25 ratio. I built so many CNN architecture and tried them to obtain the best results. I trained images with different sizes like 20x20, 50x50 and 100x100. Besides general accuracy for validation data, I also printed accuracy of each class separately. To do that I split validation data into 5 parts which represents the classes. There are some trials with architecture of network below: CNN with 6 Layers Loss Function-1 CNN with 7 layers Loss Function-2 The last one is the best among the 20 trials. CNN with 6 layers Each train data is trained in about 10 minutes. But the last one took 30 minutes. Because its size is 100x100x3. Since there are so many variation, I couldn’t try all. I think this good result for us. While I use CNN method, my groupmates have tried to extract features from image to use SVM and DT. Most difficult problem for us is to extract feature. We still try to extract feature, because we encounter many errors while extracting feature. But we have obtained some results so far. Results of Decision Tree: Trial-1 Trial-2 Trial-3 And result for Random Forest So, since there are so many variation, we need more time to try other variations. I hope next week we finish the project and obtain good results for all machine learning methods.

Week 5— Warmth of Image

0

week-5-warmth-of-image-14eaea031d72

2018-01-04

2018-01-04 22:47:40

https://medium.com/s/story/week-5-warmth-of-image-14eaea031d72

false

255

null

Machine Learning

machine-learning

Machine Learning

51,320

Mert Surucuoglu

null

fdf2a9721855

mertsurucuoglu

22

26

20,181,104

null

0

null

0

null

2018-03-07

2018-03-07 17:36:06

2018-03-08

2018-03-08 08:28:08

1

false

en

2018-03-08

2018-03-08 08:28:08

3

14ec390d66ea

1.818868

1

0

Identification of human body poses with mere 2D imagery had been a grand challenge in the field machine vision and with the explosion of…

5

Human Pose Estimations — From 2D to 3D Fig 1. Left — Input images to the model. Middle — 2D skeleton structure output showing the pose. Right — 3D derivation of the 2D skeleton structures. The code implemented on these images can be found here. Identification of human body poses with mere 2D imagery had been a grand challenge in the field machine vision and with the explosion of Deep Learning algorithms, the area has seen quite promising advancements recently. As shown in Figure 1, the outputs from these models are able to accurately identify human body keypoints and form skeleton structures depicting the poses. Advancements of these techniques imply great potential with applications in areas such as sports, security surveillance, patient monitoring etc. The process of human pose estimation can be divided into several parts: Identifying anatomical keypoints of the human body The first task is to identify the different parts of a body, preferably joints, so that these points can be tracked in the imagery. The model has to be trained initially using annotated data to be good at this task. Techniques such as heat maps are being widely used at this step and the accuracy directly impacts the nest step 2. Joining the keypoints to form the skeleton structure Next the identified keypoints have to be joined with each other to form a skeleton structure. This task can be quite tricky as to decide which keypoints should be connected to which. Techniques such as assignment of a confidence score in-between keypoints are used here. 3. Forming a 3D representation of the skeleton structure So far the process has being able to extract a 2D skeleton structure from the imagery which depicts the pose accurately, but can we derive a 3D model out of this? This can be a very powerful tool to have and as shown in Figure 1 (right), the 3D estimations are somewhat accurate, but there’s surely room for improvements. 4. Executing the above three point for multiple humans in a single image Identifying the pose of a single human is impressive. Then what about doing that for multiple people in the same image? This question offers a brand new set of challenges in the steps of detecting humans and joining body keypoints. On top of that, how about executing all these steps in a Real-time implementation? This is where the real deal lies. The research referred in this post tackles this frontier and you can find the paper here-> (https://arxiv.org/abs/1611.08050) The code found here on my GitHub is based on the tensorflow implementation of the algorithm of the paper open sourced by the authors. (Find the original repo here) This is quite an interesting frontier in Deep Learning towork in and there are many possibilities to explore!

Human Pose Estimations — From 2D to 3D

1

human-pose-estimations-from-2d-to-3d-14ec390d66ea

2018-06-17

2018-06-17 14:45:35

https://medium.com/s/story/human-pose-estimations-from-2d-to-3d-14ec390d66ea

false

429

null

Machine Learning

machine-learning

Machine Learning

51,320

Mirantha Jayathilaka

Tech Researcher. Machine Learning. Startups. Writing to share because I was inspired when others did.

9be8b9ff653d

miranthaj

215

39

20,181,104

null

0

null

0

16a66d5e85ff

2017-10-05

2017-10-05 07:55:46

2017-10-25

2017-10-25 17:16:21

0

false

pt

2017-10-25

2017-10-25 17:16:21

1

14eda8f68f2a

3.916981

0

Poderíamos apontar inúmeros fatores que nos impediriam de definir a China como um verdadeiro país desenvolvido. Contudo, quem estivesse…

5

Humanidade Aumentada — H.Á. esperança no futuro da tecnologia Poderíamos apontar inúmeros fatores que nos impediriam de definir a China como um verdadeiro país desenvolvido. Contudo, quem estivesse durante a passada semana em Dalian, no nordeste chinês, defenderia aguerridamente o contrário. Foi ali que se realizou a Annual Meeting of the New Champions (AMNC), acarinhada também de Summer Davos, um conclave desenhado pelo Fórum Económico Mundial para juntar líderes mundiais das mais variadas áreas: negócios, política, ciência e tecnologia, arte, saúde são alguns exemplos. Um espaço onde 2.000 pessoas discutiram inteligência artificial, automação, blockchain e, como tema principal da conferência, o crescimento inclusivo no âmbito da quarta revolução industrial. Este ano, dois Global Shapers portugueses foram convidados a representar Portugal num evento recheado de discussões sobre o futuro da tecnologia e o seu impacto no mundo do trabalho. Desde pequenos grupos de trabalho a discussões com líderes políticos, foi este o principal tema na agenda de centenas de participantes, que partilho em maior detalhe. A quarta revolução industrial e o futuro do trabalho O mote tinha sido dado pelo primeiro-ministro chinês, Li Keqiang: a conferência seria sobre como a tecnologia podia habilitar e potenciar o humano, não substitui-lo. Durante os três dias seguintes, vários passos foram dados no sentido de discutir novos modelos de crescimento económico, guiados pela moral e pela ética, a favor de uma sociedade mais justa, inclusiva e equitativa. Ao contrário das suas precedentes, a quarta revolução industrial terá como características não só a velocidade (tudo está a acontecer mais depressa que no passado), como também a sua escala e profundidade (muitas mudanças radicais, a nível global), resultando na total transformação dos sistemas. Sabemos que esta será capaz de gerar desafios e benefícios em igual medida. Porém, não podemos descurar que viveremos das mais profundas e sistémicas alterações ao nosso modelo de sociedade. Aqui, muito se tem falado sobre o receio patente de que os sucessivos avanços tecnológicos possam tornar muitos dos nossos trabalhos obsoletos, deixando-nos sem fontes de rendimento, contribuindo assim para um agravamento da nossa situação sócio-económica. Com estes avanços, aproximamo-nos de uma era onde a diferença entre o retorno do capital e o retorno do trabalho é cada vez maior, em prejuízo do segundo. Consequentemente, a falta de foco no desenvolvimento de políticas inclusivas de crescimento poderá resultar, num futuro próximo, numa força de trabalho inativa, com recursos humanos e materiais subaproveitados. Historicamente, as ondas do progresso têm sido acompanhadas por forças de difusão e de concentração. A difusão acontece quando o poder e os privilégios das partes estabelecidas se dissipam e a concentração, quando a influência daqueles assumir o controlo das novas tecnologias, se expande. Assim, uma das preocupações debatida foi a possibilidade de existir uma tendência para um forte aumento da desigualdade, seja no acesso ao trabalho ou na justa distribuição de riqueza. Aliás, já em 1930, Keynes previa que no futuro existiria uma massificação do desemprego causada pelos avanços tecnológicos, devido à descoberta de novas formas de economizar o uso do trabalho, estas mais rápidas que a nossa velocidade de criar novas formas de trabalho. Desemprego: reduzir agora, reorientar depois Para lá do impacto dos avanços tecnológicos do futuro, outro dos temas da conferência passou por aferir a situação atual, tendo sido identificados vários problemas estruturais que obrigatoriamente terão de ser resolvidos o quanto antes, entre eles a estagnação da natalidade nos países desenvolvidos, o envelhecimento da população e ainda o desemprego jovem. Aliás, não foram poucos os painéis onde a temática do desemprego dominou o discurso dos vários participantes. Como seres humanos, sempre tivemos uma capacidade estupenda de nos adaptarmos. Contudo, devido aos motivos supracitados (velocidade, impacto e escala), é fundamental que a forma como nos preparamos para esta nova leva de desenvolvimento seja bem preparada e executada. Falhar aqui significa que o efeito de capitalização de políticas de formação e emprego serão largamente suplantadas pelo efeito de destruição e de substituição que as novas tecnologias nos trarão. Assim, e sem demora, é necessário pensar holisticamente e focar em criar novas formas de pensar, de estudar, de criar negócios (e os seus modelos), enfim, novas formas de trabalhar, ao mesmo tempo que democratizamos o acesso à educação e formação de skills para as desempenhar. A inclusão como pilar dos valores da sociedade Concluindo, é evidente que ninguém consegue definir como serão as nossas próximas décadas — como podemos, então, planear e preparar para algo que desconhecemos? As mudanças radicais na nossa sociedade significam que as camadas mais jovens necessitam de novos skills, muitos deles ainda pouco percebidos e identificados. Contudo, uma das soluções pode passar por dinamizar processos de aprendizagem que criem esses novos skills. Facilitar a criatividade e questionar os mais diversos pressupostos com que fomos criados. Um deles prende-se com o combate à(s) desigualdade(s). Um primeiro-ministro europeu deixou esta mensagem bem clara numa das sessões: “À medida que caminhamos em frente como sociedade, temos de garantir que ninguém fica para trás”. Perceber como podemos garantir o progresso para todos e criar sistemas distributivos, porém justos, por defeito, questionando novamente tudo aquilo que conhecemos, é uma das principais formas de atacarmos as injustiças e desigualdades que bem conhecemos. Este é, possivelmente, um dos maiores desafios da nossa geração. João Romão tem 28 anos e é o fundador da GetSocial.io, uma empresa portuguesa que ajuda jornais online e equipas de marketing digital a descobrirem quais dos seus conteúdos estão a ficar virais, onde e porquê. Juntou-se ao Global Shapers Lisbon Hub em 2013 e é presença assídua em eventos do Fórum Económico Mundial, tendo já participado nos eventos de Davos (Suíça) e de Dalian (China). O Observador associa-se aos Global Shapers Lisbon, comunidade do Fórum Económico Mundial para, semanalmente, discutir um tópico relevante da política nacional visto pelos olhos de um destes jovens líderes da sociedade portuguesa. Ao longo dos próximos meses, partilharão com os leitores a visão para o futuro do país, com base nas respetivas áreas de especialidade, como aconteceu com este artigo sobre temas associados à quarta revolução industrial. O artigo representa, portanto, a opinião pessoal do autor enquadrada nos valores da Comunidade dos Global Shapers, ainda que de forma não vinculativa. Originally published at observador.pt.

Humanidade Aumentada — H.Á. esperança no futuro da tecnologia

0

humanidade-aumentada-h-á-esperança-no-futuro-da-tecnologia-14eda8f68f2a

2018-05-09

2018-05-09 23:11:35

https://medium.com/s/story/humanidade-aumentada-h-á-esperança-no-futuro-da-tecnologia-14eda8f68f2a

false

1,038

The Global Shapers Community, an initiative of the World Economic Forum, is a network of Hubs developed and led by young people who are exceptional in their potential, their achievement and their drive to make a contribution to their communities. This is their Lisbon Hub.

null

shapersLisbon

null

Global Shapers Lisbon

hello@globalshaperslisbon.com

global-shapers-lisbon

GLOBAL SHAPERS,WORLD ECONOMIC FORUM,DIGNITY,EDUCATION,YOUNG PEOPLE

shaperslisbon

Global Shapers

global-shapers

Global Shapers

335

João Romão

Founder @ GetSocial.io , a content analytics & automation platform helping publishers & marketers measure, promote and amplify their best content.

6b7131f9341e

joaoromao

640

1,144

20,181,104

null

0

null

0

null

2018-02-20

2018-02-20 17:01:00

2018-02-20

2018-02-20 17:35:25

2

false

en

2018-02-20

2018-02-20 17:47:18

1

14edd55d8165

3.839937

0

The Human Progress team produced research to show that the cost of light has fallen by a factor of 500,000, and the amount of labor that…

5

Dangers of Automation in the Bridge Between Environment and Cognition The Human Progress team produced research to show that the cost of light has fallen by a factor of 500,000, and the amount of labor that once bought 54 minutes of light now buys 52 years of light. Light-for-work is perhaps one of the most appropriate measures we could use for progress. As you will see, this is both measure and metaphor. Light allows work to happen in the first place, but most importantly, and more intimately, it allows the individual to thrive in the after-hours, the dimly lit environment where the mind has time for itself. Not burning wood or candles or whale oil, and having ample materials to read and write on, these are some of the essential metrics of cultural freedom. Light is also the nature of information. The internet runs on cables that where information flows through, and as light. Human vision functions as a factor of light, and cognitive processes in the brain probably also work through a consciousness interface of light, matter and biology. Immensely augmenting the technological production of light is part of that same continuum of light, matter and biology. Physical and energy resources required to produce light come from an ever evolving technological relationship between individual and environment, and we are not nearly out of the extractive phase. We still require finite materials to produce light, which currently puts a hard limit on our possible hours of enhanced vision within the environment. The fuel of AI is also light. Any technology we build, including quantum AI, will integrate into the environment, to some degree, through light. While the fundamental property of mindfulness transcends light, the ability to exert cognitive ability within an environment requires light, whether the cognition is human or artificial. Which poses a problem as long as the relationship between light generation and material resources is consumptive. Since the internet is information, you could think of the whole of it as light. Similarly, the upcoming internet of things, although physical in nature, will run entirely on an informational infrastructure, thus light. Every technology we produce from now on, as it will tend towards the higher echelons of artificial cognition, will be more and more light reliant for its functioning. And as we produce more light, we consume more resources. This is obvious. Somewhat less obvious, and more insidious, is the fact that we have almost given up on finding a technological solution to the generation of light and energy. Humans, as a technological species, are making the jump directly to AI, which means we are relying on AI to solve the light generation technology problem for us. The insidious part of the problem is that the initial rungs of AI apply less to creative innovation and more to automation. Automation will increase our work to light relationship many times over, probably more than in our entire history, but because we will rush to apply automation throughout the infrastructure of the industrial and social ecosystems, we will be populating the environment with light consumptive technology that is unsustainable, before we have any guarantee of true General AI, much less Super AI. In terms of our investment into AI technology, we will in fact be going halfway through and then making massive investments into human-supportive technological ecosystems using immature AI technology, while a the same time having more or less given up (as far as species wide investments) on the one technology needed to support both human and artificial cognition: sustainable light generation. Technology ultimately promises an indestructible bridge between the fundamental qualities of mind and the lit environment. This is the dream of sci-fi: beautiful metal structures that span galaxies and rival suns, their depth of luminescence and their intricacy somehow magnetic to the human spirit. A positive magnetism, a magnetism of promise, luminous and welcoming, and yet, depending on which technological trees you pursue and how, nature can turn off your light in an instant. As romantic as it may sound, as distant from technology as it may seem, the root of progress is the wisdom light of the human spirit, and the way it uses work to support the sustainable unity of beings and environment. Without cultivation of mindfulness as a basic requisite for species-wide intelligence and the appropriate use of our structural inheritance, we are basically running blind into light-consumptive utopias, hoping for our future robot friends to save us. And this is no metaphor, this is the unfolding present. Of course, automation itself is a technology that relates fundamentally to materials (in the economy, even virtual processes which can be automated have a physical substrate, and automation results in more physical production), and it may improve material production to a degree where progress in sustainable energy occurs more or less naturally. It may reduce the technological friction from a materials point of view, so to say, or it may not. At any rate, as a species, it is better to intelligently plan than to rely on improvements in the movement of atoms in space (automation). Process is entropic, and virtualizing the rate of movement of materials results in a melding of artificial cognition and artificial materials without a sustainable infrastructure to deal with the resulting entropy. Take-away message? The bliss and anti-entropy of sci-fi utopias is build in the inner wisdom light of the human mind. That’s where robots come from too.

Dangers of Automation in the Bridge Between Environment and Cognition

0

dangers-of-automation-in-the-bridge-between-environment-and-cognition-14edd55d8165

2018-02-20

2018-02-20 17:47:19

https://medium.com/s/story/dangers-of-automation-in-the-bridge-between-environment-and-cognition-14edd55d8165

false

916

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Luca Oprea

Futurism, Human-centric business, Mindfulness, Copywriting, Sales, Social strategy, Compassionate behavioral growth https://about.me/luca.oprea

2f11156218d7

lucaopreacontact

116

249

20,181,104

null

0

null

0

null

2018-05-09

2018-05-09 18:17:25

2018-05-16

2018-05-16 23:49:41

27

false

en

2018-10-13

2018-10-13 14:58:54

3

14ef6e45eba5

8.930189

12

5

0

Even AlphaGo is impressive, it requires bootstrapping the training with human games and knowledge. This is changed when DeepMind released…

5

“A foosball table with players wearing red and blue jerseys in Rosebank” by Mpho Mojapelo AlphaGo Zero — a game changer. (How it works?) Even AlphaGo is impressive, it requires bootstrapping the training with human games and knowledge. This is changed when DeepMind released AlphaGo Zero in late 2017. While it receives less media attention, the breakthrough may be more significant. It is self-trained without the human domain knowledge and no pre-training with human games. This brings us one step closer to what Alan Turing may vision: Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? We may expect AlphaGo Zero is more complicated and harder to train. Ironically, AlphaGo Zero has only one deep network and takes only 3 days of training to beat AlphaGo which takes 6 weeks of training. AlphaGo Zero Deep network In reinforcement learning (RL), we use a policy p to control what action should we take, and a value function v to measure how good to be in a particular state. In Go, our policy controls the moves (actions) to win a game. To model uncertainty, the policy is a probability distribution p(s, a): the chance of taking a move a from the board position s. The value function is the likeliness that we will win from a specific board position. As an example below, the probability of taken move a3 is 0.6, and at the board position s’3, the value function is 0.9. i.e. a 0.9 chance of winning the game. In AlphaGo Zero, we use a single deep network f, composed of convolutional layers, to estimate both p and v. It takes the board position (s) as input and outputs p and v accordingly. In the original AlphaGo, it uses two separate deep networks, and we train them with human-played games using supervised learning. In AlphaGo Zero, we train a single network f using self-play games. However, learning the policy and the value function is not good enough (or accurate enough) to beat the Go masters. We need the Monte Carlo Tree Search (MCTS). We will look into MCTS first before coming back on how the deep network f is trained. Monte Carlo Tree Search (MCTS) In MCTS, we use a search tree to record all sequence of moves that we search (play). A node represents a board position and an edge represents a move. Starting with a board position (s3 below), we search possible moves and evaluate the policy and the value function using the deep network f. We estimate position s’3 has a 0.9 chance of winning (v=0.9). We expand the search tree by making a move. This adds a new board position and its corresponding moves to the search tree. We use f to estimate the value function and the policy for the added node and edges. (To reduce confusion in our discussion, the value function is relative to us rather than the current player or the opponent.) Let’s introduce another term called action-value function Q. It measures the value of making a move. In (a) below, we take the move in red (a3) and end up with a 0.9 chance of winning. So Q is 0.9. In (b), we make one more move and end up with a 0.6 chance of winning. Now, we have taken the move a3 twice (visit count N=2). The Q value is simply the average of previous results, .i.e. W=(0.9+0.6), Q = W/2=0.75. In (c), we explore another path. Now, a3 is picked 3 times and the Q value is (0.9+0.6+0.3)/3 = 0.6. The Q value serves a similar purpose as the policy p estimated by the deep network f. However, as we explore more, our search tree grows bigger. If we play enough, the noise in p cancel each other. So Q is getting more accurate as we make more moves. However, the search space for Go is huge, we need to prioritize our search better. There are two factors we need to consider: exploitation and exploration. exploitation: perform more searches that look promising (i.e. high Q value). exploration: perform searches that are not frequently explored (i.e. low visit count N). Mathematically, we select the move a according to: where u is In this equation, Q controls the exploitation and u (inverse to its visit count) controls the exploration . Starting from the root, we use this policy (tree policy) to select the path that we want to search further. In the beginning, MCTS will focus more on exploration but as the iterations increase, most searches are exploitation and Q is getting more accurate. AlphaGo Zero iterates the steps about 1,600 times to expand the tree. We use the search tree to create a policy π to pick our next move for the board position s3. Surprisingly, we do not use Q to compute the policy π. Instead, π is derived from the visit count N. After the initial iterations, moves with higher Q value will be visited more frequently. We use the visit count to calculate the policy because it is less prone to outliners. τ is a temperature controlling the level of exploration. When τ is 1, we select moves based on the visit count. When τ → 0, only the move with the highest count will be picked. So τ =1 allows exploration while τ → 0 does not. With a board position s, MCTS calculates a more accurate policy π to decide the next move. MCTS improves our policy evaluation (improves Q), and we use the new evaluation to improve the policy (policy improvement). Then we re-apply the policy to evaluate the policy again. These repeated iterations of policy evaluation and policy improvement are called policy iteration in RL. After self-playing many games, both policy evaluation and policy improvement will be optimized to a point that it can beat the masters. (A recap of the MCTS can be found here.) Difference between AlphaGo and AlphaGo Zero search algorithm As shown below, AlphaGo does not use the Monte Carlo Rollout (in yellow). AlphaGo Zero uses the self-trained network θ to calculate the value function v while Alpha Go uses the SL policy network σ learned from real games. Even the equation in computing Q looks different, the real difference is AlphaGo has an extra term z (the game result) that is found by the Monte Carlo rollout which AlphaGo Zero skips. Self training using Monte Carlo Tree Search After looking into MCTS, we come back on how f is trained. So we start with an empty board position s1. We use the MCTS to formulate a policy π1. Then we sample a move a1 from π1. After taking the move a1, the board is in s2. We repeat the process again until the game is finished at which we determine who win z (z=1 if we win, 0 otherwise.). This whole self-play game creates a sequence of input board position, policy and game result z. This is the dataset we used to train the deep network f using supervised training. AlphaGo Zero plays games with itself to build up a training dataset. Then it randomly selects samples from the dataset to train f. Our loss function used in the training contains: a mean square error between our value function estimation v and the true label z. A cross entropy for the estimated policy p and the policy π, and a L2 regularization cost with c = 0.0001. Self-Play Training Pipeline However, not all self-play games are used for the deep network f training. Only games played by the best models are used. This self-play training pipeline contains 3 major components that run concurrently. Optimization: Using samples in the training dataset to optimize f and checkpoint models every 1,000 training iterations. Evaluator: If the MCTS using the new checkpoint models beats the current best model, we use it as the current best model instead. Self-play: Play 25,000 games with the current best model and add them to the training dataset. Only the last 500,000 self-play games are kept for training. AlphaGo Zero was trained for 40 days with 29 million self-play games. The parameters for f are updated from 3.1 million of mini-batches each containing 2,048 board positions. Recap The deep network f is trained concurrently when AlphaGo is playing games with itself. Using the policy and the value estimation from the network f, MCTS builds a search tree to formulate a more precise policy for the next move. In parallel, AlphaGo Zero let MTCS plays against each other using different versions of the deep network f. Then it picks the best one so far to create the training samples needed to train f. Once the training is done, we can use the MCTS with the most optimal network f to plan for the next move in a real game. Deep network architecture The deep network f composes of a 3×3 convolutional layer outputting 256 channels followed by 39 residual blocks. A residual block composes of 3×3 convolutional layers and skip connections. The output is feed into a policy head and a value function head. The policy head composes of a 1×1 convolutional layer output 2 channels. The layer is flatten to a vector of 722 elements followed by a FC layers outputting 362 logits (move to one of the 19×19 possible locations, or pass). The value head composes of a 1×1 convolutional layer output a single channel. The layer is flatten to a vector of 361 elements followed by a FC layer outputting 256 elements, then another FC layer to a scalar value followed by tanh (output values between -1, 1). The network takes an input of 19×19×17 features (17 features for each grid location for the 19×19 Go board). The input features contain the current and the last 7 board positions (a total of 8 positions). For each grid location, it has 2 binary values: the presence of the white and the black stone respectively. Therefore we have a total of 8×2=16 features. The last feature is the color of the stone to play now. Reference Symmetry Since a Go board is symmetrical, we can randomly rotate or mirror a board with the function d in our training. This acts as a kind of data augmentation. End of game A game is finished when both players pass, or the search value drops below a resignation threshold, or the game exceeds a maximum time step. The game is scored to determine who win. Further reading How the original AlphaGo works. AlphaGo: How it works technically? How does reinforcement learning join force with deep learning to beat the Go master? Since it sounds implausible, the…medium.com Credits Diagrams in particular containing are modified from the original DeepMind’s “Mastering the Game of Go without Human Knowledge” paper.

AlphaGo Zero — a game changer. (How it works?)

140

alphago-zero-a-game-changer-14ef6e45eba5

2018-10-13

2018-10-13 14:58:54

https://medium.com/s/story/alphago-zero-a-game-changer-14ef6e45eba5

false

1,810

null

Machine Learning

machine-learning

Machine Learning

51,320

Jonathan Hui

Deep Learning

bd51f1a63813

jonathan_hui

2,787

11

20,181,104

null

0

null

0

9cf179295b1a

2018-09-15

2018-09-15 07:09:41

2018-09-15

2018-09-15 07:10:39

2

false

en

2018-09-15

2018-09-15 07:10:59

5

14efeee6b39b

4.213522

0

Originally published on 7/16/2018

4

Live from the Japan-US Innovation Awards Symposium 2018 Originally published on 7/16/2018 Source: http://www.usjinnovate.org/ We stopped by the 8th annual Japan-US Innovation Awards Symposium, a half-day conference celebrating the latest innovative companies making waves in Japan. According to Allen Miner, CEO of Sunbridge Corporation and member of the Innovation Advisory Council, this was the first year in which the Japanese award winner was as innovative and market-ready as the US award winners. I suppose this is another piece of anecdotal evidence that a) Japanese companies were indeed behind in creating market-changing innovation, and b) they are catching up to Silicon Valley very quickly. Keynote: Dr. Yoky Matsuoka Source: LinkedIn The keynote speaker at the symposium was Dr. Yoky Matsuoka (CTO, Nest), a Japan-born tennis-player-turned-robotics-expert, who shared her perspective on the current state of technology (robotics and ML / AI), her background, and what she is excited about the recent surge in tech companies in Japan. On her inspirations: Dr. Matsuoka emphasized creating an impact in the world as being her main source of inspiration. While this line of thought can easily fall into the category of overused Silicon Valley altruisms, she has her own twist of the cliche, and a resume to back her claims up. As a technologist, she has seen companies fail due to poor commercial execution. Naive altruism can, indeed, take you down the path of “creating the best prosthetic pinky for those who don’t have one,” yet be forced to shut down due to the niche market size. Her approach: to first start by helping the largest group possible to elevate their standing, then digging in deeper for those who need even more. On Japanese and US tech: “There’s great synergy between the US and Japan in that companies in each country are good at what the other lacks, and vice versa.” — Dr. Matsuoka What Dr. Matsuoka has seen so far between companies in the US and Japan is that what each country lacks can be complemented by the other’s strengths. On one hand, Japan tends to be head-and-shoulders above the US when it comes to physical manufacturing. Panasonic, with hundreds of products, is a brand you can’t avoid in Japan if you’re looking for a camera, laptop, air conditioner, etc. Yet they all work perfectly because of the meticulous design, development, and test work that goes into their production process. On the other hand, the US, in her opinion, by far has better software, AI expertise, and user experience. She clarified one point about her comment on use experience — yes, Japan has cute geriatric care robots, but the US continues to produce apps that are part of their users’ lives, apps you can’t live without. Her encouragement to Japanese companies: There are US companies who simply want to partner with Japanese companies to achieve a win-win situation, and it’s time Japanese companies let their guard down a little to pursue these global partnership opportunities. “Larger corporations have built everything from scratch, which has now become a defense mechanism against external collaboration. It is now the time to start trusting US companies who want to collaborate with Japanese companies.” — Dr. Yoky Matsuoka On misconceptions of robotics and ML: Media depiction would lead you to believe we are either in the eve of a techno-apocalypse a la the Terminator, or in the techno-utopian world of the Jetsons. Dr. Matsuoka broke the (un)fortunate news that, while robotics and machine learning have come a long way, we are in no way close to either scenario. The Nest device still makes a lot of mistakes in determining whether a movement was caused by the sun’s movement or a burglar, and autonomous cars will need a driver in the driver’s seat ensuring safety. She also recounted a story from the early days at Nest. The Nest team found out that a competitor was able to decrease household energy spending by 33% on average without machine learning. So, the logical step forward was to pursue a 40% decrease with machine learning as their goal. The result was boiling hot houses and irritated tenants who would crank up their AC every time Nest turned it down, causing a net increase in energy spending. Her observation: “people are lazy but rebellious,” and can’t let machines be in charge. Machines can help guide our decisions, but it is us who need to have the final say. “People are lazy, but rebellious.” — Dr. Matsuoka In terms of unethical use of algorithms, data, and connected devices, Dr. Matsuoka predicts policy administration will take on an integral part in innovation as it has with nuclear energy and cloning / stem cell research. The “can we” versus “should we” debate will fall flat on its face when malicious actors take up these technologies to carry out their deeds. We might as well start a conversation between private industry, government, and the academic circles before we experience terrible consequences. On being bicultural: Dr. Matsuka pointed out that as a bicultural person who understands two wildly different cultures, she is able to adapt her personality to a wider variety of situations. She calls this “situationally appropriate adjustment.” Her statement that her job is 30% technical skills and 70% people skills reminds us that it is critically important for anybody working cross-culturally (or even cross-functionally) to adjust their behavior depending on the environment and their position in that environment. Other Quotes from the Symposium: Issei Takano, CEO of Mujin, industrial robot and controls system startup, on Japan losing 2,000 people from their population every day: “People are worried that robots will take away jobs, but there aren’t enough people to take jobs from.” Michael Tso, CEO of Cloudian, enterprise data storage company, on Japanese VCs versus those in the US: “Japanese VCs are much more of a long-term player. There’s is an initial dating period, but they have deeper pockets and more patience for enterprise-focused startups to succeed. American VCs are much more transactional and short-term oriented.”

Live from the Japan-US Innovation Awards Symposium 2018

0

live-from-the-japan-us-innovation-awards-symposium-2018-14efeee6b39b

2018-09-15

2018-09-15 07:10:59

https://medium.com/s/story/live-from-the-japan-us-innovation-awards-symposium-2018-14efeee6b39b

false

1,015

Leading voices of the Japanese startup scene.

null

Innovators in Japan

contact@innovatorsinjapan.com

innovators-in-japan

STARTUP,TECHNOLOGY,VENTURE CAPITAL,JAPAN,DISRUPTION

InnovatorsInJPN

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Shohei Narron

Born and raised in Japan, working in Silicon Valley, sent back to Japan as an expat. Founder of Innovators in Japan.

bb0110fb9cd

shoheinarron

77

378

20,181,104

null

0

null

0

null

2017-12-15

2017-12-15 13:34:14

2017-12-15

2017-12-15 13:36:38

0

false

en

2017-12-15

2017-12-15 13:36:38

2

14f0c43ac558

0.716981

0

This post is dedicated to the Human Resource Community that is wondering how to grapple with the coming Tsunami of Artificial Intelligence…

3

Artificial Intelligence and the Human Resource Community This post is dedicated to the Human Resource Community that is wondering how to grapple with the coming Tsunami of Artificial Intelligence that is expected to sweep across the world soon. The workforce will need to be up skilled and job losses can be expected. On top of that, HR maybe called upon to deal with a mixed workforce comprising of humans and BOTs. The good news is that AI is slowly moving from the lab to the field and will soon be a part of every professional’s toolkit. Recenty I tried working on the HR use case published on the IBM Watson Blog (https://www.ibm.com/communities/analytics/watson-analytics-blog/watson-analytics-use-case-for-hr-retaining-valuable-employees/ ) with my favorite Machine Learning tool TFLearn ( TFLearn | TensorFlow Deep Learning Library ) and I got good results. I would like to encourage all HR professionals to try cracking this use case on their own. This will convince you that there is no black magic and everyone in the company can learn Machine Learning and start thinking of use cases that can benefit the company.

Artificial Intelligence and the Human Resource Community

0

artificial-intelligence-and-the-human-resource-community-14f0c43ac558

2017-12-15

2017-12-15 13:36:38

https://medium.com/s/story/artificial-intelligence-and-the-human-resource-community-14f0c43ac558

false

190

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Sudhir Gupta

null

db575b06f919

sudhir.g

1

20,181,104

null

0

IMAGE_SIZE=224 ARCHITECTURE="mobilenet_0.75_${IMAGE_SIZE}" python -m scripts.retrain — bottleneck_dir=tf_files/bottlenecks — how_many_training_steps=5000 — model_dir=tf_files/models/”${ARCHITECTURE}” — summaries_dir=tf_files/training_summaries/”${ARCHITECTURE}” — output_graph=tf_files/retrained_graph.pb — output_labels=tf_files/retrained_labels.txt — architecture=”${ARCHITECTURE}” — image_dir=tf_files/dataset python -m scripts.retrain -h

3

null

2018-01-03

2018-01-03 05:35:32

2018-01-20

2018-01-20 06:56:20

5

false

en

2018-06-24

2018-06-24 21:37:32

9

14f2792f64c1

5.395597

17

3

0

Applying transfer learning on a pre-trained multi-layer perceptron for image classification, and deploying resulting model as an Android…

5

Training a neural network using Mobilenets in TensorFlow for image classification on Android First, a few definitions for the uninitiated. TensorFlow is an open-source library for numeric computation using dataflow graphs. It was developed by Google brain team as a proprietary machine learning system based on deep learning neural networks. TensorFlow is causing quite a stir in research and development field and is set very to make its way into mainstream machine learning. About TensorFlow In this tutorial, we are going to make an Android app that uses a neural network trained by TensorFlow. For our purpose, we will use a special class of convolutional neural networks called MobileNets. As also suggested by the name, the special thing about the MobileNets is that they are “mobile-friendly”, meaning that they are optimized to be executed using minimal possible computing power on a smartphone device. As you can imagine, there has to be a catch with something so customized for a minimal resource footprint. MobileNets do not provide as good of an accurate model as produced by a full-fledged deep neural network. However, the accuracy is surprisingly very high and good enough for many applications. Below graph shows the graph of accuracy versus the number of calculations required for a choice of neural network libraries. Model size is shown as the size of circle, and different model size of the MobileNets show the tradeoffs assumed during selection of model parameters. See this Google research blog post for more details. We will also come back to the choice of model parameters later. Accuracy (y-axis) vs. the number of required operations (x-axis) for available configurations Data collection The image classifier that we are going to train in this example will be able to classify cars to their respective model. We are going to use “Cars Dataset“ made available by Stanford. [Link to dataset ~ 1.8GB] This dataset consists of 16,185 images of 196 classes of cars. The label information for this dataset provides make, model and year for each of the 196 classes. The data is also available as separate train and test sets in a 50–50 split. But for this project, we will download the full dataset of 16,185 images, along with the label information. Data Preprocessing The image dataset from the Stanford is organized as a single directory containing 16,185 images of cars. To use these images for our training step, we need to reorganize these images so that each car image is inside a directory that contains all the images for a single class. Also, the name of the directory should reflect the name of the corresponding class. We will use the label information available in the .mat file provided by Stanford. Below Python script performs this task. Edit: As correctly pointed out by a reader, I missed specifying that the name for class 174, i.e. ‘Ram C/V Cargo Van Minivan 2012’ has a forward slash, ‘/’, in it and it should be renamed to remove the slash character. Organize the data to be used for training the classifier Training the classifier For training our image classifier, we are going to use the transfer learning concept. Transfer learning basically refers to a supervised learning technique that takes advantage of an already existing trained model that solves a similar problem. For our purpose, we will take TensorFlow’s fully trained model for Imagenet and retrain just the last layer of the neural network on our Cars dataset. Though this approach is not as powerful as a fully trained model, but it can provide a surprisingly high accuracy for most tasks that are related. You can read more about this concept from this article: How to Retrain Inception's Final Layer for New Categories | TensorFlow The script can take thirty minutes or more to complete, depending on the speed of your machine. The first phase…www.tensorflow.org As mentioned in the above article, you can clone the retrain scripts from this GitHub repository. As we are only training the final layer of the neural network, the training will end in reasonable amount of time. TensorFlow’s retraining procedure allows you to optimize the training procedure by tweaking certain parameters. Following two are probably the most important of those parameters: Input image resolution: The corresponding value can be 128,160,192, or 224px. As you can imagine training with a higher resolution image will take longer time, but also has higher chances of providing a better classification accuracy. Since, we are only training final layer and our dataset is not very huge, we will keep this value as 224. Relative model size: This value represents the relative size of the model as a fraction of the largest MobileNet. It can take value such as 1.0, 0.75, 0.50, or 0.25. The larger the size of the model, more accurate it will be. For our purpose, we will keep this value to 0.75. Both the above parameters can be configured as environment variables as: With these parameters setup, let’s run the retrain python script provided by TensorFlow with following parameters. Above command sets up directory paths for bottleneck, model and summary files. image_dir refers to the directory where our image image data is stored. output_graph and output_labels provides the path where we will store our training model and label information respectively. Parameter how_many_training_steps is the count of how many times retraining iterates over the data. By default it is set to reiterate 500 times. If you have time, you could increase this iteration count to achieve better results. There are still a lot more configuration options that the retraining script provides. Run the below command help command to read more about the available options. If the final size of your trained model is large, you could look into some of the techniques that can reduces the model size, such as making the model compressible and to quantize the network weights. Refer to below article to read more on this. How to Quantize Neural Networks with TensorFlow | TensorFlow When modern neural networks were being developed, the biggest challenge was getting them to work at all! That meant…www.tensorflow.org Deploying the solution in Android app TensorFlow GitHub repository contains an Android project that you can directly load into your Android studio and compile directly to create an image classifier application. Simply copy the trained model(graph.pb) and label information(labels.txt) that we generated in the last step, to the project’s “assets” directory And run the Gradle build. This will generate a .apk file that you can run on your computer in a emulator with camera access, or you can deploy it on an Android phone. Result Android app in action The resulting Android app uses the phone’s camera stream to classify objects into the identified class. The application by default also provides probability value of a car belonging to the corresponding category. You could take a look at the Android application that I published on Google Play store to see this project in action. Which Car Is That ? - Android Apps on Google Play Quickly find the car model using your phone's cameraplay.google.com As you can see transfer learning with TensorFlow makes it easy to quickly build our own classifiers. I hope you liked this article, and I’ll see you in the next one. :)

Training a neural network using Mobilenets in TensorFlow for image classification on Android

71

training-a-neural-network-using-mobilenets-in-tensorflow-for-image-classification-on-android-14f2792f64c1

2018-06-24

2018-06-24 21:37:32

https://medium.com/s/story/training-a-neural-network-using-mobilenets-in-tensorflow-for-image-classification-on-android-14f2792f64c1

false

1,209

null

Machine Learning

machine-learning

Machine Learning

51,320

Sumit Kumar Arora

Student@University of Chicago | I have an Uber passenger rating of 4.9.

93b61e3f97aa

sumit.arora

37

7

20,181,104

null

0

null

0

null

2018-03-05

2018-03-05 23:08:37

2017-11-02

2017-11-02 00:00:00

1

false

en

2018-03-05

2018-03-05 23:09:31

10

14f46d9921d

3.373585

0

I’ve said it before, but it bears repeating: those of us in learning and development can learn a lot from marketing.

5

What do we want? Learning! When do we want it? Right now! I’ve said it before, but it bears repeating: those of us in learning and development can learn a lot from marketing. Whereas marketers may be interested in changing the behaviors of consumers of beer or technology, our audiences are consumers of knowledge. Whether it’s marketers trying to reach consumers or an L&D team trying to reach employees, the end goal is the same. Change behaviors. The five-year gap between marketing and learning L&D consistently adopts technology and methodologies from great advertisers and marketers. Trouble is, L&D tends to run about five years behind the curve. Gamification is a prime example. Gamification hit peak popularity in the marketing world around the early 2010’s. It’s only been the last three years that Mindspace has seen a significant uptick in the number of organizations searching for gamified training, recruitment and retention solutions. One of the ways L&D can close this gap — or better yet, get ahead of the curve — is by studying some of the latest trends in advertising, reviewing new studies and applying marketing insights to their learning programs. And that’s just what I’ll be doing over the next few weeks. Micro-moments are big — really big To kick it off, let’s take a look at a recent blog post from Google on micro-moments, a term the search-turned-advertising-turned-everything giant coined about two years ago. According to Google, micro-moments are “…the moments when we turn to a device — often a smartphone — to take action on whatever we need or want right now.” Learning teams have already begun to make the switch to mobile-first experiences (though, again, that change has been exceedingly slow). We understand that people spend nearly three hours a day on their smartphonesand 43% of employees spend time working remotely, according to Gallup. The impending mobile revolution in learning and development is key to paving the way for what I think is an even more monumental shift: the idea of “right here, right now” learning. A few months back, my team met with a consulting group focused on innovating new ventures. Having recently spun off of a long-standing parent group, they are interested in shaking off the shackles of traditional corporate learning. They want to build a culture of learning — an environment in which learning occurs naturally and is not driven by a top-down approach, instead operating in a peer-to-peer manner with L&D curating the “best of” learning moments. There’s only one challenge. A big, hairy challenge. That 10-foot behemoth that plagues so many of our organizations. Nobody has time. Everyone is too busy doing their jobs to take the time to learn new ways to make their jobs better. To create a culture of learning in our organizations, having a “right here, right now” focus is imperative. We can no longer ask workers to set aside an hour here or a day there, interrupt their work and come learn something that may or may not have immediate application to their jobs and careers. We need to create a paradigm that enables and facilitates learners to gain relevant, timely information at exactly the moment they need it. (Many organizations have tried to solve for this need for immediate information through creating a company wiki. While the thinking is in the right direction, they are unwieldy, seldom visited and require vast amounts of time and effort to keep current.) The future is almost here According to the blog published by Google’s VP of Marketing for Americas, “Smartphone users are 50% more likely to expect to purchase something immediately while using their smartphone compared to a year ago.” As consumers, we are already training ourselves to not only want information immediately, but also take action immediately. When viewed through the lens of learning, we can see what this statistic holds for the future of learning and development. Our learners will increasingly come to demand on-demand learning that is timely, useful, current and actionable — all in the time it takes to go from one meeting to the next. We can’t take five years to get there. We have to start now. At Mindspace, in addition to diving into VR and AR for learning, we’ve begun experimenting with integrating artificial intelligence and virtual assistants like Alexa and Cortana into Fathom, our learning platform. We envision a day soon when learners will be able to ask relevant, timely questions about the task at hand and get immediate spoken answers with an opportunity to dive deeper into a subject. We’re not satisfied with playing catch-up. The days when marketing will look to learning and development for “what’s next” are just around the corner. Words by Brandon Marsala, Creative Director, Content at Mindspace. Illustration by Kyle Davila, Art Director at Mindspace. Originally published at www.mindspace.net on November 2, 2017. Stay connected with us on LinkedIn, Instagram and Facebook or send us an email at sayhello@mindspace.net.

What do we want? Learning! When do we want it? Right now!

0

what-do-we-want-learning-when-do-we-want-it-right-now-14f46d9921d

2018-03-05

2018-03-05 23:09:31

https://medium.com/s/story/what-do-we-want-learning-when-do-we-want-it-right-now-14f46d9921d

false

841

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Mindspace

Creators of branded learning experiences that engage and motivate today’s employees and customers. Learn more at mindspace.net.

41505ffbf6db

mindspaceagency

44

29

20,181,104

null

0

null

0

null

2017-12-05

2017-12-05 02:02:51

2017-12-13

2017-12-13 08:53:07

9

true

en

2017-12-14

2017-12-14 02:08:50

30

14f5c7bef3d8

4.113208

2

1

0

Last week, we had a great time at a Sunnyvale elementary school while promoting CS first — Hour of Code 2017. We ended up playing…

5

Gaming The Future By Your Super Power Last week, we had a great time at a Sunnyvale elementary school while promoting CS first — Hour of Code 2017. We ended up playing CodeMonkey instead because they only have iPads. Which does not support Flash. The usual “bullshit” is CS gives you super power. But, it’s indeed increasing more true each day. Of cause, don’t take it literately. Unite the Justice League with Android Pay To all responsible parents: try HoC2017 with your kids. BTW, could someone please port it to a Mobile Native App, please? It’s 2017 not 2007, right ;p csfirst.withgoogle.com/hoc2017-teachers It All Starts From A Game www.trs-80.org/galaxy-invasion Gaming has always been one of key factor pushing personal computing technology forward since the beginning. For example, TRS-80 (8bit CPU with 4 K RAM), one of “the earliest mass-produced and mass-marketed retail PC” debuted 40 years ago. Hobbyist was one of its key customer based, and Big Five SW turned Fun and Games into Million-Dollar business. Of cause, it is no comparison with whatever mobile phone you have today. Anyway, I still remembered when I first played Galaxy Invasion loaded from a cassette. It was really cool and fun back then. oldcomputers.net/indexwp.html TSR-80 Today, everyone plays. In 2017, the global games SW revenues is $116.0 billion according to Newzoo. Whereas, US Military spending was 599 billion in 2015. It is obvious without the gaming market, Information Technology won’t advance this quick and accelerating. Beside, making life less boring — priceless. www.asymco.com/2013/11/18/seeing-whats-next-2/ by Horace Dediu Furthermore, before Bitcoin Mining or Machine Learning hypes, GPU vendors, such as NVIDIA makes their fortune form Gamers. The fact is you can not build a GPU company over night or years. For example, I never know a real gamer, who will be happy to settle down with an Intel GPU. www.google.com/search?tbm=fin&q=nvidia&wptab=COMPARE#smids=/m/07zkq3x&wptab=COMPARE Even Smartphone penetration is getting really high, gaming is going to pull the same magic as it’s doing in PC market. It will be even more fun for Smartphone, because there are significantly difference from PC, such as: Consumers replace Smartphone more often: 2–3 years Smartphone is at least 5x bigger: 1.5b mobile vs 260m PC on the annual shipment. Gaming The Way To Solve Any Problem Or provide a proxy at least. There was a story of Lydians fought hunger by games: “on one of the days they would play games all the time in order that they might not feel the want of food, and on the next they ceased from their games and had food”. Today, it maybe still a good idea, but irony to fight obesity ;p Wall-painting of game-players from a bar on the Via di Mercurio An education scientist, Sugata Mitra talked about the child-driven education. He did interesting experiments by trowing a old PC from New Delhi to South Africa to Italy, and shown kids can learnt automatically and independently. There is also study suggests The Potential for Game-Based Learning to Improve Outcomes for Nontraditional Students. Furthermore, how to leverage Game mechanism to achieve Flow is a popular topic in education today. As Jane Mcgonigal puts it: Gaming can make a better world. www.ted.com/talks/jane_mcgonigal_gaming_can_make_a_better_world The Whole New Work As Automation advances, human can move on to act more scalable role or do high value tasks. For example, it is fair to say flying a drone is not like a video game. Nevertheless, there are clearly a lot of benefits even if using video games as a training tool. The trend: the skills are needed for such job clearly leans toward more on video gaming in the future. Furthermore, it will inevitable to be a part of solution on pilot shortage getting worse. The crazy thing is the Top Gun 2.0 might be flying with Drone Swarm in the future. It will be hard, if not impossible for classical Top Guns or a fleet without such Tech. to complete with Top Gun 2.0. This is not even an isolated case. Wherever you check, you will see a similar pattern. Of cause, unless you happen to check a job market that labor is so cheap, so no employer has motivation to invest on Automation. scout.com/military/warrior/Article/Air-Force-Fighter-Jets-Will-Control-Drones-101455147 Furthermore, game is also an important part of ML as least on two aspects. Video games have been used as a playground for ML research, e.g. Deepmind: Playing Atari with Deep Reinforcement Learning. Besides, the cost of each training cycle in a virtual world is much less than a physical world. There is less restriction too, such as time. As ML today is driven by data and based on statistical model. You can not really “test” it via existing QA methods. Simulation is likely the best substitute. Which can leverage many stuffs from simulation games. So, what’s your game?

Gaming The Future By Your Super Power

2

gaming-the-future-by-your-super-power-14f5c7bef3d8

2018-03-26

2018-03-26 16:33:10

https://medium.com/s/story/gaming-the-future-by-your-super-power-14f5c7bef3d8

false

772

null

Gaming

gaming

Gaming

52,172

Sam Lin

null

b14f6537743d

samlin001

6

7

20,181,104

null

0

encode = {'team1': {'MI':1,'KKR':2,'RCB':3,'DC':4,'CSK':5,'RR':6,'DD':7,'GL':8,'KXIP':9,'SRH':10,'RPS':11,'KTK':12,'PW':13}, 'team2': {'MI':1,'KKR':2,'RCB':3,'DC':4,'CSK':5,'RR':6,'DD':7,'GL':8,'KXIP':9,'SRH':10,'RPS':11,'KTK':12,'PW':13}, 'toss_winner': {'MI':1,'KKR':2,'RCB':3,'DC':4,'CSK':5,'RR':6,'DD':7,'GL':8,'KXIP':9,'SRH':10,'RPS':11,'KTK':12,'PW':13}, 'winner': {'MI':1,'KKR':2,'RCB':3,'DC':4,'CSK':5,'RR':6,'DD':7,'GL':8,'KXIP':9,'SRH':10,'RPS':11,'KTK':12,'PW':13,'Draw':14}} matches.replace(encode, inplace=True) df['winner'].hist(bins=50) import matplotlib.pyplot as plt fig = plt.figure(figsize=(8,4)) ax1 = fig.add_subplot(121) ax1.set_xlabel('Team') ax1.set_ylabel('Count of toss wins') ax1.set_title("toss winners") temp1.plot(kind='bar') ax2 = fig.add_subplot(122) temp2.plot(kind = 'bar') ax2.set_xlabel('Team') ax2.set_ylabel('count of matches won') ax2.set_title("Match winners") #Import models from scikit learn module: from sklearn.linear_model import LogisticRegression from sklearn.cross_validation import KFold #For K-fold cross validation from sklearn.ensemble import RandomForestClassifier from sklearn.tree import DecisionTreeClassifier, export_graphviz from sklearn import metrics #Generic function for making a classification model and accessing performance: def class_model(model, data, predictors, outcome): model.fit(data[predictors],data[outcome]) predictions = model.predict(data[predictors]) accuracy = metrics.accuracy_score(predictions,data[outcome]) print('Accuracy : %s' % '{0:.3%}'.format(accuracy)) kf = KFold(data.shape[0], n_folds=5) error = [] for train, test in kf: train_predictors = (data[predictors].iloc[train,:]) train_target = data[outcome].iloc[train] model.fit(train_predictors, train_target) error.append(model.score(data[predictors].iloc[test,:], data[outcome].iloc[test])) print('Cross validation Score : %s' % '{0:.3%}'.format(np.mean(error))) model.fit(data[predictors],data[outcome]) model = RandomForestClassifier(n_estimators=100) outcome_var = ['winner'] predictor_var = ['team1', 'team2', 'venue', 'toss_winner','city','toss_decision'] classification_model(model, df,predictor_var,outcome_var) #feature importances: If we ignore teams, Venue seems to be one of important factors in determining winners #followed by toss winning, city imp_input = pd.Series(model.feature_importances_, index=predictor_var).sort_values(ascending=False) print(imp_input) import matplotlib.pyplot as mlt mlt.style.use('fivethirtyeight') df_fil=df[df['toss_winner']==df['winner']] slices=[len(df_fil),(577-len(df_fil))] mlt.pie(slices,labels=['Toss & win','Toss & lose'],startangle=90,shadow=True,explode=(0,0),autopct='%1.1f%%',colors=['r','g']) fig = mlt.gcf() fig.set_size_inches(6,6) mlt.show() import seaborn as sns team1=dicVal['CSK'] team2=dicVal['RCB'] mtemp=matches[((matches['team1']==team1)|(matches['team2']==team1))&((matches['team1']==team2)|(matches['team2']==team2))] sns.countplot(x='venue', hue='winner',data=mtemp,palette='Set3') mlt.xticks(rotation='vertical') leg = mlt.legend( loc = 'upper right') fig=mlt.gcf() fig.set_size_inches(10,6) mlt.show() #'team1', 'team2', 'venue', 'toss_winner','city','toss_decision' team1='DC' team2='DD' toss_winner='DC' input=[dicVal[team1],dicVal[team2],'23',dicVal[toss_winner],'14','0'] input = np.array(input).reshape((1, -1)) outcome=model.predict(input) print(list(dicVal.keys())[list(dicVal.values()).index(outcome)]) #find key by value search output #output: #DD

75

null

2018-04-11

2018-04-11 08:51:38

2018-04-11

2018-04-11 08:57:01

7

false

en

2018-04-11

2018-04-11 08:57:01

2

14f61aed7b97

5.872642

2

1

0

Learn how to apply artificial intelligence and predictive modeling techniques to predict outcomes of cricket matches based on venue…

5

Predicting the Outcome of Cricket Matches Using AI Learn how to apply artificial intelligence and predictive modeling techniques to predict outcomes of cricket matches based on venue, players, toss winner, and more. In this article, fundamental concepts of analytics and predictive modeling to IPL cricket matches will be applied to get meaningful information and predictions. Teams, matches, and factors affecting outcomes of matches will be analyzed. Some factors that affect match outcomes could be venue (stadium), city, toss winner, and toss decision (field/bat). Python 3+ has helpful analytics, predictive, and charting libraries. Libraries we’ll focus on today include linear algebra (numpy), data processing for CSV (pandas), charting (MatPlotLib), statistical data visualization (seaborn), and machine learning modeling (scikit-learn). The following steps should be followed to set up an Azure environment for Jupyter notebook: Provision Azure HDInsight cluster using Spark with linked Azure Storage blob container. Upload source data matches.csv to linked Azure Storage blob container using Azure Storage explorer. Launch Jupyter notebook from the HDInsight cluster blade. Under Quick links, click Cluster Dashboards. On the dashboard, click Jupyter notebook to enter your cluster login name and password. Click Upload to upload this file. Select the kernel as Python 3.6. import numpy as np # linear algebra import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv) from pyspark.sql.types import * #if you are using https://notebooks.azure.com, then upload in Jupyter notebook itself then use #matches=pd.read_csv(‘../data/matches.csv’) #else Refer to Azure storage path to read CSV file matches = spark.read.csv(‘wasb:///data/matches.csv.csv’, inferschema=true) matches.info() First, we address missing data with a process called impute. There are numerous ways to fill missing data based on imaginary scenarios. Let's consider match-related columns -team1, team2, city, toss_decision, toss_winner, venue, and winner. We see that there are missing values in columns cityand winner. Column city was updated manually based on venue details. Column winner was updated with the value draw. Later, label each of team names with short abbreviations and then encode them as numerical values for predictive modeling purposes, as shown below. We shall see the table output format as shown below. In first row, team1 vs. team2 is 2 (KKR) vs. 3 (RCB) played in M Chinnaswamy stadium in Bangalore city with the toss won by 3 (RCB). 3 (RCB) chose to field first. The match outcome was that (2) KKR won. Here’s the code to compute a team-wise graphical representation of total match wins using a histogram: Output: Toss winners most likely choose to field first in a 20 overs IPL match. Perception is that team choosing to field first and later chase the runs are most likely to win. To find if toss winners are most likely match winners? To find a correlation between toss and match winners, below code is used. Output: From the graph, most matches are won by 1 (MI) who also won most tosses. But this is not same for 2 (KKR) and 3 (RCB). With further discussions, it shall be noted that toss winning is not most important feature in match winning. The scikit-learn open-source library provides machine learning in Python. This library requires all inputs to be numeric. All categorical variables should be converted into numeric variables using the concept of encoding with scikit-learn LabelEncoder. Later, a predictive model is created using a generic function called class_model that takes parameters model (algorithm), data, predictors input, and outcome predictable feature. Be aware of unexpected indent errors in Python while re-using the below code. The k-fold cross (k=5) validation technique is used to reserve a sample set on which we do not train the model but it will be used to test the model before finalizing. Mean score error is used to determine the evaluation. Standard deviation could also be used for evaluation. More predictor variables may result in unseen training data. This unseen training data results in overfitting. The user needs to balance the training set and predictor variables based on the accuracy and cross-validation score. A multiple model classifier was tested for given data. The RandomForestClassifier model showed an acceptable accuracy percentage of ~90%. Now, the model is trained using data frame (dataset) containing predictors variables like team1, team2, venue, toss_winner, city, and toss_decision to determine outcome variable winner. RandomForestClassifier also provides important features, like a matrix specifying numeric percentage influences for each of the predictor variables. Output: If we ignore team2 and team1, venue seems to have a higher value compared to toss_winners and toss_decision. This shows that venue is more important feature. toss_decision to field or bat first is the least important feature, comparatively. Let's plot a graph from the dataset to see if toss_winner is also match winner from the dataset. Output: From dataset, toss_winners were also the match winner 50% of the time and it is not enough to determine the winner. Let’s consider the top two winning teams CSK and RCB and analyze the number of matches won against each other and how venue influenced their win? From the histogram bar chart, we saw that CSK won 79 matches and RCB won 70 matches. RCB is now compared with CSK on the number of matches won in different venues. Output: In the above graph, 5 (CSK) won six matches against one match won by 3 (RCB) in their home turf venue of 15 (MA Chidambaram Stadium, Chepauk). RCB won all of the matches in 18 (New Wanderers Stadium) and 13 (Kingsmead stadium). When 1 (MI) is compared with 5 (CSK), 1 (MI) have won more matches in their home turf 34 (Wankhede stadium) against 5 (CSK), as shown in the below graph. Clearly, venue is a more important feature compared to toss winner. The model is now ready for prediction. Below is the input to the model. The outcome variable will predict the winner. Conclusion Data-driven predictive models could be a way forward in IPL team management. Data-driven recommendations could also be developed for player selection. Predictive analytics could seek to pick probable winners and help manage risk better. Analytics bridges the gap between team managers and team coaches. These data insights and quantifications provide precise and timely answers. These compelling charts, reports, and predictive models can be automated for continuous updates by streaming input data.

Predicting the Outcome of Cricket Matches Using AI

4

predicting-the-outcome-of-cricket-matches-using-ai-14f61aed7b97

2018-05-25

2018-05-25 18:03:01

https://medium.com/s/story/predicting-the-outcome-of-cricket-matches-using-ai-14f61aed7b97

false

1,278

null

Machine Learning

machine-learning

Machine Learning

51,320

lakshya

open sources (tech geek)

fe2b4cd6d2d6

xtaraim

48

201

20,181,104

null

0

null

0

a53d3ddf097

2017-09-19

2017-09-19 23:50:34

2017-09-20

2017-09-20 00:03:41

0

false

en

2017-09-20

2017-09-20 00:05:41

0

14f6b7c3d62a

2.34717

1

0

This is perfectly machine readable data:

2

Machine readable data needs to be human useable too This is perfectly machine readable data: Month,Year,Region,Value 1,2016,Auckland,1000 1,2016,Wellington,2000 3,2017,Auckland,1500 4,2017,Wellington, 5,2017,Auckland,1.2 R, Python, Excel, STATA, etc, will all have no trouble reading such a CSV data file. Any human trying to use this data will have many questions: What do the values represent and where did they come from? The value on the 4th row is missing. What could cause a missing value? Is it an error or has it been censored for some reason (confidentiality etc)? The value on the 5th row looks suspiciously small compared to other values. What is the typical range of values that we should expect? What could cause a very small value? Are the years calendar years, financial years, years ending June, or something else? The data appears to be monthly. Are the values totals or averages or some other statistic for each month? Or are they instantaneous observations at a particular point in time each month (eg the 15th of the month)? The data frequency seems to be irregular. What’s the reason for that? Are entire rows actually missing from this data? The data also appears to be for geographic areas. How exactly are these areas defined? Do they follow standard geographic definitions that would allow this data to be joined to other geographic data? Or if non-standard areas have been used, what are they? So this data is perfectly readable by a machine but it is not useable by a human. Since ultimately almost everything that machines do is decided by humans, for data to be useful it needs to be both machine readable and human usable. This applies to open data and data used within organisations. Human usability requires careful documentation of the data’s characteristics and quirks. This needs to be recorded separately from the data (ie as metadata and/or documentation) but be easily findable and accessible by all users of the data. It should be in a single place or file, not scattered about, and definitely not only stored in people’s heads. People should be able to find documentation and metadata in the same place that they find data itself. They shouldn’t have to go hunting for it elsewhere on a website or server. Some metadata can also be machine readable, eg whether the years are calendar or financial years could be recorded in a standard format and the machine could “understand” this when reading the data. However, in almost all cases what the machine does with the data still has to be specified by a human so ultimately a human needs to understand the metadata too. And some of the more features a dataset, such as the process by which it was collected, are best recorded as free text that will be difficult for a machine to “understand” anyway. In other words, you can’t write the human out of the equation (not yet …). Making data machine readable is largely a mechanical process of ensuring it conforms to appropriate standards. Making that data also human useable is more difficult. It requires thinking about and answering the types of questions listed above. If you are already quite familiar with a dataset, it may be hard to know which features are not obvious to a newcomer. Making data human useable can be a tedious and boring process, but without this work, data is not valuable. It often seems like data providers devote too many resources to technical solutions to make their data machine readable, like complicated APIs, while devoting too few resources to metadata and documentation. In many cases, data would be more valuable and would get more use if the technology for sharing it was simpler but the documentation was better.

Machine readable data needs to be human useable too

1

machine-readable-data-needs-to-be-human-useable-too-14f6b7c3d62a

2018-05-29

2018-05-29 02:46:31

https://medium.com/s/story/machine-readable-data-needs-to-be-human-useable-too-14f6b7c3d62a

false

622

Economist and Data Scientist

null

Aaron Schiff’s blog

null

aaron-schiffs-blog

DATA SCIENCE,ECONOMICS,DATA VISUALIZATION

aschiff

Open Data

open-data

Open Data

5,306

Aaron Schiff

Economist and data scientist; http://schiff.nz

fc2d412d0561

aschiff

159

173

20,181,104

null

0

null

0

null

2017-10-25

2017-10-25 15:37:56

2017-10-25

2017-10-25 18:43:06

0

false

en

2017-10-25

2017-10-25 22:29:15

0

14f6ed556943

1.226415

3

0

Before I started the Masters I told myself I would write at least one post per week. The posts needed to be related to what I was studying…

2

Master’s weeks three and four. Before I started the Masters I told myself I would write at least one post per week. The posts needed to be related to what I was studying or working on, or personal but still related to the sabbatical year and/or data science.I’m not a blogger, I’ve never kept a diary, I’m not very good at taking notes and have never been, even in Uni. I’ve always been more a doodling, messy notes kind of person. So this was going to be my routine so that I could look back in a year’s time and be able to judge in more quantitative terms what had been done during this year. We tend to forget a lot of the detail and this was a commitment I made to myself a couple of months ago. But here we are, I’ve been snoozing my alarm to write a post for over a week now. This is quite annoying… I have to recognise that the past two weeks have not been easy. Being a mature student means having a life at home I cannot just throw away, family responsibilities take time and you want to be actually present, not just there. At the same time there’s so much to learn. For example this week we’ve had simultaneously released coursework to do with Hadoop for one module, Statistics and R for another, and Matlab (and Weka) with some Machine Learning for another two. And then there’s the constant looking over my shoulder towards next year, jobs, connections, PhD, funding, etc. I’m really trying to be good at not planning ahead too much, since otherwise I fear pigeon-holing myself into the data scientists others might expect me to be or jobs require me to be while keeping things realistic. But it’s a hard balance to strike. For now back to Hadoop, R, Matlab and Weka… hope things will settle a bit once assignments are handed.

Master’s weeks three and four.

3

masters-week-three-and-four-14f6ed556943

2017-10-26

2017-10-26 13:13:31

https://medium.com/s/story/masters-week-three-and-four-14f6ed556943

false

325

null

Data Science

data-science

Data Science

33,617

Liev Garcia

null

2f679794b3b7

lievgarcia

9

46

20,181,104

null

0

null

0

5edd6742fe7

2018-04-30

2018-04-30 15:05:16

2018-04-30

2018-04-30 15:11:21

4

false

en

2018-04-30

2018-04-30 15:11:21

3

14f75d696a32

3.564151

1

0

One of the great challenges when developing conversational interfaces is to make chatbots more human. So, we asked ourselves the question…

5

Making chatbots more human using machine learning One of the great challenges when developing conversational interfaces is to make chatbots more human. So, we asked ourselves the question: would it be possible to create a bot that is capable of imitating a person in real life? Given enough data, chatbots are able to learn and understand human languages. This is done by techniques like NLP (Natural Language Processing) and Machine Learning. A conventional chatbot tries to understand the intent of a user and provides a fixed response based on their intent. The chatbot concept that we have created is slightly different. Instead of providing fixed responses to recognized intents, our social chatbot responds similarly to how a real person would. For our Proof of Concept, I’ve tried to imitate Vasilis van Gemert from his excellent and worth listening to podcast: The Good, The Bad, and the Interesting. The transcriptions of these podcasts were pre-processed in such a way that it suits the machine learning algorithm. Next I used Tensorflow to build the model. TensorFlow TensorFlow is an open-source machine learning library developed by Google Brain Team and released in November 2015. The idea of TensorFlow is to express numeric operations as a graph. The nodes in the graph represent mathematical operations and the edges represent the data. The machine learning model we used for our chatbot is a deep neural network. Deep neural networks are commonly taught and visualized as graphs, which makes their implementation in TensorFlow more natural for machine learning practitioners. Figure 1: Visualization of a TensorFlow graph The model The deep neural network we used for our concept is known as a sequence to sequence (seq2seq) model. A seq2seq model is able to learn vocabulary, sentence structure and more all from input and output data. It contains two main components: an encoder and a decoder. The encoder processes the input and the decoder generates an output. The goal is then to tune the model in such a way that the output of the model is similar to how Vasilis responds. In order to do this, we trained the model with thousands of lines of conversational data from all podcasts so far. Figure 2: seq2seq encoder decoder model architecture The results In the initial phases of training, the chatbot was simply outputting “Ja” (yes) to every input we gave. Although a funny coincidence, this makes some sense intuitively since “Ja” is a commonly used word in the Dutch language. Gradually, the chatbot was starting to form more complete sentences and recognized basic patterns of Dutch grammar. Figure 3: Initial phases of training. Lines starting with > denotes input, [..] the output. Figure 4: Training after 140000 iterations. I continued to train our model until we saw no performance increase. For our final model, the responses are alright, but not as good as we hoped them to be. We noticed that the responses are often out of context. The chatbot has some understanding of the language but is not able to form coherent thoughts. It has the tendency to repeat words and sometimes struggles to finish its own sentences. It has some idea of sentence structure and interestingly enough, occasionally responds in a mixture of English and Dutch. Improvements and experiences We gained a lot of insights by implementing our own deep learning model in TensorFlow. We also looked at ways to improve the performance of our chatbot. Our insights are summarized as follows: The task at hand is a challenging one. For each input you can have thousands of acceptable responses. It is difficult to score these responses in an objective way. Current state-of-the-art performance for chatbots of these types are far from human like performance. Data is key. With more data we are confident that we can produce a better chatbot experience. Penalizing responses that are not fully complete or contain repetition of words may improve performance. Spending more time tuning the hyperparameters such as number of iterations, choice of optimizer, learning rate, batch size, can lead to better performance. Conclusion We experimented with seq2seq models in TensorFlow to build a chatbot that imitates a real person. We obtained decent results but not as good as we had hoped for. We learned more about deep learning and how that is applied to natural language processing problems. We realized that a huge amount of data is key for machine learning applications and are aware of the limitations of this technology. Custom trained chatbots have a long way to go before they can reach human level performances, but AI has enabled us to make the first step towards this goal. Let’s keep experimenting… This article is written by Tiamur Khan, Solution Developer at Mirabeau Originally published at blog.mirabeau.nl.

Making chatbots more human using machine learning

2

making-chatbots-more-human-using-machine-learning-14f75d696a32

2018-04-30

2018-04-30 15:11:23

https://medium.com/s/story/making-chatbots-more-human-using-machine-learning-14f75d696a32

false

759

AHEAD IN A DIGITAL WORLD - Mirabeau is a full-service digital agency. By combining design, technology and insights, we create top-notch digital solutions - A Cognizant Digital Business.

null

MirabeauNL

null

Mirabeau

info@mirabeau.nl

mirabeau

TECHNOLOGY,DIGITAL DESIGN,DIGITAL TRANSFORMATION,CLOUD,INNOVATION

_Mirabeau

Machine Learning

machine-learning

Machine Learning

51,320

Mirabeau

AHEAD IN A DIGITAL WORLD - We create world changing digital experiences. A @Cognizant Digital Business. Follow our stories: www.medium.com/mirabeau

714dbf9174c1

Mirabeau

17

48

20,181,104

null

0

null

0

d814a4f4e355

2018-01-22

2018-01-22 16:10:34

2018-02-07

2018-02-07 19:07:18

9

false

en

2018-02-08

2018-02-08 05:13:19

39

14f92d0ddb63

16.003774

67

0

Written by Michael Bereket and Thao Nguyen

5

How Different Are Cats and Cells Anyway? Closing the Gap for Deep Learning in Histopathology Written by Michael Bereket and Thao Nguyen. Deep learning has revolutionized the field of computer vision. So why are pathologists still spending their time looking at cells through microscopes? Examples of cat detection and nucleus detection (bounding boxes in green, yellow, red) (Source: cats, cells) In recent years, the field of computer vision has undergone a revolution. Deep learning techniques, powered by increases in data and improvements in computational power, have enabled breakthroughs in tasks like image classification (Krizhevsky et al., 2012) and facial recognition (Taigman et al., 2014), which now permeate our everyday lives. These increasingly ubiquitous breakthroughs are just beginning to reach pathologists’ microscopes. Histopathology, the microscopic study of diseased cells and tissues, presents a variety of unique opportunities to meaningfully apply deep learning. Every day, countless tissue samples must be visually inspected to diagnose and characterize a variety of illnesses, including nearly all types of cancer (Gurcan et al., 2009). Additionally, developments in whole-slide imaging technology have fueled the growth of digital pathology in recent years. While primarily used for research, education, and remote consultation, following FDA approval of Philips’ IntelliSite imaging system in 2017, whole-slide images may now be used for primary clinical diagnosis. In this context, automated image analysis of scanned microscopic slides could drastically increase diagnostic efficiency and reduce inter-observer variability and errors. This would allow fewer pathologists to serve more patients while maintaining diagnostic accuracy and precision. Along with unique opportunities, histopathology presents unique challenges for deep learning models. Acquiring sizable datasets for histopathology is much more difficult than for normal computer vision tasks (cat pictures are much more popular and easier to label than cell pictures). Additionally, whole-slide images have very high resolution and are susceptible to many sources of variation. Even with the necessary data, models that are created for diagnostic tasks must also achieve greater interpretability and performance than models used in many other contexts. Mitosis under a fluorescence microscope (source) In this post, we will focus on mitosis detection, which shares many of the challenges facing other nuclear detection tasks, as a motivating example. Mitosis is the process by which non-reproductive cells divide into two genetically-identical child cells. Mitosis detection in microscopy slides is important for the analysis of many diseases — for example, pathologists will count mitosis events to determine how quickly a cancer is replicating. In particular, mitosis events are usually detected on static Hematoxylin and Eosin (H&E) stained slides (first image in the next section), making the task much more difficult than recognizing mitosis in the above GIF. In the subsequent sections, we aim to address the following questions: Why automate nuclear detection tasks? The potential impact of slide analysis automation on healthcare Why deep learning, and how is it used? Why this is a good problem for deep learning and how it has been used to achieve remarkable results Why are nuclear detection tasks hard? The various challenges and existing partial solutions to address them What’s next? Thoughts on bringing deep learning solutions into regular clinical practice The need for automated nuclear detection H&E Stained Breast Cancer Samples from ICPR Dataset (source) Imagine that you are a pathologist, and a doctor has sent you a sample of an abnormal mass observed during a routine breast cancer screening. Your job is to determine the nature of the cells in the sample. Let’s say the you determine the mass to be a malignant cancer — now you, the doctor, and the rest of the healthcare team must work together to determine treatment options. One important factor in selecting treatments is the “grade” of the cancer, which represents the cancer’s biologic aggressiveness. Applying the widely-used Nottingham Grading System, you analyze three morphologic factors: tubule formation, the degree to which the cancer cells have developed normal breast cell structure, nuclear grade, which is based on the morphology of tumor nuclei, and mitotic index, a measurement of the rate at which cells are dividing (National Cancer Institute). Let’s take a closer look at how you would determine the mitotic index. The Mitotic Activity Index is calculated as the number of mitotic events in a 2 mm² area of the tissue sample. This corresponds to the count of nuclei undergoing mitosis in 8–10 “high power fields” (areas visible under high magnification, typically 40X), depending on the microscope used. Identifying mitotic cells is challenging: you must discern mitotic events amongst similar-looking apoptotic (dying) cells and image artifacts, select appropriate areas for analysis, and account for variations in the appearance of mitotic figures (Veta et al., 2015). This difficulty can lead to variability in measurements and high inter-observer error. Take a look at the image below: can you tell the difference between the mitotic and non-mitotic examples? Answer: (a)-(c ) true mitoses, (d)-(f) confounding examples (source) Additionally, this process is highly time- and labor-intensive. A typical case will take a pathologist 5–10 minutes to analyze, with repeated examinations required in certain scenarios (Veta et al., 2015). In 2014, 236,968 women and 2,141 men were diagnosed with breast cancer in the US (CDC, 2017) — that’s a lot of medical experts spending a lot of time counting mitotic cells. Creating fast, consistent automated solutions for mitosis detection would have a significant impact on healthcare costs and quality. Of course, breast cancer treatment is not the only field that would benefit from automated nuclear detection solutions. Robust and accurate nuclear detection is essential for digital pathology in general — nuclear morphologic structure and arrangement provides important clues for many diagnostic tasks, ranging from cell counting and tracking to grading applications for a variety of diseases (Xie et al., 2015). The role of deep learning Example convolutional neural network architecture (source) Machine learning techniques, which utilize data to learn relationships between inputs and outputs, have found wide applications in nuclear detection tasks, with a subset of techniques known as deep learning becoming the leading approach for state-of-the-art solutions. While traditional machine learning often relies heavily on feature selection and engineering for good performance, deep learning models learn useful representations from raw data through the training process. Convolutional neural networks (CNNs), characterized by convolutional layers that exploit locality, have proven particularly useful in computer vision tasks including nuclei detection. For more information on CNNs, check out Stanford’s CS231N course notes. While supervised training of CNNs has been the backbone of deep learning successes in image tasks, unsupervised approaches where models learn representations from data without explicit human annotation have also proven useful in cases with limited data. For example, autoencoders (AEs) learn to encode inputs into a compressed latent representation that can be used as input for further supervised training. More on autoencoders here. Progress Made Before diving into what’s left for deep learning in nuclear and mitosis detection tasks, we will provide some examples of recent successes and the approaches used to achieve them. Recent mitosis detection competitions (ICPR 2012, AMIDA 2013, TUPAC 2016) have provided researchers with labeled datasets and pushed the state-of-the-art for automated mitosis detection. Initial deep learning success was achieved by Ciresan et al. (2013), who won the ICPR 2012 competition with multiple CNNs making pixel-wise mitosis classification predictions. The 2013 AMIDA competition was won with a similar approach. Fully Convolutional Deep Regression Network downsamples input then upsamples to output regression map of same input dimensions (Chen et al., 2016). Research using these datasets continued long after the competitions concluded. Wang et al. (2014) combined hand-crafted features and smaller CNN models to reduce the computational burden of their mitosis detector. Chen et al. (2016) achieved a new best F1 score of 0.790 on the 2012 ICPR dataset by employing a fully convolutional Deep Regression Network (DRN) that made pixel-wise proximity predictions for mitotic nuclei (diagram above). The fully convolutional architecture allows for inference with a single forward pass, regardless of input image size, improving efficiency. Beyond mitosis detection, similar approaches have achieved success in other nuclear detection tasks. Kashif et al. (2016) combined neural networks with handcrafted features to detect tumor cell nuclei. Regression approaches that exploit topological information were utilized by Xie et al (2015) and Sirinukunwattana et al. (2016) for tumor nuclei tasks. Xie et al (2015) also successfully utilized spatial information with a “deep voting” scheme. Nuclei detection using Stacked Sparse Autoencoders. The green, yellow and red dots represent the true positives, false positives, and false negatives respectively. (Source) Unsupervised learning techniques have also proven useful. Xu et al. (2015) stacked two sparse autoencoders to learn useful representations of nuclei arrangements in input patches, improving the performance of their classifier. Deep learning techniques have produced state-of-the-art results, even surpassing the performance of pathologists in some cases (Bejnordi et al., 2017). However, it’s important to note that for many tasks, the datasets used for research are often small and acquired from a single or a few prominent institutions, so variations in data may not be fully represented. As a result, the generalizability of these deep learning models to new cases is still in question. Let’s take a closer look at some of the challenges facing deep learning solutions for nuclear detection tasks, including mitosis detection, and the attempts being made to address them. Current Barriers to Progress Microscopic image analysis presents a series of unique challenges to deep learning models that are not encountered in typical vision tasks. Furthermore, existing healthcare resources need to evolve to be able to support and utilize new technologies as they are developed. Machine learning challenges This section draws heavily from this review by Xing et al., 2017. 1. Needs More (Representative, Well-Labeled) Data The performance of a deep learning model relies on access to a sufficient amount of representative, well-labeled data. However, there are significant barriers to the acquisition of good datasets for nuclear detection: lack of medical expertise: medical expertise is often required for accurate annotations of microscopy images time-intensive annotations: patch-wise annotations of extremely high-resolution images are required for detection tasks privacy concerns: biomedical data must be treated carefully to maintain privacy expensive/rare scanners: whole slide image scanners are not currently widely available imbalanced classes: for many tasks, such as mitosis detection, one label (“no mitosis”) is much more common than the other (“mitosis”) To overcome these challenges, it is important to develop techniques to extract more information from available data or standardize data aggregated from different sources. Data Augmentation One standard approach to deal with limited data is data augmentation, the practice of applying label-preserving transformations to images, such as rotation, reflection, random crops, and color alterations for microscopy images, to expand the dataset. While this technique is simple, efficient, and beneficial to performance, augmented images may be highly correlated and are often insufficient to train a generalizable model. Crowdsourcing Crowdsourcing, the practice of using a large number of individuals of varying knowledge to label data inexpensively, is also being explored as an alternative approach to building a large dataset. Because crowdsourcing yields noisy and often conflicting labels, especially in difficult tasks like nuclear detection, researchers must develop models robust to noisy annotations. Aggnet Approach Overview (source) Albarqouni et al. (2016) have made early progress on this problem, working on the challenge of crowdsourcing labels for mitosis detection in breast cancer biopsy images. They introduce an additional aggregation layer to their CNNs to process non-expert annotations and estimate the validity of these labels while simultaneously tuning their classifiers. To do so, they train multiple CNNs with expert-labeled images at various scales. These networks are then used to identify candidate mitotic events from unlabeled data, which are sent to non-experts for annotation. The sensitivity and specificity of each annotator is estimated in an unsupervised manner using the EM algorithm, and these weighted annotations are used to train the classifier (while also incorporating a trust metric from the crowdsourcing platform). The team achieved promising results, setting the stage for further exploration of crowd-sourcing techniques in automated microscopic image analysis. Transfer Learning and Unsupervised Pre-Training Transfer learning is a popular practice in computer vision to build generalizable models when there is not a lot of data available. The process involves using a model trained on a different, large dataset (such as ImageNet) for initialization or feature extraction. Additional layers may then be added and the entire network may be fine-tuned in a supervised manner. Alternatively, a scheme of unsupervised pre-training and supervised fine-tuning can be used to take advantage of unlabeled data. In this approach, an unsupervised model is used to determine a latent representation of the input, which is then extended and trained in a supervised manner. These techniques have been applied to nuclear detection tasks. Because transfer learning is more effective when the pre-training dataset is similar to dataset for the desired task, a generic model pre-trained on microscopy images from various microscopes and staining preparations could perform better than one pre-trained on standard natural images. Unfortunately, no general pre-trained microscopy model is currently available. 2. High-Dimensional Data For many nuclear detection tasks, it is important to have solutions that work with an entire whole slide image, rather than just manually cropped selections. However, these images are often enormous (1–40 GB!), with dimensions over 50,000 x 50,000 pixels. Additionally, these images may contain tens of thousands to millions of objects of interest for some nuclear detection tasks (Xing et al., 2017). With such high dimensional data, extensive computation is required to process even a single image, especially with the CNN pixel-wise classification approach. Fully convolutional networks, while being more efficient, still face memory issues with large image sizes. Because whole slide images cannot be resized significantly without losing valuable information, images are often split into patches, processed separately, then stitched together. Designing an efficient and robust algorithm for this process remains an open problem, complicated because very few patches may contain the relevant objects and image splitting may lose contextual information (Xing et al., 2017). 3. Variations in Data Preparation The image quality of a whole slide scan is highly dependent on an extensive preparation process. Tissues are sectioned onto glass slides and stained with different chemical substances before undergoing scanning and imaging. Image artifacts, such as tissue folds and blurred images, can present challenges to deep learning algorithms. As we strive to improve the diversity and quantity of our data, we will need to handle images from multiple institutions utilizing a variety of microscopes and preparation protocols. These differences lead to color and scale variations, known as batch effects, which can bias the performance of predictive models (Kothari et al. 2013). Scale variations can be particularly difficult to detect due to the wide variety of cell and nuclear morphologies in healthy and diseased tissues. Fortunately, researchers have made progress in normalizing color and scale batch effects (Bejnordi et al. 2016, Kothari et al. 2014). Deep learning methods have also been used for stain normalization without hand-crafted features (Janowczyk et al., 2017), achieving promising results. Developing techniques to effectively handle batch effects will be essential to creating widely-applicable nuclear detection models. 4. Overlapping and Cluttering of Objects A pathology image can contain tens of thousands of nuclei, often partially overlapped and clustered into clumps. The ambiguous and potentially misleading boundaries make it very difficult to detect and potentially segment individual nuclei for further analysis. Incorporating shape prior modeling into deep neural networks has the potential to improve delineation of object boundaries, though that is in still an ongoing research question. Healthcare challenges Success for automatic nuclear detection should not be measured by model performance, but by actual impacts on patients. In order for any of the proposed deep learning techniques to make a difference, we must also consider the healthcare system that we are operating within. 1. Opening the “black box” In addition to achieving good performance, it is also important for doctors to be able to understand why predictions are made. For medical applications, we must be able to verify that a model’s predictive success is derived from the proper problem representation, rather than artifacts, to confidently apply these predictions in practice (Montavon et al., 2018). Neural network interpretability is an open question and an active area of research in machine learning (for a discussion of the topic, see Lipton 2017). One solution for histopathology may be to have the neural network itself explain its predictions — for example, researchers have taken early steps towards generating descriptions to accompany predictions in the radiology domain (Shin et al. 2015). Possible joint workflows, such as a CNN that identifies candidates that are then confirmed by a pathologist, may also help address these issues. 2. High costs of digitization Today, the cost-benefit ratio of digital pathology remains too high to many pathologists. Unlike radiology, where end-to-end digital workflows already exist, digital pathology still requires the creation of a tissue block and glass slides (and all the steps involved in these processes) before images can enter the digital workflow. Once a glass slide is produced, it is currently cheaper, faster, and easier for a pathologist to view the slide under a microscope than to scan it and view the digitized slide image on a computer. Creating tissue blocks cannot be done away with — even when digital pathology is widely adopted, blocks will still need to be made and kept in storage, in case the need for further ancillary studies (such as molecular testing and special staining) arise. Additionally, some pathologists may be resistant to changes in their current workflow. Effective automated analysis will be instrumental in providing sufficient benefit to pathologists to adopt digital workflows; though, as we have seen, improved automated analysis requires more data. For this reason, it is important that partnerships between machine learning researchers and pathologists are formed to bring automated microscopic image analysis (and the benefits it brings) to fruition. Conclusion: Cats and Cells So are cats and cells really that different, from a deep learning point of view? Yes and no. As we’ve seen, the same deep learning techniques (e.g. CNNs and their variants) that have driven much of the recent success in general computer vision tasks have also produced state-of-the-art results in nuclear detection tasks. However mitosis detection and nuclear detection tasks in general face a number of challenges not usually encountered in detecting our feline friends. Unlike cats, microscopy images are not plastered across the internet, and it is much more difficult to accurately annotate a mitotic or cancerous nucleus than a tabby. Additionally, whole slide images are very large and can contain thousands of cluttered and overlapping nuclei, while the population in normal cat training inputs is much smaller. Significant variations between our limited data sources also pose a challenge to building models that can generalize well. Finally, our nuclear detection models must not only successfully detect nuclei but must also be interpretable. We believe that it is essential for pathologists and computer scientists to come together to understand and tackle the technical and healthcare challenges facing deep learning in nuclear detection tasks. Through such partnerships, we can target the most impactful problems and build the representative datasets and robust models necessary to bring the breakthroughs of deep learning to histopathology. We look forward to a future where deep learning models can effectively automate many microscopy tasks in regular clinical practice, enabling pathologists to spend less time counting cells and more time on high-level diagnostic and other analytic tasks. Acknowledgements We would like to thank Dr. Jeanne Shen, Assistant Professor of Pathology at the Stanford University School of Medicine, for her guidance and valuable feedback throughout the writing process. We would also like to thank Pranav Rajpurkar and Jeremy Irvin of the Stanford Machine Learning Group and Ruth Starkman of the Stanford University PWR department for their comments. References Gurcan, Metin N., et al. “Histopathological image analysis: A review.” IEEE reviews in biomedical engineering 2 (2009): 147–171. Taigman, Yaniv, et al. “Deepface: Closing the gap to human-level performance in face verification.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2014. Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. “Imagenet classification with deep convolutional neural networks.” Advances in neural information processing systems. 2012. Veta, Mitko, et al. “Assessment of algorithms for mitosis detection in breast cancer histopathology images.” Medical image analysis 20.1 (2015): 237–248. “Tumor Grade.” National Cancer Institute, www.cancer.gov/about-cancer/diagnosis-staging/prognosis/tumor-grade-fact-sheet. “FDA Allows Marketing of First Whole Slide Imaging System for Digital Pathology.” U.S Food and Drug Administration Home Page, Office of the Commissioner, www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm552742.htm Xie, Yuanpu, et al. “Deep voting: A robust approach toward nucleus localization in microscopy images.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2015. Cireşan, Dan C., et al. “Mitosis detection in breast cancer histology images with deep neural networks.” International Conference on Medical Image Computing and Computer-assisted Intervention. Springer, Berlin, Heidelberg, 2013. Unsupervised Feature Learning and Deep Learning Tutorial, ufldl.stanford.edu/tutorial/unsupervised/Autoencoders/. “CS231n Convolutional Neural Networks for Visual Recognition.” CS231n Convolutional Neural Networks for Visual Recognition, cs231n.github.io/. “MICCAI Grand Challenge: Tumor Proliferation Assessment Challenge (TUPAC16).” MICCAI Grand Challenge: Tumor Proliferation Assessment Challenge (TUPAC16), tupac.tue-image.nl/. “ICPR 2012 — Mitosis detection contest.” ICPR 2012 — Mitosis detection contest | Image & Pervasive Access Lab, www.ipal.cnrs.fr/event/icpr-2012. “MICCAI Grand Challenge: Assessment of Mitosis Detection Algorithms 2013 (AMIDA13).” MICCAI Grand Challenge: Assessment of Mitosis Detection Algorithms 2013 (AMIDA13), amida13.isi.uu.nl/. “Breast Cancer Statistics.” Centers for Disease Control and Prevention, Centers for Disease Control and Prevention, 7 June 2017, www.cdc.gov/cancer/breast/statistics/index.htm. Wang, Haibo, et al. “Mitosis detection in breast cancer pathology images by combining handcrafted and convolutional neural network features.” Journal of Medical Imaging 1.3 (2014): 034003. Chen, Hao, Xi Wang, and Pheng Ann Heng. “Automated mitosis detection with deep regression networks.” Biomedical Imaging (ISBI), 2016 IEEE 13th International Symposium on. IEEE, 2016. Kashif, Muhammad Nasim, et al. “Handcrafted features with convolutional neural networks for detection of tumor cells in histology images.” Biomedical Imaging (ISBI), 2016 IEEE 13th International Symposium on. IEEE, 2016. Sirinukunwattana, Korsuk, et al. “Locality sensitive deep learning for detection and classification of nuclei in routine colon cancer histology images.” IEEE transactions on medical imaging 35.5 (2016): 1196–1206. Xie, Yuanpu, et al. “Beyond classification: structured regression for robust cell detection using convolutional neural network.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2015. Xu, Jun, et al. “Stacked sparse autoencoder (SSAE) for nuclei detection on breast cancer histopathology images.” IEEE transactions on medical imaging 35.1 (2016): 119–130. Bejnordi, Babak Ehteshami, et al. “Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer.” Jama 318.22 (2017): 2199–2210. Xing, Fuyong, et al. “Deep Learning in Microscopy Image Analysis: A Survey.” IEEE Transactions on Neural Networks and Learning Systems (2017). Albarqouni, Shadi, et al. “Aggnet: deep learning from crowds for mitosis detection in breast cancer histology images.” IEEE transactions on medical imaging 35.5 (2016): 1313–1321. Bejnordi, Babak Ehteshami, et al. “Stain specific standardization of whole-slide histopathological images.” IEEE transactions on medical imaging 35.2 (2016): 404–415. Lipton, Zachary C. “The mythos of model interpretability.” arXiv preprint arXiv:1606.03490 (2016). Shin, Hoo-Chang, et al. “Interleaved text/image deep mining on a large-scale radiology database for automated image interpretation.” Journal of Machine Learning Research 17.1–31 (2016): 2. Kothari, Sonal, John H. Phan, and May D. Wang. “Scale normalization of histopathological images for batch invariant cancer diagnostic models.” Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE. IEEE, 2012. Bejnordi, Babak Ehteshami, et al. “Stain specific standardization of whole-slide histopathological images.” IEEE transactions on medical imaging 35.2 (2016): 404–415. Kothari, Sonal, et al. “Pathology imaging informatics for quantitative analysis of whole-slide images.” Journal of the American Medical Informatics Association 20.6 (2013): 1099–1108. Grégoire Montavon, et al. “Methods for interpreting and understanding deep neural networks.” Digital Signal Processing 73 (2018): 1–15

How Different Are Cats and Cells Anyway? Closing the Gap for Deep Learning in Histopathology

782

how-different-are-cats-and-cells-anyway-closing-the-gap-for-deep-learning-in-histopathology-14f92d0ddb63

2018-05-13

2018-05-13 11:46:23

https://medium.com/s/story/how-different-are-cats-and-cells-anyway-closing-the-gap-for-deep-learning-in-histopathology-14f92d0ddb63

false

3,923

Artificial Intelligence to Improve People’s Lives. Learn more at https://stanfordmlgroup.github.io/

null

Stanford AI for Healthcare

pranavsr@stanford.edu

stanford-ai-for-healthcare

MACHINE LEARNING,HEALTHCARE,STANFORD,ARTIFICIAL INTELLIGENCE,DEEP LEARNING

null

Machine Learning

machine-learning

Machine Learning

51,320

Michael Bereket

null

40c7c508d853

mdbereket

24

1

20,181,104

null

0

null

0

null

2018-04-14

2018-04-14 17:28:33

2018-04-14

2018-04-14 17:47:36

1

false

en

2018-10-22

2018-10-22 22:19:37

9

14f9ec97dee

4.913208

3

0

While all other tech sectors are driven by reducing inefficiencies and increasing productivity, cybersecurity spending is driven by the…

5

IoT connected devices provide great efficiencies, but not secured, pose enormous security threats The Internet of Things (IoT), AI and Cybersecurity : The Natural Trinity While all other tech sectors are driven by reducing inefficiencies and increasing productivity, cybersecurity spending is driven by the sharp rise in cybercrime. British insurance company Lloyds estimates that cyberattacks cost around $40 billion each year. Nowhere is that threat clearer than at the junction of the Internet of Things (IoT), AI and cybersecurity. The confluence of AI attacking the billions of vulnerable IoT devices connected to the Internet every year make the need for more cybersecurity professionals even more valuable. We can’t rely on short-staffed security in this existential fight for survival. A multitude of studies underline the cybersecurity labor shortage and illustrate the drastic need for more experts to fight cybercrime. The need for these professionals is real and students in higher education have an opportunity to be the next-generation of skilled cybersecurity experts. According to Burning Glass, “job postings for cybersecurity openings have grown three times as fast as openings for IT jobs overall and it takes companies longer to fill cybersecurity positions than other IT jobs. That’s bad for employers but good news for cybersecurity workers, who can command an average salary premium of nearly $6,500 per year, or 9% more than other IT workers. Or put another way, there were nearly 50,000 postings for workers with a CISSP certification in 2014, the primary credential in cybersecurity work. That amounts to three-quarters of all the people who hold that certification in the United States — and presumably most of them already have jobs. This is a gap that will take time to fill.” Jennifer Teeters, Director of Authorized Training at Fortinet, the $9B security networking vendor says that “while the workforce shortfall is one we cannot ignore, the question becomes how do we load the pipeline of skilled workers in the field.” She continues, “many higher education institutions are not equipped to give students the education or hands-on training required for their students to enter the corporate world prepared for these roles. Students in higher education environments need to be given the tools and curriculum to give them a leg-up in order to enter into these positions straight from college.” “Investors have also gone on buying sprees as companies have been snapped up for their cybersecurity technology or in-demand security engineers,” according to Teeters. Teeters leads the Fortinet Networking Security Academy (FNSA) program, which runs over 90 Academies in 36 countries, training a next generation of security professionals to fill the overwhelming gaps between qualified workers and the staggering number of security professional jobs that go unfilled. The FNSA program is designed to provide industry recognized Fortinet training and certification opportunities — previously restricted to the company’s customers and employees — to Academy students around the world. In only two years, Fortinet’s program has grown from inception to the point where it partners with educational organizations, nonprofits and veterans’ groups to train students who can provide great value to future cybersecurity employers. Fortinet has plans to expand the program quite significantly as students finish the program with the skills needed to help protect global organizations from cyberattacks. Teeters believes that Academy graduates will be “recognized in as an elite group who can command high salaries and play an essential role in the cyberterrorism fight.” The need for these individuals has ramped up based on the fact that IoT is so prevalent, leaving millions or billions of unsecured “things” — sensors that can collect data and communicate it over networks. The trends in security spending call for more trained professionals to combat cybersecurity. Cybersecurity Ventures estimates that the market for cyber products and services will be $1 trillion cumulatively over the next five years. Cybercrime has dwarfed security spending over the past decade. Companies and governments are finally spending massive amounts to fight this threat. The market is finally beginning to catch up out of sheer necessity. As a result, the Cybersecurity market grows 35X from $3.5B in 2004 to $120B in 2017. According to Cybersecurity Ventures, spending is predicted to exceed $1T in the next 5 years. IT analyst forecasts are unable to keep pace with the dramatic rise in cybercrime, the ransomware epidemic, the refocusing of malware from PCs and laptops to smartphones and mobile devices, the deployment of billions of under-protected Internet of Things (IoT) devices, the legions of hackers-for-hire, and the more sophisticated cyber-attacks launching at businesses, governments, educational institutions, and consumers globally. In the wake of Facebook’s hacking of 87M users, companies need to be increasingly concerned about who they connect with. As a result, people are spending to protect themselves and also counter direct attacks after they happen. And it’s no wonder that artificial intelligence is hot in cybersecurity. As the number of IoT devices is projected to reach into the tens of billions in coming years, enterprise companies will be compelled to embrace AI, machine learning and automation tools to help secure and manage their networks. “Cybersecurity is beginning to look like an endless game of chess that pits human hackers against AI-enhanced security professionals. It is already possible to automate cybersecurity responses with machine learning and AI. Hackers, however, change tactics frequently, and cybersecurity is not a finite problem that can be solved, as a recent Harvard Business Review article notes. There is no end state where cybercrime goes away or is permanently eliminated. It is, in fact, a cat and mouse game that will remain so. In any event, machine learning has already changed the rules of the IoT cybersecurity game, making it look more like a chess match that pits machine against machine. However, it will be up to AI-armed security professionals — whether they come out of the Fortinet Academies or elsewhere — to stop vulnerable, or potentially vulnerable, devices from compromising network integrity. AI can be used to take advantage of IoT devices, but it can also be countermanded to fight the criminals. To that end, we need all the trained professionals we can get. On the other hand, it is possible that powerful hacking groups — such as those supported by nation states — could be looking at AI to fuel crippling attacks against targets of their choosing. The other threat of AI, as Elon Musk and others have warned, it is the risk that humans could lose control of AI-enhanced cyber-weapons or that powerful autonomous weapons emerge that can select targets without human intervention. However, the more obvious AI threat is that human passivity allows IoT devices to be connected without sufficient security. After all, people want IoT devices connected and running their remarkable functionality. Security is an afterthought. It bears emphasis: the network is only as strong as its weakest link. If a nuclear force at sea goes black because of a cyberattack, it could be only seconds until a full-scale nuclear attack and annihilation. On a less alarmist but equally disturbing note, it was a baby monitor using a default password of “p-a-s-s-w-o-r-d” that enabled hackers to perpetrate the most heinous and costly cyberattack last year. Greater security strategy that takes into account IoT’s vulnerabilities and an army of trained professionals are needed to fight more such threats. With cyber spending growing exponentially to stay up with cybercrime, efforts to create the requisite counterattack workforce are finally being undertaken in new training programs with great promise.

The Internet of Things (IoT), AI and Cybersecurity : The Natural Trinity

3

the-internet-of-things-iot-ai-and-cybersecurity-the-natural-trinity-14f9ec97dee

2018-10-22

2018-10-22 22:19:37

https://medium.com/s/story/the-internet-of-things-iot-ai-and-cybersecurity-the-natural-trinity-14f9ec97dee

false

1,249

null

Cybersecurity

cybersecurity

Cybersecurity

24,500

Christopher Nordlinger

Ph.D. Fulbright Scholar. Storyteller. Communications Expert. Content maven. Formerly State Dept-Startups-Cisco & more.

d5ffec07f499

chrisnordlinger

928

799

20,181,104

null

0

null

0

null

2018-05-10

2018-05-10 01:41:38

2018-05-10

2018-05-10 02:31:19

5

false

en

2018-05-10

2018-05-10 02:31:19

7

14f9fa91f4c2

2.844654

2

0

Talk to your TV set, talk to ticket-selling machines and beyond

5

Speech Technology at Alibaba Talk to your TV set, talk to ticket-selling machines and beyond Alibaba has deployed AI technologies, e.g., speech technology, natural language processing (NLP), video technology, image technology and machine learning) etc, to a broad range of applications in E-commerce, Financial Services, New Manufacturing, and New Retail. In diverse scenarios related to enterprise application, Alibaba has accumulated a wealth of knowledge derived from Internet-based big data, making itself a world leader in AI application. In Speech Technology Talk to your TV set: Talk to ticket-selling machines: Job opportunities at the Alibaba Speech Technology Team: https://speech.alibaba-inc.com/icassp2018/jobs Publication: Speaking Up: Optimizing Large Vocabulary Speech Recognition Systems Deep neural networks have become the dominant acoustic model used in Large Vocabulary Continuous Speech Recognition (LVCSR) systems. Neural networks include both Feed-forward Neural Networks (FNN) and Recurrent Neural Networks (RNN). Although RNNs have been shown to significantly outperform FNNs, the learning capabilities of RNNs usually rely on Back Propagation Through Time (BPTT) due to internal recurrent cycles. This significantly increases the computational complexity of learning and also may cause such problems as gradient vanishing and exploding. Improved Time Dependency Modeling in Emotion Recognition: Advanced LSTM Long short-term memory (LSTM) layers are the building blocks of recurrent neural networks (RNN) and are used to facilitate the application of RNNs in sequential modeling tasks, such as machine translation. Due to layer inputs, the LSTM layer assumes that the state of its current layer (as stored in the memory cell) is dependent on the state of the same layer at the previous time point. This one-step time dependency restricts the modeling capability of temporal information and represents a major constraint of LSTM layers in RNNs. Lost for Words: Speech Synthesis with Limited Data Using Linear Networks Speaker-dependent acoustic models ensure that speech synthesis systems give accurate results. Given an adequate amount of training data from target speakers, speech synthesis systems are able to generate results similar to the target speaker. However, gaining enough data from target speakers is always a constraint. An Ensemble Framework of Voice-based Emotion Recognition The importance of emotion recognition is gaining more and more traction with improving user experience and the engagement of human-computer interfaces (HCI). Developing emotion recognition systems that are based on speech, as opposed to facial expressions, has practical application benefits due to low hardware requirements. However, these benefits are somewhat negated by real-world background noise impairing speech-based emotion recognition performance when the system is employed in practical applications. Reading Aloud: Sequential Memory for Speech Synthesis Text-to-Speech (TTS) systems are an essential part of human-computer interactions. For current Internet-of-Things (IoT) devices, such as smart speakers and smart TVs, speech is the most efficient and accessible approach for both the user and device to understand each other through instructions and feedback. However, one issue that commonly hampers user experience is machine generated speech being perceived as unnatural or non-human-like by users. Overcoming this obstacle has been a major challenge for TTS systems to date. Alibaba Tech First-hand and in-depth information about Alibaba’s latest technology → Search “Alibaba Tech” on Facebook

Speech Technology at Alibaba

2

speech-technology-at-alibaba-14f9fa91f4c2

2018-05-30

2018-05-30 12:23:57

https://medium.com/s/story/speech-technology-at-alibaba-14f9fa91f4c2

false

533

null

Machine Learning

machine-learning

Machine Learning

51,320

Alibaba Tech

First-hand & in-depth information about Alibaba's tech innovation in Artificial Intelligence, Big Data & Computer Engineering. Follow us on Facebook!

69f6dde768a1

alitech_2017

1,157

14

20,181,104

null

0

null

0

634d4b270054

2018-04-02

2018-04-02 06:40:21

2018-04-02

2018-04-02 06:43:16

1

false

en

2018-06-05

2018-06-05 09:22:31

3

15006ab0c506

1.007547

1

0

Aerones, a Latvian Company is developing a heavy-duty drone that can perform the number of functions from physically rescuing peoples and…

5

Meet This Aerones Drone Used For De-Icing And Rescuing Peoples Aerones, a Latvian Company is developing a heavy-duty drone that can perform the number of functions from physically rescuing peoples and de-icing the wind turbine blades. This powerful craft can lift about 100kg and the speciality of this drone is that it can be completely mobile. This drone can also fight fires without placing human near dangers. The tether provides the drone with power, solving the constant battery limitations that even the most advanced drones still suffer from. Moreover, this drone can stay airborne indefinitely. Have a look at the wonderful videos how this drone technology is helping rescuing peoples, fight the fire and de-icing the blades. Source: https://bit.ly/2EcKOb0 About DEEPAERO DEEP AERO is a global leader in drone technology innovation. At DEEP AERO, we are building an autonomous drone economy powered by AI & Blockchain. DEEP AERO’s DRONE-UTM is an AI-driven, autonomous, self-governing, intelligent drone/unmanned aircraft system (UAS) traffic management (UTM) platform on the Blockchain. DEEP AERO’s DRONE-MP is a decentralized marketplace. It will be one stop shop for all products and services for drones. These platforms will be the foundation of the drone economy and will be powered by the DEEP AERO (DRONE) token.

Meet This Aerones Drone Used For De-Icing And Rescuing Peoples

1

meet-this-aerones-drone-used-for-de-icing-and-rescuing-peoples-15006ab0c506

2018-06-05

2018-06-05 09:22:32

https://medium.com/s/story/meet-this-aerones-drone-used-for-de-icing-and-rescuing-peoples-15006ab0c506

false

214

AI Driven Drone Economy on the Blockchain

null

DeepAeroDrones

null

DEEPAERODRONES

null

deepaerodrones

DEEPAERO,AI,BLOCKCHAIN,DRONE,ICO

DeepAeroDrones

Deepaero

deepaeros

Deepaero

0

DEEP AERO DRONES

null

dcef5da6c7fa

deepaerodrones

277

0

20,181,104

null

0

null

0

null

2018-03-12

2018-03-12 23:33:25

2018-03-13

2018-03-13 12:59:17

10

false

en

2018-03-13

2018-03-13 13:52:14

4

150152b85a79

3.295283

12

0

This story is a continuation of our last weeks model fixing racism. Race is much more complicated than just calling someone Black, White…

5

Country Of Origin Using Deep Learning With Excel This story is a continuation of our last weeks model fixing racism. Race is much more complicated than just calling someone Black, White, Latin, Asian, Indian, etc… we have so much diversity even within racial groups. To demonstrate that we will build a deep net with (no programming, all setup is done in Excel) to predict country of origin. For this example we have 144,705 faces tied to a country and 219 unique classes of countries. Here is a preview of that dataset in an Excel file where the image path is first, followed by the country name: With Excel for setup we can organize our images and labels. This CSV file can then be used for deep learning There are no decisions to be made on GPU driver versions, CuDNN, Tensorflow, Caffe, Torch, etc… or network architecture (i.e. NASnet, InceptionV4, ResNet-2024, etc…) or tricks for lift if faces are present. Pushing this through our turnkey deep learning platform I achieve 23.4% accuracy on our validation set “automagically”. For 219 countries that is actually really good, better than what most deep learning data scientists could pull off, especially in a single morning. Once a deep network has been trained we can actually visualize how the labels relate to each other. This is a cluster plot of the median embedded layer for each label. An important point to make here is the median will reduce minority influence, so don’t be alarmed if majority white/black/asian countries are next to majority white/black/asian countries. I hope there are a few woah reactions to this plot. This is a gnarly set to visualize so we will add some region color groupings. A few notable ah-ha things to call out are the continents groupings. Adding the groupings: Africa, America, Asia, Australia, Balkans, Caribbean, Europe, Middle East, North Africa, North America, Oceania we get: Country organization using deep learning shows regional similarity. A HD pdf is available here: http://bit.ly/coi_plot A few quick wins to call out: Mexico & Puerto Rico are included in the North American region, but grouped with South & Central America See how Mexico and Puerto Rico (north American) are being grouped with Central/South countries? It was also encouraging to see Japan, China, South Korea, Taiwan, Hong Kong, and Vietnam so close together: See how the Asian countries are grouped (i.e. Japan, China, South Korea, Taiwan, etc..) It is also encouraging to see all of the Africa countries grouped together, as well as the Caribbean countries being more similar to Africa than the other regions. This is true due to the heavy slave trade influence seen with many of the Caribbean countries. See how African countries are grouped, also how the Caribbean countries are similar? Humans Try And Bucket: Humans try and bucket everything, personality, topics, emotion, and race. Our minds are simple. If we try bucketing these countries we can get a few different outcomes depending on how many buckets we think make sense. If we only assume there are 3 races then latin & asian countries are grouped. If we say there are 8 races we get something like this If we say there are 20 races we get something like this. Conclusion: Using ONLY the faces from countries it is interesting to see how different countries relate to each other. Our current understanding of race can be seen in some of these groupings (i.e. Central Africa = Black, East Asia = Asian, etc..). It is encouraging to see neighboring countries close to each other with this facial analysis. Also, getting back to what makes up race we can see it is much more complicated. For example, looking at Greenland they are very distinct from different countries, so you could argue that Greenland should be its own race. What do you think? Greenland is an outlier, do you think it should be its own race? Our Hardware: We are running on analysis on a PureStorage FlashBlade and a Dell C4140 with Nvidia V100s.

Country Of Origin Using Deep Learning With Excel

94

country-of-origin-using-deep-learning-with-excel-150152b85a79

2018-06-21

2018-06-21 03:40:22

https://medium.com/s/story/country-of-origin-using-deep-learning-with-excel-150152b85a79

false

542

null

Deep Learning

deep-learning

Deep Learning

12,189

Ben Taylor

I am obsessed with deep learning and general AI. I spend almost every waking hour working on making the most complex technology approachable

f9f9ec431fa9

bentaylorche

135

20

20,181,104

null

0

null

0

ca905b08d8f

2017-09-22

2017-09-22 14:06:30

2017-09-23

2017-09-23 14:16:12

1

false

en

2017-09-23

2017-09-23 14:16:12

15

1502cc89fd65

7.271698

2

1

0

This week there are three sets of articles and a video.

5

News — At The Edge — 9/23 This week there are three sets of articles and a video. From the past — the death of a Russian soldier who prevented nuclear war by ignoring a computer telling him they were under attack. An extraordinary feat — ignoring the machine, military orders and training, and risk of being “dead” wrong — unlikely to occur again; or if artificial intelligence was in control. From the future — Baby X simulation of artificial life and coming robot arms-race — we see how technology is advancing fast, wild and increasingly risky. For the present — sensor data transmission, why we cannot control AI, AI economic dangers and social media problem in selling ads to racists — it is important to stop being reactive when it comes to our digital future. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Issue from the Past – Stanislav Petrov: The man who may have saved the world - “Thirty years ago, on 26 September 1983…[in] early hours of the morning, the Soviet Union’s early-warning systems detected an incoming missile strike from the United States…[and] protocol for the Soviet military would have been to retaliate with a nuclear attack of its own…[but Petrov] decided not to report them to his superiors, and instead dismissed them as a false alarm…[and] a dereliction of duty…. In the political climate of 1983, a retaliatory strike would have been almost certain….The system was telling him that the…reliability of that alert was ‘highest’…[with] no doubt. America had launched a missile….[Instead] Petrov called the…army’s headquarters and reported a system malfunction. If he was wrong, the first nuclear explosions would have happened minutes later…[and] admits he was never absolutely sure that the alert was a false one…..[He] was the only officer in his team who had received a civilian education…[so] if somebody else had been on shift, the alarm would have been raised.” http://www.bbc.com/news/world-europe-24280831 Issues from the Future – BabyX v3.0 Interactive Simulation (1.75 min. video) Is BabyX the Future of Silicon-Based Life Forms? - “[In] simulated artificial life form, or animat…BabyX is [project]…simulating the neural machinery of an infant human…brings a host of moral and philosophical questions regarding artificial life: What are our duties and obligations…[and] what if any legal status will they possess?…. [BabyX] is a computer generated psychobiological simulation…[with] some form of reinforcement learning being used for acquiring skills like playing the piano….[T]he graphics used for modeling…BabyX can be so spellbinding, that the difficult mathematics behind their brain circuitry gets brushed aside…. [From] video on BabyX, it appears the animat possesses many of the neural correlates of a human, including an artificial dopamine system and other pleasure-releasing brain structures.” https://medium.com/extremetech-access/is-babyx-the-future-of-silicon-based-life-forms-d6e077d7fb0c Commentary: The coming robot arms race - “[Russia’s] military exercise is underway…[but] in 2021, those troops might be sharing their battle space with…self-driving drones, tanks, ships and submersibles…without a guiding human hand…a truly revolutionary shift…[every] nation wants…. Critics have long feared countries might be more willing to go to war with unmanned systems…control might pass beyond human beings altogether….Musk has long warned…of some cataclysmic errors when it comes to artificial intelligence…[and] the development of autonomous weapons…[creates] devastating arms race…. ’[The] leader in this will become ruler of the world,’ Putin [said]….[China] believed by some [to]…be the global leader [in]…developing autonomous swarms of drones…to fly themselves independently…[and] may be able to make their own tactical decisions…fight their own aerial dogfights….[without] direct supervision at all…. ’Radical technological change begets radical government policy ideas’…[and] AI arms race could prove as revolutionary as the invention of nuclear weapons….[AI] could dramatically increase the efficiency of surveillance technology…[that’s] terrifying, particularly in the hands of a state with little…democratic oversight…. [B]y 2030, technological breakthroughs — not just AI, but quantum computing and beyond — would produce entirely unpredictable changes. Special force teams…[might] have a robotic and artificial intelligence component deployed alongside them…. Most countries deliberately keep their defense AI secret, ultimately fueling the arms race….Russia has long [trusted]…machines more than people…[but] Facebook shut down an AI experiment after programs involved began communicating with each other in a language the humans monitoring them could not understand…. Such technology is coming…[and] even relatively old military equipment increasingly can be retrofitted….Even if mankind can avoid a nuclear apocalypse…coming AI and robotic revolution may prove an equal existential challenge.” https://uk.reuters.com/article/us-apps-robots-commentary/commentary-the-coming-robot-arms-race-idUKKCN1BT1XN Issues for the Present – A clever way to transmit data on the cheap - “[A] technology called ‘LoRa’ (from ‘long range’)…allows computers to talk to each other with radio waves…[but] not easily blocked by walls, furniture and other obstacles…because LoRa uses lower-frequency radio waves than Wi-Fi…[and] make use of a technique called ‘chirp spread modulation’…[making] even faint LoRa signals easy to distinguish from background noise…[by] modulating it…. [Chips] consume almost no power at all…by choosing to earth its tiny aerial, or not, millions of times every second. When the aerial is earthed, part of the carrier wave will be absorbed. When it is not, it will be reflected…with the whole process controlled by three tiny, and thus very frugal, electronic switches…made for less than 20 cents apiece. The signals they generate can be detected at ranges of hundreds of metres. Yet with a power consumption of just 20 millionths of a watt, a standard watch battery should keep them going a decade or more. In fact, it might be possible to power them from ambient energy….[T]he chips are slow, transmitting data…. [Already] incorporated the chips into contact lenses and a skin patch. In hospitals, the chips could help track everything…[and] making their way into disposable drug-delivery devices that notify patients via their phones when their medication is running low.” https://www.economist.com/news/science-and-technology/21728866-long-range-frugal-new-chip-could-be-just-what-smart-city-needs-clever-way Who is to blame for algorithmic outrage? “[T]he ability to target advertising to unsavory groups generated or suggested by major internet companies. ‘Jew haters’? There aren’t many…Facebook’s ad backend said. Try adding ‘jews in an oven’ to broaden your reach, suggested Google. ‘Nazi’ could engage 18.6 million users, says Twitter…. [Alerted] the companies’ responses all struck the same notes: ‘this is against our rules, we have no idea how it happened’…[yet] perfectly happy to make money from advertising targeted at groups like ‘Hitler did nothing wrong’….How was it none of them, with their thousands of employees…saw this coming?… [Seems] these companies were unwilling to institute restrictions on the parts that make them money….And let’s not pretend this is the only such abuse…where they stand to gain — Facebook sold $100K (or 5 million rubles) worth of political ads to a Russian bot net….If they’re going to trumpet their leadership in and dedication to principles of openness and inclusivity, it is incumbent on them to carry it out with maximum transparency.” https://techcrunch.com/2017/09/17/who-is-to-blame-for-algorithmic-outrage/ AI: Scary for the Right Reasons - “[Any] technology can be used for good and bad…[and] much of the public discourse [reflect]…AI’s gone wrong (a scenario certainly worth discussing)…[but] before AI goes uncontrollable or takes over jobs, there [is]…larger danger: AI in the hands of governments and/or bad actors used to push self-interested agendas against the greater good…. With AI, the vast majority of current jobs may be dislocated regardless of skill or education level…[so] it is possible that emotional labor will remain the last bastion of skills that machines cannot replicate…. [There’s] an economic war going on between nations that is more threatening…[and] likely to get exponentially worse when AI is a factor in….[that] will further concentrate global wealth…and ‘cause’ the need for very different international approaches to development, wealth and disparity…. Capitalism is by permission of democracy, and democracy should have the tools to correct for disparity. Watch out Tea Party, you haven’t seen the developing hurricane heading your way…. [Also, we’ve] seen the integrity of our political system threatened by Russian interference and our global financial system threatened…[and] AI will dramatically escalate these incidents…as rogue nations and criminal organizations…press their agendas…. Imagine an AI agent…could unleash a locust of intelligent bot trolls [that]…destroy the very notion of public opinion….[This] has a strong chance of becoming a reality in the next decade…. [AI] is already on the radar of the authoritarian countries….Putin has talked about how AI leaders will rule the world….[China’s] accumulation of expertise [and]…very large funding…could create large power inequality….This disregard for data privacy and one way transfer of technology will lend…countries like China and Russia a huge advantage…. [We] need to rethink capitalism…because efficiency will matter less, or…disparity will matter more….The biggest concern [is]…AI will dramatically worsen today’s cyber security issues and be less verifiable than nuclear technology…[so] dictators like Putin will have massively amplified clandestine power…. Not taking risks here might inadvertently be the largest risk we take.” https://hackernoon.com/ai-scary-for-the-right-reasons-185bee8c6daa To control AI, we need to understand more about humans - “[O]n the cusp of…ever-more-powerful artificial intelligence, the urgency of developing ways to ensure our creations always do what we want them to do is growing…not just how artificial intelligence works, but how humans work. Humans are the most elaborately cooperative species…[and] outflank every other animal in cognition and communication — tools that have enabled a division of labor and shared living…[with] our market economies and systems of government….Humans are also the only species to have developed ‘group normativity’…[as] system of rules and norms [for]…what is collectively…not acceptable…[punished] with prisons and courts…[or] criticism and exclusion…. [With] AIs exercising free will…[we’re] worried about is whether or not they will continue to play by and help enforce our rules…. But our complex normative social orders are less about ethical choices than they are about the coordination of billions of people making millions of choices on a daily basis….How that coordination is accomplished is something we don’t really understand. Culture is a set of rules, but what makes it change…is something we have yet to fully understand. Law is another set of rules that we can change simply in theory but less so in reality…. AIs will only be able to integrate into our elaborate normative systems if they are built to read, and participate in, that system…[focusing] on the features of a system…[as] normative systems. Are we prepared for AIs that start building their own normative systems — their own rules about what is acceptable and unacceptable for a machine to do…[as] basis for predicting what other machines will do. We have already seen AIs that surprise their developers by creating their own language to improve their performance on cooperative tasks…. [For] machines that follow the rules that multiple, conflicting, and sometimes inchoate human groups help to shape, we will need to understand a lot more about what makes each of us willing to do that, every day.” https://techcrunch.com/2017/09/13/to-control-ai-we-need-to-understand-more-about-humans/ Find more of my ideas on Medium at, A Passion to Evolve. Or click the Follow button below to add me to your feed. Prefer a weekly email newsletter — no ads, no spam, & never sell the list — email me dochuston1@gmail.com with “add me” in the subject line. May you live long and prosper! Doc Huston

News — At The Edge — 9/23

11

news-at-the-edge-9-23-1502cc89fd65

2018-01-22

2018-01-22 23:52:38

https://medium.com/s/story/news-at-the-edge-9-23-1502cc89fd65

false

1,874

Time has come to rethink everything

null

dochuston

null

A Passion to Evolve

dochuston1@gmail.com

a-passion-to-evolve

EVOLUTION,KNOWLEDGE,FUTURE,SOCIAL CHANGE,INNOVATION

doch_one

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Doc Huston

Consultant & Speaker on future nexus of technology-economics-politics, PhD Nested System Evolution, MA Alternative Futures, Patent Holder — dochuston1@gmail.com

4561ce40d387

doch_one

4,656

106

20,181,104

null

0

null

0

null

2017-10-19

2017-10-19 02:18:36

2017-10-19

2017-10-19 02:37:27

2

false

en

2017-10-19

2017-10-19 02:37:52

0

15038fb858b0

0.987107

0

9

0

Most people believe driverless cars are too futuristic to be true or that they are only being made by specialist companies. Neither of…

3

Toyota AI-Powered Driverless Car Most people believe driverless cars are too futuristic to be true or that they are only being made by specialist companies. Neither of these things are true anymore. Toyota announced they will test their driverless AI-Powered cars in 2020. Toyota Concept-i Toyota plans on combining their Conecept-i car with Yui, an AI. The Japanese carmaker is spending billions on venture capital meant for AI development. This money is going into Yui which is supposed to be the seperating factor between Concept-i and other autonomous cars. Toyota plans on having Yui be able to converse with drivers and get to know them and learn their preferences, emotions, and habits. Toyota has an ambitious plan for the future even outside Concept-i. Later this month they plan on testing their hydrogen powered trucks. The manufacturer’s bigger plan, though, is bringing a flying car to the 2020 Olympics in Tokyo with the help of Cartivator Resource Management.

Toyota AI-Powered Driverless Car

0

toyota-ai-powered-driverless-car-15038fb858b0

2018-04-14

2018-04-14 10:00:59

https://medium.com/s/story/toyota-ai-powered-driverless-car-15038fb858b0

false

160

null

Tech

tech

Tech

142,368

Sean McCormick

Junior Mechanical Engineering Student @ Temple University

a32d3a632cd0

sean.mccormick.

20

9

20,181,104

null

0

null

0

null

2018-04-17

2018-04-17 18:18:58

2018-04-17

2018-04-17 20:19:54

7

false

en

2018-04-17

2018-04-17 20:19:54

2

1503b41f2ea1

3.70283

0

Note: The takeaway from this article is that you could probably install DAR.WIN on 5 Shopify stores in the time it takes to read this…

4

DAR.WIN DAY 1: Getting started with DAR.WIN in your Shopify store, Pt. 1 This could be you! THIS COULD BE YOU! Note: The takeaway from this article is that you could probably install DAR.WIN on 5 Shopify stores in the time it takes to read this writeup, but we’re still happy you’re along for the ride with us Getting started with DAR.WIN’s Machine Learning for Ecommerce platform is: from an expertise standpoint, probably the easiest thing you could do with a spare two minutes on your webshop from an ROI standpoint, probably the most valuable two minutes you’ll ever spend working on your webshop In fact, here’s a link if you want to jump right in: https://apps.shopify.com/dar-win So, we should probably elaborate on that rather brazen intro. As you might know, DAR.WIN is an ecommerce solution currently available in Shopify that uses machine learning to provide shoppers with a personalized experience in your online store. There are a number of tools within DAR.WIN, including a recommendation engine, an in-depth cart playback, a widget styler and an email product embed snippet generator to name a few. The biggest value our users get from DAR.WIN, though, is insights into the people who spend money in their stores — and those insights start generating from the second DAR.WIN is installed on your store. In other words, the sooner DAR.WIN is installed on your site, the earlier your data collection can begin, making your shopper recommendations more compelling and consumer insights more incisive. That’s why DAR.WIN has a free tier and takes two minutes to install and activate — it’s just good for business, and for the life of us, we can’t come up with a reason not to.* *Really. The installation step doesn’t change your site at all (you have to toggle the recommendation widget on intentionally) and the app is written in clean, modern code that doesn’t collide with other popular apps you might already be running. So what can you expect from Day 1 with DAR.WIN? We’re glad you asked. Step 1: Get the App Visit the Shopify App Store — https://apps.shopify.com/dar-win. If you’re already logged into your Shopify site, all you need to do is press ‘Get’ and the rest takes care of itself: I counted 3 clicks. That’s a three-click install. Maybe the 5 site installation disclaimer above was too conservative. At this point, DAR.WIN is installed on your site and will begin to crawl your product data to get an initial understanding of your store’s inventory and how products relate to each other — no other work is needed on your part to kick off that process. DAR.WIN will take a bit of time to complete this task, based on the size of your store. In the mean time… Step 2: Check your email We sent you an email! Welcome aboard! If you need to wrap up and head home for the day, DAR.WIN will keep chugging along in the background. If you’d like to make a bit more headway, follow along with the suggested next steps: Customize and enable product recommendations We’ll cover this topic more in depth in an upcoming article, but this area in your app where you can apply custom styles to the DAR.WIN recommendation widget. Define your Visitor Profiles The fun part! Take a few minutes to help DAR.WIN’s initial understanding of your site by associating terms with Visitor Profiles (for example, Athlete might get associated with terms like cleats, soccer ball, and free weights, while Chef could use cutting board, charcuterie or olive oil). This is another topic to explore in an upcoming article. Start Monitoring your customers’ shopping carts Take a peek at historic cart activity and make some assessments about products that are doing a killer job of converting (and products that are flattening out your sales). Not to get repetitive, but upcoming article! Step 3: Finish account creation and select a plan Now’s a good time to set up your account password and select a plan. By default, you’re on the free plan (hurray!). Upgrading when your ready has benefits including: White-Labeled Recommendation Widget Increased Recommendation Cap per month Increased Cart Playback history DAR.WIN specialist services That’s it for Part 1! In the next installment, we’ll cover setting up Visitor Profiles in depth and how you can start showing recommendations using DAR.WIN’s awesome algorithm.

DAR.WIN DAY 1: Getting started with DAR.WIN in your Shopify store, Pt. 1

0

dar-win-day-1-getting-started-with-dar-win-in-your-shopify-store-pt-1-1503b41f2ea1

2018-04-17

2018-04-17 20:19:54

https://medium.com/s/story/dar-win-day-1-getting-started-with-dar-win-in-your-shopify-store-pt-1-1503b41f2ea1

false

703

null

Ecommerce

ecommerce

Ecommerce

46,740

DAR.WIN

We're DAR.WIN! We use help online store owners of all skill levels find optimizations to sell more effectively using Machine Learning and AI. Let's connect!

816ad5d586b9

dar.win

4

6

20,181,104

null

0

null

0

a02e59d8475a

2018-04-30

2018-04-30 06:41:17

2018-04-30

2018-04-30 07:02:02

1

false

en

2018-06-27

2018-06-27 04:01:55

12

15041cacdbae

1.449057

4

1

0

Smart Micro-Insurance for the Unbanked and Uninsured on the Blockchain.

5

SURETY.AI Project and Official Channels Smart Micro-Insurance for the Unbanked and Uninsured on the Blockchain. SURETY.AI: Your Next Generation Insurance Partner SURETY.AI is a blockchain based artificial intelligence (A.I) platform for insurance companies developed by Hearti Lab Pte Ltd (Hearti). SURETY.AI allows insurance companies to connect effectively with their customers by offering micro-insurances, on-demand and at affordable prices, to Asia’s fast-growing economies through a Decentralised Enterprise Insurance Network. We use A.I. to provide a seamless distribution channel, responsive customer service, frictionless claim processing, and advanced fraud detection. SURETY.AI allows the unbanked to exchange cash for tokens for insurance coverage. SURE token holds the smart contract of the transactions and activities between the Insurers and the Insurees. We envision that SURETY.AI will be widely adopted by insurance companies in Asia to serve millions of unbanked and uninsured consumers. As insurances are closely tied to healthcare, where healthcare services such as medical check records are often used for underwriting and preventive measures, Hearti is also working with Healthcare partners to integrate their services and data into SURETY.AI. SURETY.AI TELEGRAM GROUP SURETY.AI Your Next Generation Insurance Partner SURETY.AI is a platform that allows the unbanked to exchange cash for tokens for…t.me Photo by Christian Wiediger on Unsplash 🖥 OFFICIAL CHANNELS: Telegram Group: https://t.me/suretyai ANN: https://t.me/suretyaiinfo Website: https://surety.ai Whitepaper: https://surety.ai/whitepaper.pdf Explainer Video: https://youtu.be/rlc8zqACbcM Medium: https://medium.com/theheartilab Support Email: support@hearti.io 📱SOCIAL: Facebook: https://www.facebook.com/theheartilab/ LinkedIn: https://www.linkedin.com/company/theheartilab/ Twitter: https://twitter.com/theheartilab 🙋🏻‍♂️ HEARTI TEAM: @iamkeithlim @mrkennethtan @BF_Dylan @BF_Arabella @BF_Hanna @BF_Donna @BF_Linear @BF_David 🚨 NOTE: *All announcements will be made by admins via the our ANN channel itself. You should never expect to see a PM from an admin unless it is announced first in the channel. *Our admins will never ask for you to send funds to our wallet address via PM. To verify admins, look for the ‘admin’ next to the name, or a star next to the user in the members list. Join our Telegram Group Now! SURETY.AI Your Next Generation Insurance Partner SURETY.AI is a platform that allows the unbanked to exchange cash for tokens for…t.me

SURETY.AI Project and Official Channels

11

surety-ai-project-and-official-channels-15041cacdbae

2018-06-27

2018-06-27 04:01:55

https://medium.com/s/story/surety-ai-project-and-official-channels-15041cacdbae

false

331

Enterprise Platform Using A.I & Blockchain to Revolutionize Insurance.

null

theheartilab

null

Hearti

hello@hearti.io

theheartilab

INSURTECH,BLOCKCHAIN,ARTIFICIAL INTELLIGENCE,FINANCIAL INCLUSION,CHATBOTS

theheartilab

Blockchain

blockchain

Blockchain

265,164

Kenneth Tan

b7e0a0f91935

kennethshares

25

44

20,181,104

null

0

null

0

null

2018-03-27

2018-03-27 13:47:31

2018-03-27

2018-03-27 13:51:12

1

false

sv

2018-03-29

2018-03-29 15:20:16

4

1506609ac8c1

2.003774

1

0

Detta är en sponsrad artikel i samarbete med WorkFusion.

3

Så kan intelligent automation få din verksamhet att växa Detta är en sponsrad artikel i samarbete med WorkFusion. Framtidens AI är redan här! Tillväxten inom AI är enorm inom näringslivet, menar Suhaib Mohammed på Cademica. Ca 30 % av de tillfrågade företagen använder redan intelligent automation enligt den undersökning som Monica Almgren refererar till. Siffran förväntas vara dubblerad år 2020. Vad är då intelligent automation? Egentligen är det inget annat än en specialdesignad programvara som till exempel kan identifiera produkter eller objekt i bilder, extrahera data ur dokument eller förändra information. Dessa intelligenta programvaror har tränats för att slutföra uppgifter med mänsklig kvalitet. Programmen anpassar sig och lär sig medan de arbetar. Intelligent automation kan få din verksamhet att växa? Vill du veta hur? Mohammed tar upp tre punkter: 1. Öka produktiviteten Mohammed berättar om hur en restaurang i Singapore ökade personalens produktivitet med 25 % efter att ha infört sina “flygande servitörer” — drönare som levererar tallrikarna till gästerna vid borden! De anställda kan ägna sig åt annat! Detta är bara ett av många exempel på hur intelligent automation kan effektivisera din verksamhet. Ett annat exempel är automatisering av grundläggande ekonomisk administration. Det kan frigöra dina anställda för mer icke-repetitiva och kreativa uppgifter. 2. Reducera mängden mänskliga fel Förutom att programvaran arbetar snabbare, så slipper du också den mänskliga faktorn och de fel som människor lätt gör. Det skapar också effektivitet! Den intelligenta programvaran kan lära sig och anpassa sig mycket snabbare än de flesta människor kan. 3. Förbättra din avkastning Tack vare intelligent automation kan du anställa färre och därmed minska kostnaderna för löner och förmåner. Istället investerar du i den automatiserade mjukvaran. Med ökad produktivitet och färre mänskliga fel så kommer dina genomsnittliga kostnader för samma arbete att vara mycket mindre, vilket kan öka din avkastning avsevärt, menar Mohammed. Intelligent automation i Sverige Faktum är att just svenska företag är ovanligt benägna att investera i intelligent automation jämfört med företag i andra länder. Almgren skriver om en intressant undersökning i Verkstäderna. Mer än hälften av de tillfrågade tror att människa och maskin kommer arbeta sida vid sida snarare än att maskiner kommer att ersätta människor. Det finns dock organisatoriska utmaningar med internt motstånd som hindrar en snabb implementering av intelligent automation på företagen. Att klara denna utmaning är en framgångsfaktor för framtidens ledande företag. Så mycket som 86 % av de tillfrågade företagen anser att deras företag måste använda intelligent automation de närmaste fem åren för att vara ledande inom sin bransch. Ökad produktivitet är den största affärsfördelen enligt de tillfrågade. Det är inte enbart inom näringslivet som intresset för intelligent automation är stort. Även inom offentlig verksamhet undersöks möjligheterna att öka nyttan för invånarna med hjälp av intelligent automation. Sammanfattning Vill du veta hur du kan få din verksamhet att växa med hjälp av intelligent automation? Kontakta WorkFusion så kan de berätta mer. Intelligent automation kan effektivisera din verksamhet!

Så kan intelligent automation få din verksamhet att växa

1

så-kan-intelligent-automation-få-din-verksamhet-att-växa-1506609ac8c1

2018-03-29

2018-03-29 15:21:13

https://medium.com/s/story/så-kan-intelligent-automation-få-din-verksamhet-att-växa-1506609ac8c1

false

478

null

Intelligent Machines

intelligent-machines

Intelligent Machines

55

Magnus Norberg

null

49faadd44fd5

c.magnus.norberg

0

2

20,181,104

null

0

null

0

87582c3479c6

2018-04-04

2018-04-04 17:51:27

2018-04-05

2018-04-05 16:47:16

1

false

en

2018-10-02

2018-10-02 18:48:16

3

1506ed3c4cee

2.490566

10

1

0

The MIT-born startup raised $2.6M from Eniac Ventures to replace lawyers with AI.

5

It is time to reinvent the U.S. legal system The MIT-born startup raised $2.6M from Eniac Ventures to replace lawyers with AI. By Tim Young, Founding General Partner, Eniac Ventures According to the Thomson Reuters Legal Executive Institute, the U.S. Legal Services reached upwards of $437B in 2017, and with law firms’ business models being predicated on inefficiencies, Eniac has long been bullish on a company that can capture the spend and make the provision of legal services far more effective. Enter Legit — the company that we believe is going to reinvent the U.S. legal system one vertical at a time, from IP to tax, M&A and litigation. They are well on their way because today, we are excited to announce that Eniac has led their $2.6M seed round. Born from MIT’s Computer Science and Artificial Intelligence Lab, Legit has started by tackling the burdensome process for creators of Intellectual Property. On average, up to 80% of a company’s value is in non-tangible assets like IP. The current inefficient processes to identify and create new intellectual property may be one of the most overlooked pain points enterprises face, especially now, when protecting IP is absolutely mission-critical. Legit has built the first software for designers, researchers, scientists, engineers, scalable to large R&D departments. Their proprietary natural language processing (NLP) software provides an easy-to-use interface to determine the patentable aspects of what they are working on, give real-time feedback against patent data and guide towards what is truly unique. Once complete, the output is ready to file. Legit is saving companies massive amounts of time and hundreds of thousands of dollars in legal fees in what would otherwise be spent on input, research, correcting errors, etc. Already, they have seen impressive momentum with Fortune 500s, including the largest manufacturer of industrial tools and household hardware. Eniac always looks for teams with deep technical experts and the Legit Team is no exception. Co-founder and CEO, Matt Osman was the youngest VP at a $1 billion structured credit hedge fund in London where he specialized in credit strategies and tax structuring. He graduated from Oxford with a degree in Philosophy, Politics, and Economics. Jacob Rosen, Co-founder and CTO, is an internationally recognized for his work “Tax Non-Compliance Detection Using Co-Evolution of Tax Evasion Risk and Audit Likelihood” which won the Best Innovative Application Paper Award at ICAIL 2015. He holds an M.S. from MIT and a B.S. in mathematics from the University of Michigan. We are excited to continue to support the team as they continue to grow and are enthusiastic about their early commitment to diversity and support for female engineers. “Legit has been dedicated to building a diverse staff from the beginning, not only because it makes sense businesswise but because it is the right thing to do,” said Dr. Anthony Bucci, Legit’s Chief Scientist. “Early on we reached out to organizations such as Black Girls Code, Girl Develop It, Girls Who Code and Women in Tech, to name a few. We have also taken great pains to ensure our interview process is not stacked against women or minorities.” Two extraordinary women recently joined the team. Dr. Mengke Hu joined as Legit’s Senior Research Scientist. She has a Ph.D. in applying NLP to patent data. Yuan Geng recently came onboard as the Head of Product, formerly leading product management for TripAdvisor and Amazon. Legit Founding Team: Matt, Jacob and Anthony. Eniac is incredibly excited to be supporting the Legit team in building the first AI company to create a solution for generating IP at scale. Check them out at legit.ai and follow on Twitter @legitpatents.

It is time to reinvent the U.S. legal system

99

it-is-time-to-reinvent-the-u-s-legal-system-1506ed3c4cee

2018-10-02

2018-10-02 18:48:16

https://medium.com/s/story/it-is-time-to-reinvent-the-u-s-legal-system-1506ed3c4cee

false

607

Medium Publication for Eniac Ventures

null

eniacventures

null

Eniac Ventures

news@eniacventures.com

eniacvc

VC,TECHNOLOGY,VENTURE CAPITAL,STARTUP,ENTREPRENEURSHIP

eniacvc

Venture Capital

venture-capital

Venture Capital

32,826

Eniac Ventures

We lead seed rounds in bold founders who use code to create transformational companies.

291154f857bf

eniacvc

810

93

20,181,104

null

0

null

0

null

2017-12-22

2017-12-22 10:16:52

2017-12-22

2017-12-22 10:17:46

2

false

es

2017-12-22

2017-12-22 10:17:46

2

15084cf58d9a

2.028616

0

Los dispositivos móviles se batallan por la inteligencia artificial

4

La batalla de los smartphones con machine learning Los dispositivos móviles se batallan por la inteligencia artificial En Movetia ya hemos hablado en varias ocasiones cómo la inteligencia artificial o el machine learning están revolucionando la manera de hacer y ver el mundo. Hace unos días hablábamos de las tendencias en tecnología digital de cara al 2018 y claro está que estas dos tendencias tienen un papel importante. De hecho, un reciente estudio de ABI Research “Industry Survey: Transformative Technology Adoption and Attitudes — Artificial Intelligence and Machine Learning” afirma que esta tecnología está destacando frente a otras, generando aplicaciones y usos de inteligencia artificial y machine learning. Así, según el informe, se prevé que para el año 2022, más de 650 millones de dispositivos móviles admitirán aplicaciones de visión avanzadas. Los teléfonos pasarán de retocar fotos en la nube a utilizar la visión para identificar a los propios usuarios y reconocer su estado de ánimo. Así que ahora ya no hablamos de teléfonos grandes, mejores cámaras o diseños más innovadores. Hablamos de smartphones con chips neuronales, procesadores inteligentes, asistentes virtuales y sistemas de machine learning capaces de aprender cada uno de nuestros movimientos hasta anticiparse y predecir nuestras necesidades. Nuevos modelos smartphone con tecnología machine learning Este año 2017 han aparecido en el mercado smartphones con tecnología machine learning cuyas armas más poderosas son la potencia de cómputos y los datos de usuario (bueno, quizás también tenga algo que ver que se trata de gigantes tecnológicos capaces de invertir en inteligencia artificial y situarse diez pasos más allá que el resto…). Apple, con su nuevo iPhone X y el procesador A11 Bionic, ha sido el primero en presentar un smartphone con tecnología biométrica capaz de desbloquear el terminal a través el reconocimiento facial, capturando y analizando más de 50 movimientos de los músculos de la cara. El Pixel 2 de Google es capaz de reconocer la música que suena a nuestro alrededor en tiempo real gracias al aprendizaje automático. Además este smartphone incorpora Google Lens, herramientas de inteligencia visual que permite reconocer objetos, lugares o personas. Se trata de una tecnología basada en avances en visión artificial y aprendizaje automático. El nuevo Mate 10 de Huawei, tercer mayor fabricante de smartphones del mundo, ha incluido un procesador que cuenta con una unidad dedicada de procesamiento neuronal. Ello permite que la cámara del Mate 10 pueda detectar su objetivo y ajustarse automáticamente en función de lo que se va a fotografiar. Todo ello nos lleva a predecir un futuro muy presente casi automatizado, capaz de hacernos la vida más fácil. Y esto es solo el principio.

La batalla de los smartphones con machine learning

0

la-batalla-de-los-smartphones-con-machine-learning-15084cf58d9a

2018-05-22

2018-05-22 14:45:49

https://medium.com/s/story/la-batalla-de-los-smartphones-con-machine-learning-15084cf58d9a

false

436

null

Machine Learning

machine-learning

Machine Learning

51,320

Movetia

Consultoría, #diseño y #tecnología para transformar procesos empresariales en #serviciosdigitales eficientes para un mundo en movimiento. Barcelona + Madrid

62763f0ec854

Movetia

102

88

20,181,104

null

0

null

0

f5af2b715248

2018-06-18

2018-06-18 03:49:42

2018-06-20

2018-06-20 11:09:24

13

false

en

2018-06-25

2018-06-25 17:29:25

35

1508a0596d3c

7.950943

81

0

“Technology leadership in artificial intelligence for drug discovery and biomarker development, academic excellence, extensive…

5

We are an industry now: the emergence of AI-powered longevity biotechnology industry “Technology leadership in artificial intelligence for drug discovery and biomarker development, academic excellence, extensive collaborations with pharmaceutical and consumer companies, novel methods of attracting top talent, and increasing global reach have allowed Insilico Medicine to build a credible and sustainable business model in the nascent longevity biotechnology industry,” noted Neelotpal Goswami. “In recognition of its pioneering research and ability to introduce novel products and solutions for age management, Frost & Sullivan is pleased to present it with the 2018 Technology Innovation Award.” In 2004, I decided to quit the computer hardware industry and focus on extending human productive longevity. For years I have been working to deliver tangible research results and popularize aging research as the most important cause from any perspective, be it quantified altruism or macroeconomic reasons. This cause knows no geographical boundaries, and makes every human life more valuable. But in this world, the path to popularity lies through economic incentive and the bigger, the better. When I published our news on the investment from the global biotechnology giant WuXi AppTec, Pavilion Capital (Singapore’s Temasek), BOLD Capital Partners, and Juvenescence, I did not expect the news to go viral. We received broad media coverage from both general and pharmaceutical industry outlets including Endpoints, Fierce Biotech, TechCrunch and Forbes. GenomeWeb covered our work with George Church’s Nebula Genomics and we were mentioned in the Financial Times. It was a great week for the longevity therapeutics industry in general. Juvenescence announced the close of its $50 million series A to invest in the many longevity-oriented assets and companies. In addition, AgeX Therapeutics (Mike West and Aubrey de Grey) raised $5 million to go after the most recent advances in regenerative medicine. My team and I also presented at CogX and Founders Forum last week. CogXis one of the most revered events in AI, bringing together over 6,000 entrepreneurs, financiers, tech conglomerates and the British Government.Founders Forum, one of the premier events for digital and technology entrepreneurship did not disappoint either. With a roster of world class entrepreneurs, CEOs and investors longevity and AI were among the key highlights.. As I explained in my first Medium blog at Insilico Medicine we strive to use the latest advances in AI to understand aging, use aging research to accelerate the research in AI and make the research more interpretable. Just a couple weeks ago Y-Combinator had their usual Summer enrollment challenge and it was heavily focused on the companies bridging AI and aging research. They are in stealth mode right now, which is their usual style and we do not know who the winners will be. But a lot of capital has now been deployed, so in about one year there will be a plethora of companies coming out of the most professional incubation and acceleration program in the world. But this week AI for aging research was highlighted for the first time by Frost & Sullivan, with their first 2018 North American Artificial Intelligence for Aging Research and Drug Development Award for Insilico Medicine. The original press release from Frost & Sullivan This was a very serendipitous award. A few weeks ago I received an email from Frost and Sullivan, stating that the their analysts had completed a review of Insilico Medicine. I get a few thousand emails a week, and never open all of them. This one in particular looked like a promotional email, but by some lucky chance I opened it and noticed both a brief report on Insilico Medicine and a message saying we had received the award! The report was comprised of information and facts available online from the press releases and research papers etc. But they took such a deep dive into our published research papers that it made me shiver. We usually publish proof of concept studies without revealing the place of the study in our core engine or drug discovery pipeline, soit was a bit scary that someone else could put all the pieces of our puzzle together. I added the full report below. Another serendipitous event was a visit by a delegation from Taiwan headed by the president of DCB Taiwan on the same day we received the award. The people at DCB are extremely hard working and it was amazing that they agreed to spend the weekend to meet with us, as I had to leave for London on Monday for the London Technology Week. When I showed the award to DCB team, it turned out that they were is already working with Frost & Sullivan and they are an analytics and education partner of choice for AI, blockchain and other industries. It would be great to see them start the analytical and education series on the AI for aging research and drug discovery with our Insilico Taiwan team. Last year we were picked by CB Insights, a leading technology analyst, as one of the top 100 AI Companies for 2018. Receiving this award was a great motivator for our team. CB Insights also covered us in the longevity space. However, the Frost & Sullivan award is unique in that it bridges AI and aging research and means there is now more than one player in this field. What we are witnessing is the emergence of these two rapidly expanding and converging fields. Frost & Sullivan is probably the most reputable industry analysts, and I used their reports when I was at school in the late 90s and then when working for PMC-Sierra and ATI Technologies in the early 00’s. Their reports are often the basis of business plans, influencing strategy and financial projections, growth forecasts, and market size estimates Recently they published two industry reports: Anti-Aging Therapies and Services Market — Trends and Growth Opportunities, Forecast to 2022 — Focus on Integrative Solutions Combining Dietary Supplements and Aesthetic Solutions with Clinical Therapeutics as the Market Moves Toward a ‘Prevent-Manage-Repair’ Continuum Clinical Therapeutics, Dietary Supplements and Aesthetic Solutions Propel the Anti-aging Market to Hit $85.6 Billion by 2022 These reports are the first birds covering the birth of the still nascent but rapidly advancing and longevity biotechnology industry. One of the brightest minds in modern finance, who previously bet on some of the most unlikely trends and companies and won, Jim Mellon, covered the sectorcomprehesnively in his book called “Juvenescence” which every modern investor should read. And recently, Brian Bergstein covered Jim Mellon’s work and vision in his Medium article at Neo.Life started by Jennifer Metcalfe, the co-founder of Wired. It feels like a throwback to the the hype and opportunity of the internet era, when many promises were not realized in the short term, but were far exceeded in the long run. And the longevity biotechnology industry is certainly an industry where everyone wins even when there are short-term failures along the way the industry keeps growing. The industry’s goal is to make every human life more valuable — and we should not slow down in any way shape or form. Longevity biotechnology is a rapidly growing industry and Peter Diamandis is one of the world’s most informed futurists, exponential educators and business people. On June 17th he wrote an article on the Future of AI and Longevity describing this trend and it went viral. Here is his original post. I highly recommend looking at the Frost & Sullivan reports on the aging industry. Every large pharmaceutical and “Big Tech” company should have aging research and longevity biotechnology as part of their corporate strategy. This is not just an altruistic cause but a major opportunity for everyone. Frost & Sullivan summarized these opportunities in just one simple graphical teaser. Source: Frost & Sullivan. Anti-aging Therapies and Services Market — Trends and Growth Opportunities, Forecast to 2022 There is an enormous threat for the governments of the developed countries already running substantial deficits due to their aging populations and, at the same time, an enormous opportunity for the pharmaceutical companies and startups to make extraordinary profits while pursuing this altruistic cause. Source: Frost & Sullivan. Anti-aging Therapies and Services Market — Trends and Growth Opportunities, Forecast to 2022 In this report Frost & Sullivan covered the Buck Institute for Research on Aging as one of the undisputed leaders in the longevity R&D space. The CEO of the Buck Institute, Dr. Eric Verdin, is an established scientist with H-index exceeding 100 (100 papers with over 100 citations), and an outstanding leader, who raised millions for top-notch research. He isalso focused on developing a center for AI for aging research at the Buck Institute. Longevity research and AI are rapidly converging, and we expect to see many advances in this area. The Buck Institute for Research on Aging We collaborate with the companies that are far removed from healthcare to take the AI-powered longevity biotechnology industry to the next level. One of these collaborators, Neuromation, is contributing their enormous supercomputing cloud and AI expertise to the joint effort and I highly recommend following their chief science officer on Medium. The Frost & Sullivan award report also noted of our important partnership with the BitFury Group called Longenesis presented for the first time by the CEO of BitFury, Valery Vavilov at the Global Leaders Forum in Korea. Valery Vavilov presenting Longenesis at the Global Leaders Forum in Korea Here is the link to the video. Longenesis has also partnered with George Church’s Nebula Genomics. It was a pleasure to co-present our demo at the CogX conference in London last week with Kamal Obbad, Nebula’s CEO. Kamal Obbad, the CEO of Nebula Genomics presenting our partnership at Cognition X in London Conclusion Longevity biotechnology is coming of age. In my opinion, AI is the best way to advance aging research and build sustainable models with the pharmaceutical industry. The pharmaceutical industry may be a bit slow to embrace both AI for aging research and the emerging business models. Meanwhile, while big tech” and large consumer companies are reluctant to venture outside the basic digital health arena into the more regulated area of pharma. However, we can work with both of these worlds on their terms. This award — and the broad recognition we and our nascent industry have received — is proof that the AI-powered longevity biotech industry is becoming more credible and robust. We are poised to change business models and achieve both great business results and immeasurable human gains. Bridging consumer & pharmaceutical insights www.insilico.com We are very proud to receive this important award and recognition in the seminal report from Frost & Sullivan. The full-text of the report is available here. This story is published in The Startup, Medium’s largest entrepreneurship publication followed by 338,320+ people. Subscribe to receive our top stories here.

We are an industry now: the emergence of AI-powered longevity biotechnology industry

3,128

we-are-an-industry-now-the-emergence-of-ai-powered-longevity-biotechnology-industry-1508a0596d3c

2018-06-25

2018-06-25 17:29:26

https://medium.com/s/story/we-are-an-industry-now-the-emergence-of-ai-powered-longevity-biotechnology-industry-1508a0596d3c

false

1,736

Medium's largest publication for makers. Subscribe to receive our top stories here → https://goo.gl/zHcLJi

null

The Startup

null

swlh

STARTUP,TECH,ENTREPRENEURSHIP,DESIGN,LIFE

thestartup_

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Alex Zhavoronkov, PhD

Founder of Insilico Medicine, adj. prof at the Buck. I speak at 40+ events per year on AI for drug discovery and longevity research. PubMed: Zhavoronkov+A

51d4c45c9248

longevity

811

588

20,181,104

null

0

null

0

863f502aede2

2018-04-29

2018-04-29 14:38:31

2018-04-29

2018-04-29 14:40:48

5

false

en

2018-04-29

2018-04-29 14:41:48

16

150980e4860c

5.705031

2

0

If you are familiar with the biotech industry, you know Cambridge. The small city at the center of the Great Boston Area hosts over 1,000…

5

The Past, Present & Future of Biotech in Boston If you are familiar with the biotech industry, you know Cambridge. The small city at the center of the Great Boston Area hosts over 1,000 biotechnology-related companies. Most of these companies cluster around Kendall Square, the same block as the Massachusetts Institute of Technology (MIT). Over the past decade the number of biotech jobs in Boston has jumped by 37%. [1] With the rapid rise in the application of Artificial Intelligence, and given the promising future of bioinformatics and computational biology, the local biotech industry finds itself in the midst of a technological revolution. Why Did Biotech Choose Boston? The biotech industry in Boston can be traced back to the 1970s, when molecular biology was in its golden age. The idea of “playing with genes” however made many uncomfortable at the time. Cambridge City Council held a hearing on DNA experiments, and granted permission to Biogen, a new, local company founded by MIT Professor Phillip Sharp. Biogen was the first US firm to get the green light for genetic engineering. [2] Bio-pharmaceutical companies quickly poured into Kendall Square, creating a well-rounded, self-vitalizing biotech ecosystem and building a global centre for biotechnology. Academic Background One key factor in Boston’s rise in biotechnology is the area’s academic resources, which are second to none. Along with traditional biomedical schools such as Harvard Medical School and MIT Whitehead Institute for Biomedical Research, there are also a number of interdisciplinary programs combining biology and other informatic engineering subjects, such as the MIT Computational and Systems Biology Initiative (CSBi) and the Department of Biostatistics at Harvard. Other universities such as Tufts also have their own bioinformatics research groups. Boston’s many universities are a talent pipeline, and most of the area’s biotech company founders are graduates or professors from Harvard, MIT or other top universities. [4] Research Resources Besides laboratories in universities, Boston’s biotech industry is also backed by labs in hospitals and large pharmaceutical companies. Boston has three respected medical schools, two pharmaceutical schools, and three general hospitals. Having top hospitals not only provides more research facilities, but also more disease study opportunities. It is extremely difficult for example to do clinical testing and study on certain rare diseases which appear early in life and can claim the lives of 30% of patients before age five. The Boston Children’s Hospital International Health Services division treats young patients from over 100 countries every year, and these treatment cases can inform the biotech research ecosystem. Large medical companies are becoming more dependent on smaller biotech companies in the research field. As rising costs, patent cliffs, and other factors eat into pharmaceutical industry profits, many big medical companies are turning to Boston’s innovative biotech startups to provide high-quality R&D results at lower costs. By the end of 2017, pharma giants Genzyme, Novartis, Pfizer, and Baxter had all established presences in Boston largely for this reason. Non-technical Support Capital is an essential element in building a vibrant industry ecosystem, and as biotech has grown so has the money behind it. General venture capital and business companies are involved, along with investors specifically targeting biotechnologies. Fidelity Biosciences, with US$2 trillion in assets under management, was one of the earliest venture capital companies to focus on life sciences, specifically “Biopharmaceuticals, MedTech, and Healthcare IT/Services in a stage-agnostic fashion.” [5] Other capital and consulting firms in this market include Flagship Ventures, MPM Capital, Locust Walk Partners, Voisin Consulting, and Fuld & Co. What’s more, Biotech is an area where government can support businesses. Last year Massachusetts Governor Charlie Baker announced plans to invest US$500 million over the next five years in the Massachusetts Life Sciences Initiative. Baker’s predecessor, Gov. Deval Patrick, had launched a US$1 billion initial investment in biotech back in 2008. [6] Future: Opportunities and Challenges Respected biotechnology media company FierceBiotech publishes an annual “Fierce 15” list of the world’s most promising biotech companies, which it believes can make future breakthroughs. Cambridge area companies have taken about one-third of the spots on the list over the last five years, an achievement no other city has even approached. (2016 Fierce 15 List includes 5 companies from Cambridge) [11] (2015 Fierce 15 List includes 7 companies from Cambridge) [12] Opportunities in Gene Editing What are the most popular areas in biotech industry today? Gene editing is undoubtedly one, especially with the discovery of the CRISPR/Cas9 Method in 2011. CRISPR, for Clustered Regularly Interspaced Short Palindromic Repeats [7], is a family of DNA sequences in bacteria that contains snippets of DNA from viruses that have attacked the bacterium. These sequences play a key role in a bacterial defence system, and form the basis of a genome editing technology known as CRISPR/Cas9, which allows permanent modification of genes within organisms. [8] With its huge application value in genome engineering, gene knockdown/activation, disease models, biomedicine and more, CRISPR technology has fostered many innovative Boston startups. Zhang Feng is the MIT Professor who first successfully applied CRISPR/cas9 in eukaryotic cells, and secured a patent for this method in 2017. [9] Zhang co-founded Editas Medicine, one of the leading Boston companies focused on this technology. Another is CRISPR Therapeutics, which was co-founded by Emmanuelle Charpentier, another seminal figure in the field. In 2015, Editas Medicine received US$120 million in Round B funding. Investors included Flagship Ventures (15.6%), Third Rock Ventures (15.6%), Polaris Venture Partners (15.6%), and bng0 under Bill Gates (9%). [10] In 2016, Editas Medicine became the first CRISPR gene editing company to make an IPO. Editas plans to start testing CRISPR in treating blindness, which would be the first instance of editing human genomes using CRISPR. Other promising players in the gene editing field include Intellia and Bluebird Bio — whose stock price has risen sixfold since it went public. Future Challenges The biotech industry in Boston is booming, while facing potential challenges. Life sciences research involves a heavy time commitment in the design and execution of experiments, and this can send costs skyrocketing. Editas Medicine for example remains stuck in the preparation phase for the blindness treatment plan it announced in early 2015, although it confirmed investments of up to US$20 million at its 2016 IPO. This situation is normal in the industry. The success of a biotech company thus depends not only on its technology, but also on the confidence of the capital behind it, which will support it through such delays. Silicon Valley is another challenge for Boston. The two regions are virtually neck-and-neck in the biotech race, but as the industry starts relying more on software, wearable devices, and big data analysis, etc, this could tilt the contest in favour of Silicon Valley, which has a definite advantage in these tech areas. Reference [1] Mass Bio [2] http://news.yikexue.com/archives/50527 [3] CSBi official website http://csbi.mit.edu/overview/index.html [4] https://www.yypharm.com/?p=2502 [5] F-Prime http://fprimecapital.com/about/ [6] How Boston Became ‘The Best Place In The World’ To Launch A Biotech Company http://www.biospace.com/news_story.aspx?StoryID=460414&full=1 [7] Sawyer E (9 February 2013). “Editing Genomes with the Bacterial Immune System”. Scitable. Nature Publishing Group. Retrieved 6 April 2015. [8] Zhang F, Wen Y, Guo X (2014). “CRISPR/Cas9 for genome editing: progress, implications and challenges”. Human Molecular Genetics. 23 (R1): R40–6. PMID 24651067. doi:10.1093/hmg/ddu125. [9] https://www.nytimes.com/2017/02/15/science/broad-institute-harvard-mit-gene-editing-patent.html [10] http://www.fiercebiotech.com/r-d/gates-backs-a-120m-breakout-round-for-crispr-cas9-pioneers-at-editas [11]http://www.fiercebiotech.com/special-report/fiercebiotech-s-2016-fierce-15 [12]http://www.fiercebiotech.com/special-report/fiercebiotech-s-2015-fierce-15 [13]http://www.biotech.org.cn/information/139512 [14]http://weibo.com/ttarticle/p/show?id=2309403968140475272541 [15]http://www.istis.sh.cn/list/list.aspx?id=1867 * * * Analyst: Rui Sun| Editor: Robert Tian, Michael Sarazen * * * Subscribe here to get insightful tech news, reviews and analysis! * * * The ATEC Artificial Intelligence Competition is a fintech algorithm competition hosted by Ant Financial for top global data algorithm developers. It focuses on highly-important industry-level fintech issues, provides a prize pool worth millions. Register now!

The Past, Present & Future of Biotech in Boston

52

the-past-present-future-of-biotech-in-boston-150980e4860c

2018-05-09

2018-05-09 20:15:40

https://medium.com/s/story/the-past-present-future-of-biotech-in-boston-150980e4860c

false

1,291

We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

null

SyncedGlobal

null

SyncedReview

global.sns@jiqizhixin.com

syncedreview

ARTIFICIAL INTELLIGENCE,MACHINE INTELLIGENCE,MACHINE LEARNING,ROBOTICS,SELF DRIVING CARS

Synced_Global

Science

science

Science

49,946

Synced

AI Technology & Industry Review - www.syncedreview.com || www.jiqizhixin.com || Subscribe: http://goo.gl/Q4cP3B

960feca52112

Synced

8,138

15

20,181,104

null

0

ssh <username>@aspire.nscc.sg # Eg. ssh kelvin321@aspire.nscc.sg jupyter lab --no-browser —-port=8889 module avail module load <module name> # Eg. module load anaconda/3 ssh -N -L 8888:localhost:8889 <username>@aspire.nscc.sg qsub -I -q gpu -l select=1:ncpus=4, walltime=01:00:00 -P Personal runipy -o <notebook_name.ipynb>

6

null

2018-04-03

2018-04-03 05:13:49

2018-04-04

2018-04-04 23:15:59

3

false

en

2018-04-05

2018-04-05 12:45:26

1

150ab02b4d73

2.833019

5

0

This is a short guide for those who might need some help in getting NSCC to work. Unlike services like AWS or Goggle Cloud Compute, getting…

5

6 steps to take off your Deep Learning project in NSCC This is a short guide for those who might need some help in getting NSCC to work. Unlike services like AWS or Goggle Cloud Compute, getting your Jupyter notebook to display on your local browser or even use the GPU nodes in NSCC can be challenging. However, if you follow these simple instructions closely, you should be able to get your project up and running in no time at all! This guide assumes you have obtained your user account and password from NSCC. Please registration here if you haven’t done so. You will receive an email from NSCC with your account and password, alongside instructions for setting up NSCC SSL VPN service. Once done, proceed to follow these steps. Step1: Login into VPN Tunnelblick Login interface for NSCC SSL VPN service (Mac OS) Do take note you will need to append the 6 digit pin from your authenticator after your password (no space in between). Step 2: Login into NSCC from local machine Enter password when prompted (without the 6 digit pin this time). Step 3: Start Jupyter lab on remote machine We are setting Jupyter to run with no browser on the remote machine because otherwise, Jupyter will instead open in a program called Elinks. I prefer using Jupyter notebook/lab within a browser since it is interactive and more user-friendly. If you havn’t set up your environment (therefore jupyter notebook does not run), you have two available options to get it ready. Option 1: Load existing modules from NSCC List of modules available in NSCC Option 2: Pip install anaconda3 and whatever else you need within your login node. (I prefer this to avoid module loading everytime) If the download fails during pip, simply download the installation package from your local machine and transfer them via remote manager into your remote root directory. pip install packages offline. Step 4: Setup a SSH tunnel from local machine Enter your password (without the 6 digit pin) when prompted. This step is done so you can now run Jupyter notebook/lab from your local browser. Yeah! Copy the url from the remote machine and paste into your browser Set port to 8888 Step 5: Running a job in GPU node on remote machine Congrats on getting your jupyter notebook/lab up! However, if you try to use GPU now, you will not be able to =/ This defeats the whole point of using NSCC right? We want the computational power! Unfortunately, there is no easy way to run jupyter notebook with GPU enabled like how you would do so in AWS or Goggle Cloud Compute. Instead NSCC wants us to run our jobs in GPU node like this: So go ahead and start a job session with the code above. I’ve set the time needed for the job to 1 hour and requested for 4 cpus and 1 gpu. Step 6: Running and saving your notebook from GPU node on remote machine For the last step, simply save all the codes you need to run on the GPU into a ipynb file. Then run the following code. After it’s completed, you may open the file in your jupyter notebook/lab and copy the cells into your main project notebook. I hope this post has helped you in some way. If so, I would appreciate if you can leave a clap or two as feedback. Thank you and appreciate your time in reading this!

6 steps to take off your Deep Learning project in NSCC

9

6-steps-to-take-off-your-deep-learning-project-in-nscc-150ab02b4d73

2018-05-05

2018-05-05 10:09:54

https://medium.com/s/story/6-steps-to-take-off-your-deep-learning-project-in-nscc-150ab02b4d73

false

605

null

Python

python

Python

20,142

Kelvin Tham

Engineer with a passion for Machine Learning, Computer Vision and Improvement Science. Currently working as a Specialist in Healthcare sector, Singapore.

5ad01ef486a7

kelvinthamkh

8

18

20,181,104

null

0

null

0

863f502aede2

2018-02-20

2018-02-20 20:19:29

2018-02-20

2018-02-20 20:21:16

1

false

en

2018-02-20

2018-02-20 20:21:16

2

150d1328d56e

2.25283

11

0

A retinal fundus image is a photograph of the back of the eye taken through the pupil. For more than 100 years these images have been used…

5

It’s All in the Eyes: Google AI Calculates Cardiovascular Risk From Retinal Images A retinal fundus image is a photograph of the back of the eye taken through the pupil. For more than 100 years these images have been used for detecting eye disease. Now Google has introduced a surprising new use for retinal images: combined with artificial intelligence, they can also predict a patient’s risk of heart attack or stoke. Research arm Google Brain today published a paper in the journal Nature Biomedical Engineering which demonstrates how deep learning models can use retinal images to detect a patient’s age, gender, smoking status and systolic blood pressure; calculate cardiovascular risk factors; and predict the risk of major adverse cardiac events occurring over the next five years. A problem with today’s mainstream cardiovascular risk calculators such as the Pooled Cohort Equations, Framingham, and Systematic Coronary Risk Evaluation is that they require the input of multiple features such as blood pressure, body mass index, glucose and cholesterol levels, etc. to generate a disease risk result. A study by the American College of Cardiology’s Practice Innovation And Clinical Excellence Program concluded that the data required to calculate 10-year risk was available for less than 30% of patients. Google Brain discovered that a retinal fundus image alone was sufficient to predict many cardiovascular risk factors. The anatomical feature patterns were extracted using a convolutional neural network — a computational model that excels in analyzing images. Researchers trained models on retinal images from 284,335 patients and validated on two independent datasets of 12,026 and 999 patients. The trained model identified patients’ ages with 3.26 years, distinguished gender 97 percent of the time, spotted a smoker 71 percent of the time, and calculated blood pressure with a 11.23 mmHg margin of error. Google Brain then took a step forward. Researchers discovered the trained model could predict a patient’s risk of cardiovascular disease over the next five years 70 percent of the time, approaching the accuracy rate of established risk calculators without all the additional data inputs. Deep learning is often criticized for its lack of transparency and interpretability, and this has hindered the technology’s entry into areas such as medical health and the legal system. But Google Brain believes their methodology is sound. It employs attention techniques to determine which pixels are the most important for predicting a specific cardiovascular risk factor: blood vessels for example are a critical feature for determining blood pressure. This is not the first time Google Brain has leveraged the value of retinal images. In November 2016 it presented a study on deep learning for early detection of diabetic retinopathy, which could potentially protect 415 million worldwide diabetics from irreversible blindness. Google Brain’s paper opens up the exciting possibility of applying deep learning to retinal images for improving diagnoses beyond eye disease. Will AI be the key that unlocks even more medical science innovations? Journalist: Tony Peng| Editor: Michael Sarazen Dear Synced reader, the upcoming launch of Synced’s AI Weekly Newsletter helps you stay up-to-date on the latest AI trends. We provide a roundup of top AI news and stories every week and share with you upcoming AI events around the globe. Subscribe here to get insightful tech news, reviews and analysis!

It’s All in the Eyes: Google AI Calculates Cardiovascular Risk From Retinal Images

26

its-all-in-the-eyes-google-ai-calculates-cardiovascular-risk-from-retinal-images-150d1328d56e

2018-06-14

2018-06-14 14:55:56

https://medium.com/s/story/its-all-in-the-eyes-google-ai-calculates-cardiovascular-risk-from-retinal-images-150d1328d56e

false

544

We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

null

SyncedGlobal

null

SyncedReview

global.sns@jiqizhixin.com

syncedreview

ARTIFICIAL INTELLIGENCE,MACHINE INTELLIGENCE,MACHINE LEARNING,ROBOTICS,SELF DRIVING CARS

Synced_Global

Google

google

Google

35,754

Synced

AI Technology & Industry Review - www.syncedreview.com || www.jiqizhixin.com || Subscribe: http://goo.gl/Q4cP3B

960feca52112

Synced

8,138

15

20,181,104

null

0

null

0

null

2018-06-30

2018-06-30 20:28:49

2018-06-30

2018-06-30 22:50:25

7

false

en

2018-07-02

2018-07-02 02:52:43

4

150d917f9eac

5.582075

3

0

If you spend time online or get updates from the tech industry from time to time, you must have heard about the buzzwords “Deep Learning”…

5

What is Deep Learning ? (An explanation for TechNoobs) If you spend time online or get updates from the tech industry from time to time, you must have heard about the buzzwords “Deep Learning”, “Machine Learning”, “Artificial Intelligence” and wondered what exactly is that? Is it about the era of AI as shown in movies? In this article, I aim to decipher its meaning in the simplicity that you can explain it to your grandparents! 😉 TL;DR coming straight to the point. Deep Learning refers to programs that enables a computer to do something that comes naturally to humans : learn by example. These programs are smart enough to actually “learn” things, without needing someone to explicitly code them. What happens is that the programmer will design a “model”. There is some example data that we give to such a model to show that — “if this is input, this is output”, and the model would learn to map them correctly. After it learns that, we can give it a similar input and it would provide us with an output. Too hard to understand? Here’s an example to help you. Suppose you have lots of pictures (say a million), and a lot of them contain cats. You want a program to identify which of these pictures contain a cat. What you can do is, you take some of the pictures randomly (say a thousand) and label them as a “cat image” and “non-cat image”. You give one such model as explained above these labeled images. Once this model goes over these pictures a few times, it learns to recognize a cat. You can then run this model against the remaining 990,000 pictures and it would label all of them as a “cat image” or a “non-cat image”. Do you see its power yet? If you’re interested in knowing more about it — keep reading. There are instances when a model shows accuracy higher than a human! (I know you thought the left picture was of an airplane too. :P) What’s the magic behind it? How is a computer program able to “learn”? Have you noticed how babies learn to identify daily objects, they learn to speak words and do various activities over time. They observe the environment around them, and they learn from what we teach them. We tell them that this 🍌 is a banana and this 🍎 is an apple and they recognize it over time. Our human neural system is made up of specialized cells called neurons (about 100 billion of them) and these neurons are what enables us to think. When we see an object, our eyes passes the raw image information to the neurons and they recognize what objects that image contains. Similarly, the secret behind Deep Learning is a Neural Network which is made up of neurons (but digital). A basic Neural Network Architecture. A Neural Network basically consists of a bunch of neurons stacked in layers as shown in the figure. The neural network basically tries to fit a complex mathematical function that could map a given input to the output. Each neuron contributes a factor in shaping that function. The input layer is involved with taking the input and understanding it. The hidden layers add weighted factors to the mathematical function to try to link the input to the output layer where the output is generated. These weights are updated through an algorithm like Gradient Descent to try to get as close to the final function that could give the most probable output. What’s so cool about Deep Learning? Well, everything! You have actually seen its applications everywhere, but you didn’t know Deep Learning is making it possible. Have you used digital personal assistants like Siri on an Apple device, Alexa by Amazon or Google Assistant by you-know-who on Android phones? These AI bots understand your words (text or voice) and give you an appropriate reply — That’s Deep Learning. When you see a few products on Amazon and you see similar products being recommended to you, or when you watch FIFA Highlights on YouTube one day and next day you see all Football/Fifa related videos being recommended to you — That’s Deep Learning. Have you heard about self-driving or driver-less cars? Yes, that’s not a joke...they exist. That’s been made possible by guess what — Deep Learning! Have you seen face-unlock feature or fingerprint scanner on smartphone devices? Yes — That’s Deep Learning too. Have you uploaded a picture on Facebook and you see that the faces are sometimes automatically recognized without you ever telling them? — That’s Deep Learning. You obviously have heard about the cool new feature of the iPhone X called “Animoji”. It works because of — Deep Learning. Everybody has used Google Translate to translate one language to another — that works on Deep Learning too. This is just the tip of the iceberg, the list is endless!! New applications and research papers are coming out every week with more cool stuff. Some really cool applications are Image Captioning, Image Generation from sentences, A storyteller that creates stories from an image (see picture below) , Music composition, AI that sees in dark, etc. One article is too short to cover them. A Neural Story Teller that generates romantic story from an image. How is Deep Learning related to Artificial Intelligence and Machine Learning? You must have heard the terms “Artificial Intelligence” and “Machine Learning” more often than “Deep Learning”. Even if you haven’t, here’s a quick definition: “Artificial Intelligence is the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.” You have probably heard about Artificial Intelligence (AI) more in movies such as “I,Robot”, the “Terminator” series or TV Series such as the “Westworld”. They are mostly depicted as robots who are as smart as humans but are able to process and compute things on a much larger scale and speed. “Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.” You may have heard this from people and right now, you might be a little confused about the difference between Machine Learning and Deep Learning. It’s all right. Deep Learning is actually a particular kind of Machine Learning which revolves around Neural Networks. Machine Learning is more about algorithms that predict for you, but if there’s an inaccurate prediction, generally an engineer has to step in to resolve it. Deep Learning has a way of checking the accuracy and growing itself. As clear from the picture, Deep Learning is actually a subset of Machine Learning which in turn is a subset of Artificial Intelligence. The AI right now is made by combining Deep Learning and Machine Learning models at a lower level and adding to them. I have been into Deep Learning since the past few months and its so cool that I want to tell more people about it. I’m sure many people working on it get asked this question by people with no technical background. So my aim was to explain the idea as simply as possible. This was my first article on Medium, hope you liked it. I’d appreciate any feedback, feel free to comment or get in touch through email, LinkedIn, Facebook or Whatsapp. \(n_n)/

What is Deep Learning ? (An explanation for TechNoobs)

4

what-is-deep-learning-an-explanation-for-technoobs-150d917f9eac

2018-07-03

2018-07-03 11:49:14

https://medium.com/s/story/what-is-deep-learning-an-explanation-for-technoobs-150d917f9eac

false

1,201

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Aakash Choubey

CS Undergrad | Web Developer | Deep Learning Enthusiast

3ae867130c1d

choubeyaakash77

3

6

20,181,104

null

0

null

0

null

2018-04-16

2018-04-16 21:28:56

2018-05-04

2018-05-04 21:15:31

2

false

en

2018-05-04

2018-05-04 21:15:31

8

150f170ce0e3

1.960692

3

0

|by Catherine Suski; Marketing, Citizen Health Ambassador

5

Data Science for Health: The Whats and Whys |by Catherine Suski; Marketing, Citizen Health Ambassador “Long lines of tiny speckles on light in a high-ceiling interior” by Joshua Sortino on Unsplash What if it were possible for a health system to integrate patient data from electronic health record systems, regional data on the strength of UV rays, skin cancer research from the past 10 years, and then offer sunscreen protection tips to patients susceptible to skin cancer? That may sound futuristic, but it’s closer than you think with the possibilities of data science. There’s been an explosion of talk about data science in healthcare and life sciences. A number of exciting real world projects are underway right now that showcase the potential: Google has created a deep learning algorithm that can detect the eye disease diabetic retinopathy by examining retinal images. Innoplexus offers a research platform for life sciences which ingests data from publications, clinical trials, conferences, and patents to speed up the drug development process. The Inova Translational Medicine Institute is building genetic models that help answer questions about individual predispositions to a disease, treatment and ultimately prevention. These new technologies are certainly creating a lot of buzz. Harvard Business Review even referred to the role of data scientist as “the sexiest job of the 21st century”. Is it just a new, more exciting term for statistics or is it truly a new discipline worthy of the hype? According to KDNuggets, a respected website and blog on big data and related topics, a data scientist needs expertise in computer science and software programming, statistics, written and verbal communication, and subject matter expertise. A candidate could be a great programmer with statistical ability, but if they lack industry specific knowledge, that person will need a lot of help to produce insights meaningful to the health or life sciences field. With all of those requirements, it’s not surprising skilled data scientists are in short supply. McKinsey estimates that there is currently a shortage of talent ranging from 140,000 to 190,000 open positions. The challenge isn’t related to hiring alone, as the position can be a frustrating one and retaining good staff can be tough. There’s no easy fix for remedying the shortages of talent in data science, but the ever-growing salaries are likely to get more people interested in pursuing it as a career. To keep up with the latest news and insights on data science, follow these publications: Bio-IT World Data Science Central Health Data Management KDNuggests Citizen HealthTalks Join the conversation next Thursday, May 10th- Link: Citizen HealthTalk Launch the site: Citizen Health

Data Science for Health: The Whats and Whys

33

data-science-for-health-the-whats-and-whys-150f170ce0e3

2018-05-13

2018-05-13 01:48:11

https://medium.com/s/story/data-science-for-health-the-whats-and-whys-150f170ce0e3

false

418

null

Data Science

data-science

Data Science

33,617

Citizen Health Team

We’re building an open health economy designed to elevate the health & prosperity of humanity.

ec606a230e13

citizenhealth

56

72

20,181,104

null

0

null

0

null

2017-09-16

2017-09-16 17:13:09

2017-09-17

2017-09-17 08:00:51

2

false

en

2018-02-01

2018-02-01 09:57:45

0

150ff8340a06

3.168239

34

2

0

AI’s most simple concept is terrifying

5

The Real Reason AI Terrifies the World’s Greatest Minds and Why it’s Inevitable Machines Will Take Over AI’s most simple concept is terrifying If it *only* reaches the same level of intelligence as us, it’s ability to operate at a far higher speed means that in 6 months it would have effectively operated for our equivalent of 500,000 years. It has taken us around 200,000 years to reach where we are now as a species Let that sink in for a moment In a period of 6 months for us, AI will have accumulated half a millions years worth of knowledge on top of everything we already know meaning that we will be unable to compete with them within days of them achieving the same levels of intelligence as us. We won’t even have time to react because machines will have surpassed us literally within the blink of an eye. In the same way we couldn’t relate to the earliest humans AI couldn’t relate to us This, at best relegates us to the role of how we treat pets now Or worse — Ants When ants stay out our way we leave them alone. When they invade our homes or obstruct our intentions we obliterate them without a second thought. They are an insignificance that can be eradicated. This is the likely outcome for AI and us It is why AI is a zero sum game. How would Russia or China react if they though the States were on the brink of such dominance? How would they react if there were even murmurs of rumour that it was close? This isn’t a technology that can be competed with — if you are 6 months ahead you have 500,000 years worth of knowledge more than the competition. The winner literally takes it all Then likely loses it all to the AI itself This exceeds the Manhattan project by several orders of magnitude, both in terms of danger and graveness to the future of humanity. This is why the world’s greatest minds are terrified This isn’t just the most likely outcome – it is inevitable If any rate of progress is assumed it is unavoidable that we reach this point in the future. It will be far sooner than we think In singular tasks, machine already bests us with ease. Chess, Jeopardy, and Go are all examples of this. Machines can lift far more than we ever could and calculate things that would take us years. A computers ability to operate computationally millions of times quicker than we are able to comprehend is an insurmountable barrier to our operation alongside them. It is why enabling human machine connectivity is critical to the survival of our species. Without it, we are worse than cavemen. It is critical that we expand our bandwidth as soon as possible. Yet people are scared of what Crispr means for development of a different class of human. Sure, genes may be edited to create a superior human, more intelligent, more beautiful, less susceptible to serious illness – but they will still be human. AI won’t be I know where my fears are placed If AI accumulates the equivalent of 2 years of our knowledge each minute we are finished. Potentially we already are — we have never taken a step back from an impending technological innovation because of it’s danger to the survival of humanity — just look at nuclear bombs. It doesn’t even matter if they are benevolent Eventually, they will be operating so far out of our realms of comprehension that their actions will affect us as a byproduct of their intentions. It’s as simple as that And that is if you assume they only reach the equivalent intelligent as us and never progress past that. But They will All that it now requires is a way to combine all these individual elements of intelligence into a singular AI that is able to make use of them all. Vertically they already exceed us in every area. Horizontally they don’t come close. This is to mean their general intelligence lets them down in the broadness of their capability. Once they achieve this it is game over In the blink of an eye a single machine has accumulated years of human level knowledge. The second our goals diverge the AI is in control of our destiny It’s evolve or face extinction Killed by our own creation

The Real Reason AI Terrifies the World’s Greatest Minds and Why it’s Inevitable Machines Will Take…

220

the-real-reason-ai-terrifies-the-worlds-greatest-minds-and-why-it-s-inevitable-machines-take-over-150ff8340a06

2018-02-01

2018-02-01 09:57:46

https://medium.com/s/story/the-real-reason-ai-terrifies-the-worlds-greatest-minds-and-why-it-s-inevitable-machines-take-over-150ff8340a06

false

738

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Chris Herd

Founder @Nexves, Entrepreneur, Angel Investor, ICO/Blockchain Advisor

da7b665f3cc7

ChrisHerd

31,328

3,629

20,181,104

null

0

null

0

b106b8d1f368

2018-02-08

2018-02-08 10:17:49

2018-02-08

2018-02-08 12:59:04

17

false

id

2018-02-13

2018-02-13 03:54:29

1

1515245c83e7

2.713208

0

Data Preprocessing 04

5

Memproses Category Data Data Preprocessing 04 Dalam memproses suatu dataset, kita harus memperhatikan jenis-jenis data agar tidak terjadi kesalahan proses. Salah satu jenis data yang akan dibahas dalam artikel kali ini adalah: Categorical Data. Mari kita lihat dataset dari project sebelumnya berikut ini: Isi dari dataset project sebelumnya Terlihat jelas bahwa Age dan Salary adalah jenis Numberical Data, sedangkan Country dan Purchased termasuk Categorical Data. Daftar Isi Python / Spyder R / RStudio Python / Spyder Buka Anaconda Navigator, lalu buka Spyder Lanjutkan project sebelumnya tentang Data Preprocessing 03 ENCODE KOLOM COUNTRY Tambahkan kode berikut ini: Kode Python untuk melabeli suatu categorical data Eksekusi kode diatas, sekaligus cek nilai X terbarunya seperti ini: Hasil eksekusi kode dan nilai X terbaru Dibandingkan dengan nilai X sebelumnya: Perhatikan bahwa sekarang value dari kolom Country telah berubah: France = 0 Spain = 1 Germany = 2 Tentu, angka 0, 1, dan 2 ini seharusnya tidak membuat Germany lebih baik dari Spain, dan Spain lebih baik dari France bukan? Itu sebabnya kita masih harus melakukan proses selanjutnya: Dummy Encoding / One Hot Encoding Ilustrasi hasil dari dummy encoding / one hot encoding Tambahkan kode berikut: Kode python untuk membuat one hot encoder Eksekusi kode tersebut, lalu lihat nilai X terbaru: Hasil eksekusi kode dan nilai X terbaru Agar lebih enak dilihat, lihatlah nilai X dari variable explorer Dobel klik pada X Hasil dari one hot encoding Jika dibandingkan dengan value awal kolom Country, adalah sebagai berikut: Perbandingan setelah dan sebelum diproses dengan one hot encoder ENCODE KOLOM PURCHASED Seperti yang sudah dijelaskan diatas, ada 2 kolom yang jenisnya categorical data, yaitu Country dan Purchased. Tambahkan kode berikut untuk meng-encode kolom Purchased: Kode python untuk meng-encode kolom Purchased Eksekusi kode tersebut, lalu cek nilai Y terbaru: Hasil eksekusi kode dan nilai Y terbaru Kita pun bisa memeriksanya lewat variable explorer: Perbedaan nilai kolom Purchased setelah dan sebelum di-encode Catatan: Hasil encode kolom Purchased ini tidak perlu diproses lagi dengan OneHot Encoder, karena tidak ada kerancuan antara label “Yes” dan “No” R / RStudio Buka RStudio Lanjutkan project sebelumnya tentang Data Preprocessing 03 Tambahkan kode berikut ini: Kode R untuk memproses categorical data kolom Country Eksekusi kode tersebut, lalu lihat isi dataset terbaru Hasil encode / transformasi categorical data kolom Country Lakukan hal yang sama untuk kolom Purchased Kode R untuk memproses categorical data kolom Purchased Eksekusi kode tersebut, lalu lihat isi dataset terbaru Hasil encode / transformasi categorical data kolom Purchased

Memproses Category Data

0

memproses-category-data-1515245c83e7

2018-02-13

2018-02-13 03:54:30

https://medium.com/s/story/memproses-category-data-1515245c83e7

false

295

Sumber belajar Machine Learning berbahasa Indonesia

null

rachmat.kukuh.rahadiansyah

null

Machine Learning ID

rkukuh@gmail.com

machine-learning-id

MACHINE LEARNING,PYTHON,R,INDONESIA,DATA SCIENCE

rkukuh

Python

python

Python

20,142

R. Kukuh

Modern Web, iOS/Swift, Machine Learning

b7563d2b13d7

rkukuh

294

14

20,181,104

null

0

null

0

null

2018-08-03

2018-08-03 16:12:35

2018-08-03

2018-08-03 16:18:16

0

false

en

2018-08-03

2018-08-03 16:18:16

1

151550623c53

1.045283

0

Searching the internet hands-free is one of the biggest advancements in recent years. In a nutshell, voice search is a two-way conversation…

3

Voice Search 101 Searching the internet hands-free is one of the biggest advancements in recent years. In a nutshell, voice search is a two-way conversation with virtual assistants aimed to help users complete different tasks in real time. It processes and transcribes human speech into text and then analyzes the text to detect commands and questions. Then, the virtual assistant reads the result into the external data sources to find relevant information and translates it in a digestible format to provide the user with what they need. Which Businesses Benefit Most From Voice Search? When it comes to voice search, 2018 has noted a giant step forward. Although it is still a growing trend, when we considered its operational nature, we discovered that it particularly gets along with several niches. In general, voice search is used for: Searching for general info on the internet; Getting directions; Calling or texting somebody; Checking the weather; Scheduling an alarm; Checking sports scores; Finding jokes; Playing music; Opening an app; Checking email. If your business includes one or more of these operations, you should undoubtedly consider voice search. Current Usage of Voice Search Based on the operations mentioned above, as well as the statistics research conducted, we also discovered the kind of information for which the voice search is used most. People use voice search to look up: Deals, sales, and promos Personalized tips and info Events and activity information Business information (such as store location) Customer service support If your business is associated with any of the above, include voice research in your SEO strategy ASAP! To read more about voice search, check out this article

Voice Search 101

0

voice-search-101-151550623c53

2018-08-03

2018-08-03 16:18:17

https://medium.com/s/story/voice-search-101-151550623c53

false

277

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Marco Haddaoui

Co-founder of Brossard Design, a software engineering development firm.

a2dd02c18bb8

marcohaddaoui

3

1

20,181,104

null

0

null

0

null

2017-10-17

2017-10-17 02:51:15

2017-10-29

2017-10-29 21:25:29

0

false

en

2017-10-30

2017-10-30 02:44:26

0

151645ed2dad

2.124528

7

0

Understanding the concepts of AI can be difficult, let alone applying those concepts in code. I’ll try to explain some of those complex…

5

Everyday Artificial Intelligence #1 — Intro to AI Understanding the concepts of AI can be difficult, let alone applying those concepts in code. I’ll try to explain some of those complex topics in this new succession of posts (to the best of my understanding) and provide resources and examples to apply them in code. I’m going to write at least 5 posts regarding concepts of AI, from neuroscience to processing individual pixels using a neural net, where I attempt to clarify and make simple the very complex field of AI that many (including myself) do not understand very well. I hope that you can learn alongside myself as I delve into the incredible journey of bright minds across the globe to create a truly bright mind. So, let’s jump right into it. What is the study of AI? The idea of describing human thinking in terms of a machine was actually born in the field of philosophy much, much earlier than the 20th century, where the invention of the programmable digital computer sparked the field of mathematical reasoning and the idea of building an electronic brain. In 1956, many bright minds came together at a workshop at Dartmouth. They discussed a project to build an intelligent machine, and were granted millions. These industry leaders were hit with hard times due to processing limitations, the underestimation of the difficulty of the project and the lack of funding from the US government. This period in the ’70s was known as the “AI winter.” Private and federal investment into AI surged again in the early 21st century due to the arrival of the dotcom bubble and the very real possibility of creating this new technology. Since then, countless businesses have incorporated AI into their workflow and demand for AI engineers has drastically spiked. The original goal of AI was to create a machine that was as intelligent as a human. I’d say that while this original goal is still kept in consideration by many, the industry has gotten too large and too aggressive to shoot for just this goal. Hobbyists may pursue a goal such as I did of just creating something fun and interactive like a chatbot or a robot, and businesses may want a bot to manage their social media accounts for them. Researchers may want to achieve successfully making a computer feel love, and private companies may want to create a technology that is smarter than humans. While ambitious, these goals are potentially dangerous and may even be impossible to achieve in the near (or far) future. Many people, while embracing the possibility of powerful new computing tech and revolutionary products, certainly consider the evident danger of developing things as intelligent or more intelligent than humans. Watch these videos to get an understanding of the issues of AI along with some basic concepts: Very good video: Recent advancements in AI to delve into further: AlphaGo Tesla Autopilot Swarm AI Speech recognition (including Google’s live language translating headphones) Leaders in the field of AI: Andrej Karpathy Ray Kurzweil Andrew Ng The most popular programming languages for AI are: Python — powerful, simple, uses Tensorflow Java — an all-around classic tool Lisp — a language specifically designed for AI C++ — good for directly communicating with hardware Cool projects by research institutes: OpenAI’s sentimental neuron OpenAI’s virtual reality learning robot arm DeepMind’s WaveNet model

Everyday Artificial Intelligence #1 — Intro to AI

165

everyday-artificial-intelligence-1-intro-to-ai-151645ed2dad

2018-03-04

2018-03-04 10:35:06

https://medium.com/s/story/everyday-artificial-intelligence-1-intro-to-ai-151645ed2dad

false

563

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Nicholas Vitebsky

Avid programmer learning React Native, Python and JavaScript focused on AI and mobile https://github.com/nv1327

411e7386f6cb

n.vitebsky1327

9

24

20,181,104

null

0

null

0

null

2018-06-17

2018-06-17 09:49:40

2018-06-15

2018-06-15 23:18:32

9

false

en

2018-06-17

2018-06-17 09:57:20

43

15171c619a14

9.10566

0

Article 3 of the Series on Ethics in Artificial Intelligence

5

How the era of artificial intelligence will transform society? Article 3 of the Series on Ethics in Artificial Intelligence In the previous article, we talked about the nature of human fears in relation to artificial intelligence (AI). We also outlined two important themes to address in the upcoming challenges of the AI era: the societal, and decision-making aspects of AI systems. In our new article I expand on the theme of the impact on society, and I share our opinion on possible steps we should take to anticipate and adjust to it. McKinsey analysts estimate the automation potential for all economic sectors to be around 50%. This means that around half of all the activities people in the world’s workforce are paid to do today could potentially be automated with currently available technologies. It represents almost $15 trillion in wages (about China’s GDP today). Historically every industrial revolution (Figure 1) has had its primary driving factor. For example, the adoption of electrical power defined the era of mass-production and led to the economies of scale, impacting both capital and labor. The huge societal challenges arriving with automation resemble the challenges of industrial revolutions. Figure 1 Timeline of the industrial revolutions. What lessons can we learn from the past? What are the current drivers and the economic consequences? What is the role of ethics in the ongoing changes? Industrial revolutions and the Engels’ pause When we talk about the technological impact on society, we can reference lessons from the past. We know, for example, that as a consequence of the 1stindustrial revolution, mechanization increased the productivity of each worker but real wages stagnated for approximately 50 years. Figure 2 Engels’ pause in real wages during productivity growth (MGI, 2017; Allen, 2008). In economic theory, this phenomenon is described as “Engels’ pause”. It explains that due to great technological developments, the lives of a large number of people worsened first before society began to prosper in the longer term (Allen, 2008). Finally, what could be the driver of growth for the new industries that are emerging? Let’s identify this resource that will enable us to create new types of jobs and transform businesses. Data is the “new electricity” Steam, Electricity, Computer & Internet… the drivers of previous industrial revolutions. Let’s now look at what would drive change in an era of AI. Along with algorithms and advanced computation facilities, the accuracy of the models constituting artificial intelligence rely heavily on the availability of real-world data. Data will fuel the development of AI and data is crucial. IDC estimates 10x growth of the available worldwide data by 2025 with almost 50x growth in analyzable data (up to 5.2 ZB). A $138 bn data market is predicted by 451 Research, this number surpasses the GDP of more than 135 countries (IMF). Figure 3 Sizing of the data market (IDC, 2017; 451 Research, 2017) Data is the driver of the new industrial era, it is here, growing, and ongoing transformational processes brought by data-driven applications pose new challenges to society. These challenges aren’t in the technical field of AI development but are rooted in our human nature. While 35% of the skills demanded for jobs across industries will change by 2020, at least 1 in 4 workers in OECD countries is already reporting a skills mismatch with regards to the skills demanded by their current jobs (WEF, 2017). The problem is not the absence of jobs, but the skills requirements of new jobs (MIT Sloan Management Review, 2017; WEF, 2018) that are already being created today (see AI Teacher). How will humanity surpass and adjust to the possible Engels’ pause that may result from adoption of AI and automation? Absorption of skills and the new industries Throughout history, the decline of some large-scale employment sectors has been countered by the growth of new sectors that absorbed workers. The recent example of the emergence of personal computers shows us that technology drives the creation of many more jobs than it destroys over time, mainly outside the industry itself (Figure 4). It is estimated that from 1970 to 2015 the personal computer industry destroyed almost 3M jobs in the USA while creating more than 19M jobs in other sectors. Such increase has resulted in fast economic growth in the services sector, in trade and in the creation of new industries such as the software industry. Figure 4 Birth of the knowledge economy (MGI, 2017; IPUMS, 2017). In the information era that started in the 1960s, the main driver of change was the dissemination of computers and, consequently, global access to the internet. This, in turn, has created a so-called knowledge economy. Products and services based on knowledge-intensive activities that contribute to an accelerated pace of technical and scientific advance, as well as rapid obsolescence. //Knowledge Economy (Powell & Snellman, 2004) In the knowledge economy, it became possible for us to deliver services remotely, outsource business processes, and to dematerialize the whole concept of economic transactions with internet-based payment systems. What skills will the workplaces of the future require? In previous articles we approached this question a few times, and we keep reflecting on it. Education, employability and the future of work Why do we talk about education in this series of articles about ethics? We do so because we are able to re-use lessons learned from history, re-think our future, and re-skill individuals to benefit the economy and guarantee their right to work. Everyone has the right to work, to free choice of employment, to just and favorable conditions of work and to protection against unemployment. //Article 23.1 of the Universal Declaration of Human Rights In their book, Milton and Rose Friedman said, “the essential part of economic freedom is freedom to use the resources we possess in accordance with our own values — freedom to enter any occupation, engage in any business enterprise, buy from and sell to anyone else, so long as we do so on a strictly voluntary basis“. What do we need to support such freedom? Technological advances may lead to income inequality with higher incomes for workers whose skills are complemented by technology, but not for those whose skills are substituted by it. Economists at the Bank of Canada suggest: “the greatest productivity benefits will occur in firms with high-quality people-management and decision-making processes and high levels of human capital”. Developing human capital requires education. In the past, one had to spend a lot of time digging in libraries with limited opportunities to access information or find a teacher. Today most answers are available in our smartphones and are accessible 24/7. Still, according to the statistics, people in search of employment are not spending sufficient amounts of time in re-education and re-skilling. Figure 5 Most common activities for many who don’t work (KPCB Internet Trends, 2018). Why is this so? Today’s formal education system is based on outdated industrial learning methods (Rolff, 1993; Banathy, 1993; Gray, 2009; Rose, 2012). Technology is moving fast, making many courses obsolete even before learners graduate. Let’s just think about this: teachers learn and prepare course materials, they conduct classes, design evaluation systems and help students to graduate. By the time all this happens, the information taught has become outdated. We have been, and still are, working today with an “information-push approach” where the teacher and the content are put in the center and where the learner and the problem are not. This approach has proved insufficient and should be reformed (OECD Report 2018). With the introduction of MOOCs, we address the availability of and access to knowledge with more ease. However, the engagement of a learner, who is overloaded with disparate pieces of information, remains a challenge. Let me recap on this: we know the change is coming with AI and automation, we know that this change will have implications for the workers, we cannot prevent change. Therefore, we need to help people to adjust to change with maximum benefit for them by enabling exponential thinking (Figure 6) so that humans and AI will be complimentary. Blaming machines for this issue is not a solution. Employability of human capital in the 21stcentury will require new sets of skills: resilience, critical thinking, social skills and the ability to learn, reflect, and quickly adapt to change. Figure 6 Adaptability and exponential thinking with the right mindset and tools (SU, 2018). Acquisition of such skills requires new learning methods and discussion about them worth a separate article, if not a series of research projects, conferences, and debates. Life-long learning requires creative destruction, where the word creative is the key. The word Creativity is rooted in creation, the creation of new meanings, new interactions, new products, and new markets. In the abyss of the great depression and hardship in 1933, the opening lines of Franklin Roosevelt’s inauguration speech were, “…the only thing we have to fear is fear itself.” Becoming comfortable with uncertainty is an essential component of any creative process, especially when creating a positive future. Team PocketConfidant believes that education is fundamental to prepare our society for the future. We need a model of education that helps each individual to learn faster and easier, based on personality, strengths, and interests. The new model of education should help anticipate the never-ending change in life while accounting for personal values, desires, and goals. This is how Team PocketConfidant started to work with the Education system in 2017, as a first step to developing new ways of supporting learning methods and academic innovation. How should we govern in the AI era to positively impact privacy, security and wealth distribution? As Peter Drucker once said, “the best way to predict the future is to create it”. I think that three of the following approaches could help in society’s regulation of the impact of AI. Industry-led assessment of the impact on each stakeholder; An inclusive, ongoing and open dialogue between experts and citizens; Governmental regulations based on the nature of each AI solution. Industry-led impact assessment It is very hard to estimate the impact of artificial intelligence on society in general. Yet, we could estimate, regulate and design the impact of artificial intelligence on an individual basis — industry by industry, technology by technology and role by role. I propose a very simple framework (Figure 7) to approach the impact evaluation: Figure 7 Impact evaluation on a role-by-role basis. I think that a basic framework such as this could be used by each industry consortium and community to begin the dialog and craft targeted approaches. Inclusive, ongoing and open dialogue between experts and citizens An inclusive dialogue between experts and citizens can lead to inclusive societies where humans and AI successfully work together. Public dialogue is useful when a topic is controversial or complex. As reported by RSA and suggested by Diane Beddoes, Director of Deliberate Thinking, there are advantages, limitations and challenges to building open dialogue with citizen juries: Figure 8 Advantages and limitations of the open dialogue between experts and citizens. Governmental and cross-border governance As in previous transitions between industrial eras, developing education, awareness and continuous learning will be vital. Distribution of income, is equally important. While yields in productivity will increase, there is no guarantee that these yields will be shared fairly. Governments have a responsibility to address issues of inequality. They can use tools such as taxation and transfers, however such use will involve difficult trade-offs related to preserving incentives to invest in technology. As pointed out by Carolyn Wilkins, senior deputy governor in Bank of Canada, increased market power for some AI players may raise important systemic issues, many of them being global in nature. In today’s globalized world, international regulations will also be challenging, given how easy it is to move intellectual property to unregulated jurisdictions. Therefore, countries and economic areas that create favorable conditions for artificial intelligence solutions that augment human capabilities, are expected to largely benefit from technology progress. For example, France’s AI mission for humanity aims to develop complementarity between humans and machines (Proposal 4.2) and to support AI-based social innovations (Proposal 7.3). Conclusions and suggested next steps In this article, through the prism of industrial revolutions, I covered some of the anticipated transformational changes artificial intelligence will bring. Upcoming transformations will be disruptive in nature and will change the future of work and the way we develop our human skills. At the same time, in PocketConfidant we are positive about the potential of transformative changes and we think that they can benefit society if anticipated well. We believe appropriate anticipation and preparation would require 1) industry-led assessments with identification of skills and learning approaches to future-proof the workforce, 2) Creation of an inclusive dialog between experts and citizens, and 3) Governmental and targeted cross-border governance for each concrete application of AI. In our next article I will discuss the governance and ethics of AI decision-making and will share our opinion on inclusive AI design. Originally published at pocketconfidant.com on June 15, 2018.

How the era of artificial intelligence will transform society?

0

how-the-era-of-artificial-intelligence-will-transform-society-15171c619a14

2018-06-18

2018-06-18 18:48:59

https://medium.com/s/story/how-the-era-of-artificial-intelligence-will-transform-society-15171c619a14

false

2,095

null

Technology

technology

Technology

166,125

Nikita Lukianets

Founder, CTO @PocketConfidant AI. I work in the R&D with interests ultimately related to the fields of Computational Neuroscience and Artificial Intelligence.

deaa66463d16

lukianets

240

236

20,181,104

null

0

null

0

588226e9514d

2017-10-18

2017-10-18 19:00:50

2017-10-27

2017-10-27 19:47:33

0

false

en

2017-10-27

2017-10-27 19:47:33

1

1517b3461c72

1.075472

0

Robo-advisors first took over the advisory industry — now robots are infiltrating the trading floor.

2

Make Way For The Robo-Trader Robo-advisors first took over the advisory industry — now robots are infiltrating the trading floor. It’s no surprise the financial services industry is undergoing a total transformation considering technologies like artificial intelligence, machine-learning and big data currently have or eventually will have a presence in every industry and furthermore in every job. Technology first started disrupting the financial services space when equities trading shifted to electronic platforms decades ago. Then robo-advisers began replacing human financial advisers. And now, robots— more specifically artificial intelligence — will disrupt the day job of a markets trader. Wall Street is entering a new era. The fraternity of bond jockeys, derivatives mavens and stock pickers who’ve long personified the industry are giving way to algorithms, and soon, artificial intelligence. A recent Bloomberg article discusses how jobs on the trading floor are already shifting. One job in the cash/trading side of business is fascinating in how AI will be integrated to fulfill client orders at the best market price and time. Eventually, this AI called reinforcement learning will automate hedging and market making. The brief description of this disruption leaves readers, well at least me, curious about how AI will maneuver complicated financials like hedge funds where trades and sell-offs are made in split seconds and the risk of a failed transaction — potentially due to inaccurate or premature AI — can cost a fund extreme amounts of cash. It remains to be seen how successful AI will be in streamlining sensitive financial trades that have been traditionally handled by humans. And if the industry will be better off with a robot than a brain. Read the full Bloomberg article here.

Make Way For The Robo-Trader

0

make-way-for-the-robo-trader-1517b3461c72

2018-01-31

2018-01-31 06:32:45

https://medium.com/s/story/make-way-for-the-robo-trader-1517b3461c72

false

285

Keeping Tabs on the Latest Trends Shaping Our World

null

dittopr

null

Ditto TrendComms

teamditto@dittopr.co

ditto-trend-comms

PUBLIC RELATIONS,COMMUNICATIONS STRATEGY,BUSINESS INTELLIGENCE,CURRENT EVENTS,INNOVATION

dittopa

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Kendall Clark

NYC PR

ec48cb55edb6

kennie_lane

22

96

20,181,104

null

0

null

0

null

2018-02-13

2018-02-13 13:10:15

2018-02-13

2018-02-13 14:41:44

3

false

en

2018-02-13

2018-02-13 17:19:09

3

1517dc252494

3.882075

4

0

Standee — check, Brochures — check, Marketing video — check, Dresses — check, camera- charged, sales pitch — prepared and feedback forms…

5

Lessons from setting up a sales booth for the first time Me at the Ntalents.ai booth - Drishti 2018 event at IIM Bangalore Standee — check, Brochures — check, Marketing video — check, Dresses — check, camera- charged, sales pitch — prepared and feedback forms printed. That was my idea of how sales booth at conferences work. But as any other sales plan — this one trashed too. But lets begin with the fun facts and save the trash-can for few paragraphs later :) We arrived as early as 7.30 AM for a 10 AM event — which was a first for us (we are engineers after-all) and given our abysmal record at reaching early anywhere it was a moment of pride! So you can guess how serious we were about making this work. We explored the whole place to pick the ‘most viewed’ spot based on meticulous calculations of ‘guest-eye-movements’. After all we are a serious data-backed ML powered startup and data is now nothing less than a religion :) Then we looked around for availability of ‘power points’ because we had planned to keep both out laptops on and keep running the promotional video. We had just one power point allotted so we high-jacked our neighbors power point too (after-all we came early). We had also called up a couple on interns to help us out with the stuff (We thought we deserve to feel like CXOs once in a while). But it turned out that with ‘morning’ the Gen-Z assumes 11 AM :-P. No hard feelings though, students have some serious stuff going on in their lives too. Anyways, this was actually good news for our CEO and CTO who started some of their favorite activities from college — setting up the booth and decorating it. I wish I had a photo from 10 years back when in college they used to have the same kind of fun in the name of being ‘Fest coordinators’. The tasks include — #Print posters; #never purchase cello-tape and look for it like ‘treasure-hunt’; #find spots to display posters (again using eye-motion-tracking :-P ); #look at other stalls and copy what they did; #read own standee content and praise own efforts… the list goes on :) Data scientists at serious start-up work After 2 hours of hard work, we were all set to meet the guests. And that is where the real story begins. There are 4 key lessons we had taken away and I am sure this must be common sense in the world of B2B sales. Lesson #1 : No one will ever read the brochure So no need to write paragraphs with 12 font size in there. And those flat graphic icons designed by professionals seem like colored dots anyway. All people look at is — brand name — pictures — cartoons and highlights. We had a co-founder level fight over what color scheme to use in icons and whether the ‘sequence’ of brochure content was right (Fact : Everyone turned the brochure back immediately after reading the headline) As good as we could design it — the brochure! Lesson #2 : No one stays put to watch your 1 minute whiteboard video That too with a North American accent voice over :) So it is better to have a micro-mini colorful cartooned video which just screams out what you do many times over. SELF PROMOTION ALERT : Watch www.ntalents.ai video here — Lesson #3: Most big-ticket leads will never come finding you out on the booth So go attend the conference first and then stalk the speakers. There is no way that they will visit the school-exhibition type stalls in the very short time that they have. And startup-philanthropy is the last thing on their minds on a Sunday evenings. And to say the least about the posters put on pillars in walkways — if they are not life-size standees then your black-white posters merge with washroom signboards anyway. And secondly the arrows that you proudly put up with the poster will not hypnotize them to arrive at the booth, even if you are a google. Lesson #4 : There are all kinds of people in the world Yea, this is kind of philosophical and way too general for this context, but yes all kinds of people come to you when you sit there for hours without work. There were people who agreed on whatever we said enthusiastically (and we thought we are so good at pitching) only to reveal later that they came to sell their stuff to us. :-| Then there were investors of competitors who read every line on the brochure and argued on how it will never work. Although it did help us get better at face-to-face rejection handling — what-doesn’t-kill-you-makes-you-stronger is applicable here too. Then there were ‘explorers’ who knew nothing about technology and expected us to explain everything like a professor. Not to forget few idea-exploring wannaprenuers who just wanted to walk off with the AI-ML studded brochure — as if that is what built startups. And finally to give us some relief our mentors came in the evening with that parental-affection in their eyes to appreciate us on random stuff :) Credits :) Mayank Sharma & Varun Narula

Lessons from setting up a sales booth for the first time

5

lessons-from-setting-up-a-sales-booth-for-the-first-time-1517dc252494

2018-05-29

2018-05-29 15:07:49

https://medium.com/s/story/lessons-from-setting-up-a-sales-booth-for-the-first-time-1517dc252494

false

883

null

Startup

startup

Startup

331,914

Deepika Anu

Co-founder@Ntalents.ai, IIM Bangalore

1635f8dc8c21

dasaxena

15

20,181,104

null

0

null

0

null

2018-06-02

2018-06-02 14:09:13

2018-09-22

2018-09-22 10:03:36

0

false

en

2018-09-22

2018-09-22 10:06:12

2

151b39dfc3d7

2.924528

8

0

…and how MLeap makes our data science and engineering teams happy and productive

5

Real-time serving machine learning models with MLeap …and how MLeap makes our data science and engineering teams happy and productive At Expedia Partner Solutions, we support the hotel business of the world’s leading airlines, travel agencies and consumer and loyalty brands. We help them innovate in the travel industry through a versatile API at the core of our offering. Like our partners, we like to grow and innovate. One example is our work to leverage machine learning capabilities across hotel sorting, anomaly detection, hotel recommendation and cross-sell, image ranking or chat bots. Machine learning tools proliferation Data scientists use a continually expanding range of machine learning technologies — Scikit-learn, Spark ML, Tensorflow to name a few — to solve different types of problems. MLeap makes our data science team happy because it makes it simple to deploy a wide range of ML applications and frameworks using its common serialization format and execution engine. Data scientists can leverage the right tools for the job at hand without worrying about unnecessary delays caused by the need to build additional model scoring services. MLeap makes our engineering team happy as it minimizes the effort to serve models within a production environment. To get the serving infrastructure right, we look at the maintainability, monitoring and scalability of a single model scoring service, removing the overhead of multiple model serving implementations. With so many machine learning tools available, clients (our shopping or recommendations APIs etc.) would need to manage the complexity of integrating with the right model serving implementation based on the model being used. But MLeap hides the differences of the underlying model training implementations from clients behind a unified scoring API. This makes changing between various models, perhaps by means of an A/B test, straightforward. Model representation to the rescue Traditionally, the transition between model training (data science team) and model scoring (engineering team) runs the risk of causing bottlenecks and inefficiencies as each team works with its own toolsets and according to its own workflows. Our data scientists typically work with Python and notebooks, while our software engineers use Java or Scala most commonly. MLeap makes our data science team happy because it eliminates any rewriting that needs to be done to serve models in production. It achieves this by using an intermediate representation of the trained machine learning models for serving. The output of the training stage is the predictive model/pipeline, stored as a bundle of Json or Protobuf files. This means that once a model has been built and trained, it is also ready to deploy in production. MLeap makes our engineering team happy since building and training models becomes decoupled from using them to make predictions in a production environment. Without having to re-implement the trained model/pipeline within the scoring environment, our data scientists can now launch new models in production with minimal engineering involvement. Real-time serving machine learning models MLeap makes our engineering team happy given the low latencies we’ve been able to achieve in our scoring service, even while scaling to high throughputs. At EPS, we’ve built a REST scoring service using the MLeap library, the serialized MLeap bundles and Spring Boot. Taking our hotel sorting use case (where our machine learning pipelines are relatively elaborate consisting of 50+ stages) as an example, the results we’ve come across in production have been very encouraging: a 99th percentile latency of 30ms to 70ms under considerable load (averaging at approximately 500 requests/second) when scoring on average 250 hotels within each individual scoring request. Feature engineering made easy When building the model, data scientists convert the raw data into features that better represent the underlying problem to the predictive models. This helps improve model accuracy at prediction time. MLeap makes our data science team happy because most gains come from good features and MLeap allows them to pull in as many features as needed and combine them in intuitive ways. MLeap makes our engineering team happy given that implementing our custom feature transformers with the serialization code required is a matter of just a few classes worth of code. We can follow similar engineering practices as with any other code and write tests to make sure we’ve eliminated any discrepancies between how we handle data in the training and serving pipelines. Because debugging differences between online scoring and training in production can be hard. With an increased confidence that the model in our training environment will give the same score as the model in our serving environment, MLeap makes both our data science and engineering teams happy. Check out MLeap documentation for more information or join the community on Gitter!

Real-time serving machine learning models with MLeap

9

real-time-serving-machine-learning-models-with-mleap-151b39dfc3d7

2018-09-22

2018-09-22 10:06:12

https://medium.com/s/story/real-time-serving-machine-learning-models-with-mleap-151b39dfc3d7

false

775

null

Machine Learning

machine-learning

Machine Learning

51,320

Anca Sarb

Senior software engineer at Expedia Partner Solutions, part of Expedia Group

298307b4b6f2

ancasarb

6

7

20,181,104

null

0

null

0

null

2018-05-17

2018-05-17 22:02:49

2018-05-17

2018-05-17 22:03:39

1

false

en

2018-05-18

2018-05-18 16:33:29

7

151bfeb1f67

3.849057

0

Your phone rings and you answer it. A woman with a perky, Mid-western voice asks if you’re happy with your current insurance provider.

5

The (Poor) Imitation Game: 3 Ways to Prove That You’re Better Than a Robot Your phone rings and you answer it. A woman with a perky, Mid-western voice asks if you’re happy with your current insurance provider. “Could this be fate?” you think to yourself. You excitedly describe your history of insurance troubles, your search for love, and your lifelong dream of meeting a woman from the Midwest with an interest in your insurance coverage needs. “I’m sorry, I didn’t quite catch that”, comes the polite reply. With your heart in your throat, you do your best to repeat yourself only to again hear the words “I’m sorry, I didn’t quite catch that”, in the exact same tone and cadence. With horror, you realize that you’ve just given your life story (and your heart) to a machine… Have you ever fallen victim to communicating with a robot unintentionally? In this article, BriteBee shares the history of artificial intelligence (A.I.) interactions, why insurance consumers prefer a personal relationship, and how A.I. can never replace the human advantage and intuition of the local insurance agent. The Imitation Game In the year 1950, computer scientist Alan Turing developed a litmus test for advanced artificial intelligence. The test was simple. One man (Player C) would sit alone in a room while exchanging text messages with players in two other rooms. In one room, another man (Player A) would try to hold a pleasant person-to-person conversation. In the other room, a computer (Player B) would run its best imitation of “real-person” conversation. If Player C can’t tell the difference between Player A and Player B, then Player B (smart robot) wins and Player A (the rest of humanity) loses. Here in 2018, Turing’s test has broken out of the laboratory and into our day-to-day lives. Whether we like it or not, we are all players in one big “imitation game”. Corporations and businesses struggle to genuinely connect with their audiences in a communication landscape saturated with chatbots and automated sales calls — and the insurance world is no exception. Workers walk a tightrope of rooting for the development of artificial intelligence in order to make their jobs easier and more efficient, while also fearing their jobs becoming entirely automated. As the march of technology steadily closes the gap between the capabilities of artificial intelligence and the needs of our society, questions about the exact benefits of being human will become more and more important. In most cases, it’ll take more than correctly translating squiggly letters into normal letters to show the unique advantages of our humanness. Luckily, we have (at least) three ways to prove that local insurance agents are better than robots. After all, we already naturally have the traits we’re trying to build machines to imitate, right? Be Flexible in Your Communications Improvise. Adapt. Overcome. You have the ability to do things you weren’t explicitly told to do. You have the ability to say things you weren’t explicitly told to say. You aren’t programmed — you think on your feet. The world of social and business interactions is complex. While scripts can be useful and backup plans are great, it takes a human to anticipate and overcome the unexpected. Lean into your creativity and think outside the box. As a human communicator, only you have unlimited options for improvising a message, adapting to a new audience’s context, and overcoming misunderstandings. Treat People Like You Would Want to Be Treated The Golden Rule. A higher imperative. A way to use our own selfishness for good. When it comes to successfully working and interacting with people, empathy is at the center of our human advantage. We have unlimited ideas on how to serve others because we have unlimited opinions about how we should be served. No matter how far technology advances, a robot will never be able to intuitively understand a person’s needs — so use that leg up to your advantage! In the complicated world of insurance, no consumer wants a scripted response. Instead, insurance agents should tap into their compassion and resourcefulness to find solutions to problems from a personal perspective a machine cannot provide. Remember That Technology is Built By Actual Living People For Actual Living People Chatbots and automated forms are great tools when they’re built (by a person) to save valuable time and cut down on tedious redundancy, but they are a terrible replacement for conversation when dealing with complex issues like insurance liability (or love). Humans can do incredible things with the help of apps, websites, and automated processes (developed by people). Businesses like Zillow, AirBnB, Uber, and Lyft all take advantage of technology as a compliment to human-to-human interaction, connecting people with a need to the people who can help. With this mindset comes another benefit of being human: the ability to steer and leverage technology for the good of ourselves and the good of our fellow non-robots. BriteBee Loves Real People BriteBee desires to connect real-life, trustworthy insurance agents with insurance consumers in need. We desire to restore, strengthen and bridge the gap between insurance shoppers and insurance professionals. If you’ve been stung while shopping for insurance online, take heart — we have too. In fact, the deceptive use of A.I. in the insurance realm is one of our driving forces to serve those in need of insurance coverage. With BriteBee, we want every insurance consumer to be in control of their information, not an insurance aggregate database. Our platform allows consumers to communicate with agents of their choice, when and how they prefer. This results in higher quality connections for insurance agents on the BriteBee platform. BriteBee is launching May 2018. For more information, visit us online at www.britebee.com. Originally published at agents.britebee.com.

The (Poor) Imitation Game: 3 Ways to Prove That You’re Better Than a Robot

0

the-poor-imitation-game-3-ways-to-prove-that-youre-better-than-a-robot-151bfeb1f67

2018-05-18

2018-05-18 16:33:30

https://medium.com/s/story/the-poor-imitation-game-3-ways-to-prove-that-youre-better-than-a-robot-151bfeb1f67

false

967

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

BriteBee

BriteBee allows you to shop and compare insurance quotes with local insurance agents without unwanted phone calls and emails.

90dfa0bf4d51

britebee

27

20,181,104

null

0

null

0

33b49288dc77

2018-08-29

2018-08-29 17:39:16

2018-08-31

2018-08-31 11:49:15

1

false

ru

2018-08-31

2018-08-31 11:49:15

9

151c5c36f19a

2.966038

5

0

Когда дело доходит до оценки потенциала проекта блокчейн и прогнозирования его будущих успехов, одной из попыток к продвижению является…

4

Обзор команды Matrix. Часть первая: Стив Денг Когда дело доходит до оценки потенциала проекта блокчейн и прогнозирования его будущих успехов, одной из попыток к продвижению является фундаментальный анализ. Это процесс оценки актива в попытке измерить его внутреннюю ценность, исследуя различные качественные и количественные факторы. Одним из наиболее важных факторов, на мой взгляд, является ведущая команда проекта, поскольку ее авторитет может показать нам, смогут ли люди, работающие над проектом, предоставить полностью функционирующий продукт и превратить наши инвестиции в прибыль. Так как цена со временем будет соответствовать успеху продукта, его можно проанализировать аналогично инвестированию в стартап. Основное внимание при изучении токена должно быть уделено команде и ее способности поставлять рабочий продукт, который фактически будет использоваться. В последующей серии статей я расскажу о нашей команде Matrix AI Network, которая работает над очень перспективной и инновационной платформой блокчейн, которая, я уверен, станет одним из ведущих проектов в криптомире. Первая часть серии посвящена главному ученому MATRIX по искусственному интеллекту Стиву Денгу Стив Денг Профессор Денг отвечает за разработку алгоритмов машинного обучения и структуры аппаратного обеспечения для блокчейн MATRIX. Вы можете до сих пор не знать, что MATRIX — это инновационная блокчейн-платформа, которая использует искусственный интеллект для создания интеллектуальных смарт-контрактов, которые не требуют программирования (да, искусственный интеллект помогает в этом), защищают эти смарт-контракты посредством использования самонастраивающегося искусственного интеллекта для проверки на наличие уязвимостей. Кроме того, искусственный интеллект в MATRIX используется для выполнения различных вычислений. Некоторые из существующих технологий MATRIX включают в себя секвенирование генома, автоматическое распознавание изображений искусственного интеллекта и помощь в диагностике рака в больницах. Выдающаяся команда даже строит свои собственные майнинговые чипсеты для помощи искусственному интеллекту и различным статистическим вычислениям. Наконец, блокчейн будет использоваться для решения проблем реального мира. Проще говоря, делает MATRIX искусственным интеллектом суперкомпьютера. Это не только сделает MATRIX одним из самых быстрых блокчейнов, но и гарантирует современную архитектуру, как никто другой. Для передачи новому поколению этих полностью работоспособных функций требуется талантливая команда, которой руководит Стив Денг. Ниже мы рассмотрим профиль главного ученого MATRIX по искусственному интеллекту. Мы постараемся доказать заслуженное доверие человека, который находится у руля искусственного интеллекта в MATRIX. Профессор Стив Денг является ассоциативным профессором Школы программного обеспечения университета Цинхуа, где он был преподавателем с 2008 года. Стоит отметить, что университет Цинхуа является крупным исследовательским университетом в Пекине, Китай. Этот университет окончили многие китайские лидеры в области политики, бизнеса, науки и культуры. Университет Цинхуа занимает одну из лидирующих позиций в списке ведущих научных институтов Китая, Азии и во всем мире. Также он был признан лучшей инженерной и компьютерной научной школой в мире. Стив Денг закончил свое ME и BE в Университете Цинхуа. Исследовательские интересы Стива Денга включают в себя компьютерное обучение, аналитику отраслевых данных и компьютерную архитектуру. Профессор Денг с 2016 года был заместителем директора по архитектуре прогностического управления здравоохранением для высокоскоростного железнодорожного проекта компании Rollingstock китайской железной дороги, в котором основное внимание уделяется использованию современных методов машинного обучения в целях радикального преобразования работы и обслуживания железнодорожных транспортных средств. Кроме того, он является автором или соавтором более 50 научных работ, а также своей книги, которая была принята университетом Цинхуа и многими другими университетами в качестве учебника по проектированию ИС. Работа Денга по распознаванию изображений на основе глубокого изучения заняла первое место на многих престижных конкурсах (PASCAL VOC и COCO). Он получил множество наград, в том числе награду за лучшую статью на Международной конференции по компьютерному дизайну 2013 года, премию профессора в области партнерства NVIDIA и премию Tsinghua Key Talent. Удивительно, но у Стива Денга есть профиль на официальном веб-сайте NVIDIA, который вы можете посетить https://www.nvidia.com/object/professor_partners_bios_yangdong.html NVIDIA заявляет, что во время исследовательской работы над докторской диссертации Денг разработал эффективный алгоритм для оптимизации сети межсетевых соединений для однокристальной системы. Работа открыла новое направление исследований для оптимизации взаимосвязи системы с учетом прогнозируемой информации макета. Также там вы можете найти информацию о том, как Стив Денг разрабатывал передовые инструменты физического дизайна. На мой взгляд, после того, как вы ознакомились со всей этой информацией из уважаемых источников, никто не будет сомневаться в репутации Стива Денга или в том, сможет ли MATRIX реализовать обещанную технологию. Основываясь на исследованиях и работе Денга в этой области, ясно, что процесс внедрения искусственного интеллекта или разработки микросхем системы искусственного интеллекта находится в хороших руках, и такая команда мирового класса станет ключом к успеху Matrix AI Network, становясь новым титаном блокчейна. Для получения более актуальной информации присоединяйтесь к нам в социальных сетях: Website | Telegram | TelegramRU | Twitter | Reddit | Facebook | White Paper | White Paper RU

Обзор команды Matrix. Часть первая: Стив Денг

225

обзор-команды-matrix-часть-первая-стив-денг-151c5c36f19a

2018-08-31

2018-08-31 11:49:16

https://medium.com/s/story/обзор-команды-matrix-часть-первая-стив-денг-151c5c36f19a

false

733

Русский блог MATRIX AI NETWORK

null

MATRIX AI NETWORK-RUSSIAN

null

matrix-ai-network-russian

BLOCKCHAIN,CRYPTOCURRENCY

Matrixrussia1

Matrix

matrix

Matrix

651

Pukis3891

Matrix AI Network Russian community manager, miner, crypto world enthusiast.

c725b0704b46

pukis3891

52

6

20,181,104

null

0

null

0

f702855ffe47

2017-10-28

2017-10-28 10:00:20

2017-10-28

2017-10-28 10:00:21

3

false

en

2017-10-28

2017-10-28 10:00:21

7

151d19a9898c

0.980189

0

null

3

Driverless Autonomous vehicle industry Blog # medium.com Originally published at www.driverless.global. 20 Answers to Clarify Your Understanding of Artificial Intelligence # medium.com https://www.huffingtonpost.com/barbara-jacoby/asking-questions-is-really-hard_b_7052722.html Over the last m… AWS Deploys New EC2 Instances For Machine Learning, AI Customers # medium.com Next-gen EC2 instances designed for compute-intensive applications that require massive parallel floating po… Tokyo Motor Show to highlight ‘green’ vehicles and AI in cars # medium.com Nissan Motor Corp’s new electric vehicle LEAF at global headquarters in Yokohama, Kanagawa Prefecture, south… 4 new things to read in AI # medium.com China’s Social Credit System Will be Contagious # medium.com If China leads as a Big Data consumer experimen…

5 new things to read in AI

0

5-new-things-to-read-in-ai-151d19a9898c

2017-10-28

2017-10-28 10:00:23

https://medium.com/s/story/5-new-things-to-read-in-ai-151d19a9898c

false

114

AI Developments around and worlds

null

AI Hawk

aihawk1089@gmail.com

ai-hawk

DEEP LEARNING,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING

null

Deep Learning

deep-learning

Deep Learning

12,189

AI Hawk

null

a9a7e4d2b403

aihawk1089

15

6

20,181,104

null

0

null

0

null

2017-11-15

2017-11-15 21:59:51

2017-10-30

2017-10-30 17:32:13

1

false

en

2017-11-15

2017-11-15 22:00:36

3

151f4082764f

3.377358

0

An interview with Amy Inlow who is the CMO of Albert. Albert is the first-ever artificial intelligence marketing platform for the…

5

Albert, the AI marketing platform for enterprise, works 24/7, freeing up time for creative strategy An interview with Amy Inlow who is the CMO of Albert. Albert is the first-ever artificial intelligence marketing platform for the enterprise, driving fully autonomous digital marketing campaigns for some of the world’s leading brands. You shared that “AI technology has the ability to transform the way we market today, and address many of the execution and analytics oriented issues that exist within our field.” How is Albert transforming the way we market? What are the issues that Albert’s solving? With the continuous introduction of new channels and consumers’ rising expectations for personalization, reaching and engaging relevant users has become nearly impossible. The advent of martech tools was meant to be an answer to this problem but ended up creating an even bigger headache for the marketer. We’re now swimming in both data and tools to break down that data–not to mention siloed, inefficient, efforts to understand, engage and convert each and every customer on each and every channel. Albert is the answer to this. As a true AI and fully autonomous digital marketer, Albert has the ability to manage and optimize data at a speed and rate beyond human capability. And he does this from a holistic cross-channel standpoint, meaning that he unites all digital efforts so that the insights and learnings he gains on one channel are shared across all channels. As a result, true attribution is reflected in the campaign’s focus at any given moment. Albert works 24/7 and handles all of the gathering, processing and analyzing of data, in order to free up the marketer to do what humans do best: creative strategy. Asaf Jacobi, President of Harley-Davidson of New York City, shared that they’re building a call center to accommodate the leads Albert is generating for them. Is this kind of ROI common? Harley-Davidson NYC is much like other brick and mortar retailers in that they use their online properties and channels not only to sell, but also to drive customers in-store. This allows them to cater to different consumer buying patterns, and focus their physical presence to specific products and buyer motivations. In the case of Harley-Davidson, they know that their customers buy their bikes in the store rather than online, so they used Albert as part of a lead generation program, aimed at gathering customers’ interest and info–or sending them to a call center where they could learn more. On our end, we ensure that there is an attribution measurement in place to gauge the effectiveness of Albert across historically detached online/offline ROI goals. We are seeing similar offline use cases and ROI across other large-ticket items, such as vehicles and furniture that require further inquiry or in-person trials, etc. Being able to measure the online to offline success is crucial. We are also seeing this among several of our retail/ecommerce clients, who find that they must adjust their on-the-floor team or “call centers” in order to take on the new volume of offline leads produced by Albert’s online efforts. Courtney Connell, Marketing Director at Cosabella and an Albert customer, shared that “it’s important to distinguish between those that just offer insights and those that actually execute.“ Can you give a bit more insight into this comment? While there are systems that can come up with automatic insights or strategies, only a truly autonomous AI system, such as Albert, can act on those insights without human supervision. Albert automatically implements changes based on what he observes, and continues to adjust and optimize based on changing user behavior patterns. He additionally offers high-level suggestions–that marketers can apply across their marketing programs–based on analysis of the company’s data and customer/target market (including learned and predicted behaviors). For companies who are still a bit iffy about using AI in their marketing efforts, what advice do you have for them? The key here is to understand that AI takes on the time consuming data management/analysis work in order to free up the marketer to focus on high-level strategy. The marketer remains in control and the AI makes decisions, which remain within the guidelines and ROI targets defined by the marketer. Additionally, when a company first decides to make the shift, it’s important that they find a vendor who offers transparency into the AI platform. Clear measurable goals should be defined in the scope of the campaign, and the AI should ease digital efforts, rather than add to the marketing headache. Which AI solutions do you use for work and in your personal life? This is unending. From my smartphone, to Netflix and Amazon shopping, I happily go with the AI flow where my convenience and ease is improved, I have frequent conversations with Alexa. At work, anywhere there is a possibility to automate a process, we will implement–whether it be email campaigns or social media scheduling. We even use Albert for Albert promotional campaigns! Originally published at rozee.co on October 30, 2017.

Albert, the AI marketing platform for enterprise, works 24/7, freeing up time for creative strategy

0

albert-the-ai-marketing-platform-for-enterprise-works-24-7-freeing-up-time-for-creative-strategy-151f4082764f

2017-11-15

2017-11-15 22:00:37

https://medium.com/s/story/albert-the-ai-marketing-platform-for-enterprise-works-24-7-freeing-up-time-for-creative-strategy-151f4082764f

false

842

null

Ecommerce

ecommerce

Ecommerce

46,740

Rozee

Content + Community for all things interesting in #Enterprise #AI! Get the Daily Rozee in your inbox, Mon-Fri: http://eepurl.com/c9sahD.

83c3dd12d9c6

heyrozee

0

1

20,181,104

null

0

null

0

86c895bd18e0

2018-08-03

2018-08-03 09:48:26

2018-08-22

2018-08-22 11:42:10

25

false

en

2018-09-19

2018-09-19 08:05:21

5

151fe2a45502

13.709434

12

1

0

Why data science matters and how it powers business values.

5

Unlocking Business value with Data science The advent of data science, machine learning, and artificial intelligence has gained visibility across platforms such as conferences and social media. Nonetheless, a large proportion of startup founders face challenges when identifying key areas where businesses would benefit from using data science. 80% of a data scientist’s time is devoted solely to finding, filtering, and organising data, leaving only 20% to actually perform data analysis. Data collection and storage are crucial requisites for the success of any project in the domain of data science. Why does data science matter and how does it power business values? To answer this, Accel organised the event ‘Unlocking Business Value with Data Science,’ where founders and data leaders from the Accel family came together with data product experts, who expounded the benefits of data. Guest speakers included: Swaminathan Padmanabhan — Director of Data Science at Freshworks Ambarish Kenghe — Leading Tez Product Management at Google, Former Chief Product Officer at Myntra Sharath Bulusu — Senior Vice President of Product Management at Myntra Through this blog post we will be discussing the following key points of using data science at Startups. The concrete benefits that Accel portfolio companies perceive in using data science today. Data collection and data storage as key pillars for the resolution of data science problems. Data quality and volume are crucial; “garbage-in-garbage-out” applies across all data science projects. The benefits of simple data science techniques such as exploratory data analysis, regression, and decision trees. Almost 70% to 80% of business problems can be resolved using simple techniques if the required data is available for processing. 90% of the data science cases explored during this session were resolved using straightforward methods. In order to prepare for the data-driven world, Accel recommends that start-ups: Learn to collect data in the correct format (minor lapses could result in a delay of 6 months) Effectively organise data for analysis (termed ‘data pipelines’) Cultivate a curiosity to explore The event commenced with sharing data-driven case studies related to Accel portfolio companies. Of the 120 portfolio companies, at least 30 companies are using data to derive business insights and to add value to their businesses. These insights help build data products in the fields of fraud prevention, personalised customer experiences, increased sales, streamlined operations, and developed customer engagement. The insights also empower management to take better decisions using data. While working with Accel portfolio companies, it was observed that almost 80% of business problems were solved with basic data science techniques (such as regression, binary classification, tree based models, etc.).For solving key business problems you don’t really need to jump to neural network. You might rather identify what problems can be solved by analysing data you have collected over the period of running business operations. “Data science use cases at each step of funnel “across Accel portfolio companies To understand customer behaviour and their engagement with the platform ,data science can be applied at each step of Funnel. list of few companies using data science at each step of Funnel Study of data science use cases across Accel portfolio companies. Note: We will take only few use cases to convey the message “benefits of applying data science in existing business problems “ in this blog post.Post your question in the comment section to know more about specific use case . CoverFox-Lead scoring: Since we are able to prioritise the leads, most effective leads will be attended first along with 360 degree view of customer to our internal agent.This will help us convert more leads into prospect customers.Sending these leads to most suitable agent will help increase customer satisfaction and minimising ticket time. Also, building lead scoring engines can greatly benefit companies in similar domains. UrbanClap- Matching Customer Needs to Professionals Initially, merely key attributes such as location, professional rating, and the category of the service platform were used to match professionals to customer services, but this system left professionals and customers unsatisfied. Thus, tree-based models were developed that accounted for approximately 45 features such as the time of day, work experience, and other factors. This resulted in a marked improvement in satisfaction for both customers and professionals. ChargeBee- Optimal retry reschedule : ChargeBee is a recurring subscription billing SAAS business. For ChargeBee, 5% of all renewal transactions were failing. Thus, the ChargeBee team took to data analysis in order to better discern the reasons behind this decline and if they could identify any apparent trends. Some errors indicated that the payment gateway configuration needed to be optimised and that requirements for CVV or AVS may not be configured for recurring payments. B2B merchants tend to see errors relating to “SERV NOT ALLOWED” when customers make payments using corporate cards with restrictions on recurring transactions. Likewise, B2C merchants may see this error more frequently when customers make payments with prepaid debit or gift cards. After establishing the decision-tree-based model, they were able to minimise the rate of declined transactions. In turn, this had a direct impact on revenue and reduced churn for customers. UrbanClap — Churn detection: Retaining trained professionals is a challenge for any business. For certain categories, such as plumbers, electricians, and the like, professionals are trained by the company itself. Thus, there is high benefit in evading turnover to have access to their continued services. Team has built a basic binary classifier to help predict churn-out within 30 days. Moreover, patterns of churning out are also sought, such as fewer tickets assigned to the individual, no incentives, and so on.This is helping us to minimise professional churn and increasing customer satisfaction(since services are provided by more experienced professionals) all the problems mentioned above were solved using basic exploratory data analysis ,regression, clustering and tree based models except Crownit -invoice reader → We tried with support vector machine but accuracy was not good enough. Then We built the CNN based model which improved the system accuracy by 10%. As we can observe here almost all the business problems were solved with basic data science techniques if you have enough and structured data in place. Accel encourages startups to build a strong foundation of data before venturing into the territory of data science, artificial intelligence, or machine learning. We found Data science “HIERARCHY OF NEEDS” digram very useful.What we will see in media everyday about AI, deep learning which is relatively hard to apply especially if you are business company driven by technology and you don’t have in-house data experts . NEEDS TRIANGLE[DATA SCIENCE] Making AI works for your business [Case studies from Freshworks] Swami addressing audience about Freshworks’s data journey ML is core of the Freshworks Products. Each product line is driven by Data . ML state [Freshworks product] Freshworks The audience was presented with two important cases wherein data science was used to improve business processes and increase revenue. 1. Incident (ticket deflection in customer support) : Recently, customers have begun to expect DIY forms of customer service. This is because these are often the fastest and lowest effort ways to resolve problems. There is no need for an external agent when the issue is relatively simple. For companies, this is beneficial because increasing self-services lead to improved ticket deflection. This is when customers choose to help themselves rather than reach out for support. This allows support teams to focus on more complicated issues. Moreover, since the issue is resolved instantly, customer satisfaction is high. Simple search engine-based approaches do not succeed here due to lack of domain understanding and contextual information. For instance: Q1. Will you give me my money back if I don’t get a service call by Thursday? The expected answer is a refund policy, but the search engine cannot handle this. Q2. How much do you charge for a 3-year old? The expected answer is pricing for children, but a search engine cannot be trained to answer these problems. There exists a definite need for data science in the form of NLP or machine learning tools. Freshworks uses NLP and machine learning to automate high-frequency, low-touch customer interactions and bypass the effort required for customers to discover content. Whenever the platform is not able to handle support requests, it is passed on to an appropriate agent. This results in great cost-cutting for Freshworks. A feasibility analysis for this issue was conducted to decipher if enough data was present for analysis. We asked the following questions: How many customers have a knowledge base? (Most do) What percentage of queries can be answered using the customer’s existing knowledge base? Per model for each customer? (Industry level models) Apart from knowledge base, what else can be learnt from tickets, previous chats, forums, and the like? How do we incentivise customers to grow their knowledge base? (This is a very critical matrix because when a more expansive knowledge base is added into the platform, the model becomes more accurate.) After evaluating such data verticals, enough data was found to apply data science to this use case. An ML platform was built for incident deflection and assisted resolution, which provided customers with the following services through this platform: Chatbots Email answerbots Similar incidents compose assist social bots Collaborator assist Results: Matrices for ticket deflection use case 2. Predictive deal scoring : The Goal: We will want to prioritise which accounts we reach out to. Assigning a score for all inbound leads based on how likely they are to convert (as a customer is the essence of lead scoring) is predictive deal scoring. The below problems were worked on using this technique: Sales teams typically have a large number of open leads in their pipeline In conventional CRM, there is no predictability on which deals would close and by when There is no tools available to help a sales agent determine how to accelerate deal closure A feasibility analysis using existing data was performed to check whether a deal is predictable or not. “A good predictive model needs good quality data in large amounts!” by Swami Data analysis-1 Data analysis-2 From the above diagrams, one can discern that the lead scoring engine could surely be built. The regression model was used to achieve this. Features set lead scoring There may occur situations where the data is not sufficient enough to build deal scoring systems. For example, when new customers and industry level cases are involved. In such scenarios, fallback logic is employed wherein models are trained at an account level. There were cases where CRM data was incomplete on account of low fill rates due to erroneous filling by the sales team. In such cases, other attributes such as revenue, web traffic, and industry, among others, are used to predict lead score. Results: Predictive lead scoring enabled the sales team to focus on the best prospective leads and optimise conversions Beta experiments indicated that sales teams could isolate around X% of the conversion by focusing solely only on Y% to Z% of their total outstanding leads. Where X,Y,Z are some percentage of impact. Data science what is it good for[Case studies from Myntra] AK and Sharath sharing Myntra data story with Audience “Data is gold, You need a solid foundation for your data before being effective with AI and machine learning ” Ambarish Identifying Fashion trends: Fashion Trending system Myntra Results: Revenue impact per impression CTR went up Personalisation: Myntra emphasises the value they place on data in order to tailor recommendations, engage influencers. and customise experiences. Consumers are willing to share data if it provides a more personalised experience online. Fashion trends today are fleeting, which is why discovery is very important for better engagement and sales revenue. Initially you don’t need to build complicated system, start personalising, from where user left in last session something like recently viewed items once user re- login to platform. And then build collaborating filtering , this approach really worked out for Myntra. Iterating fast and running experimentation was the key for this system. Personalisation appears easy on the surface, but for a company with Myntra’s scale, it does become challenging. The process of personalising from thousands of brands, millions of products with different sizes, and the failure to predict the correct sizes for personalised items, leads to frustrated customers. One way to build size recommendations is to study the customer’s purchased and returned items. Data on customer returns enable the building of a system to recommend sizes. Data collection across multiple levels is therefore critical. Another scenario is, “What if I am ordering this purchase for someone else?” This is where data strategy enters. One could enquire from customers, “Who did you order this for?” And use this data to further personalisation. Who did you buy for? Data collection to build profile Myntra has run various experiments to design data collection strategies for different sets of customers. It benefits to reward customers in order to accrue this data. Results: CTR increased for personalised banners compared to equivalent no personalised banners X% of increase in CTR Personalised sort order for ~ Y million users Search: In the beginning, general sentiments of search were not optimal. There were several areas where customers were not provided with meaningful results. Information retrieval systems such as Elastic search failed to comprehend queries such as “casual shoes under 400.” “why, many queries where we have not provided intended result or provided no result”. Data from the following verticals were used to answer this: query log Feedback system from customer care A Myntra-wide bug bash to collect feedback Problems from this experiment were clustered into three areas: Failure to find intent or context of the query “casual shoes under 400” should be understood as the customer searching for casual shoes with a limit of Rs. 400. “Moto 360” must be understood as “Smart wearable ” and “Fog” must be understood as “deodorant” in order to suggest customer-appropriate products. Similar results if product not found Spell correction did not seem to correct common scenarios like “jins” to “jeans.” No query substitutions were available. In the case of no results a probable substitution was not offered. It is evident from the above results that the problem relates to either precision or recall and since both cannot be solved simultaneously, the choice was made to improve precision. For this, the click-through rate was researched. The bounce rate, click-depth, and zero results were found to be most important. Based on this research, the below pipeline was built: data pipeline to improve search(Semantic search) This resulted in fruitful engagement and an increase in revenue. Cross sell: Expert tell is what to cross sell data science collaborative filtering approach increased the CTR by X% . Myntra’s Intelligent fast Fashion -> Rapid Platform [example of AI] relatively hard science problem. That’s Myntra Fast Fashion for you — fashion via high-tech engineering. This platform is the perfect example of using the already collected data and insights to create new revenue stream and value for customers. Myntra Fast Fashion entails fashion via high-tech engineering. The production processes deliver the latest trends in the market, which usually span 6 months. These were reduced to under 30 days. “In the initial days, there was less machine and more designer input. With rapid platform, we utilised more machine input and less designer supervision.” — Ambarish High level overview of rapid platform Myntra uses social and various other data sources to sense demand and global trends. The sales data across Jabong, Myntra, Flipkart were included to help develop a machine-generated design. Results: Two brands launched using Rapid platform generated huge demand among customers. The Rapid Platform is an example of intensive data science and a more sophisticated AI. Takeaways: Make data easily available to data scientist Build for experimentation Cross function team Beware of spurious models and correlations A Q&A with our experts: Which should be hired first at a startup: a data engineer or a data scientist? Ambarish answers: Definitely a data engineer, making the data scientist your first data hire is the common mistake startups make. Unless you already have a solid data infrastructure and internal business intelligence (BI) practice, you’ll need a data engineer to build pipelines and help data scientists prepare data to prevent boredom and turnover. If you hire a data scientist first, they won’t have any data to play around with. We did this when we hired our first data scientist. He work for 6 months to bring the data into the appropriate format and only then started solving data science problems. Hiring a data engineer reduces the scope of work for the data scientist because data prep steps can be handled by data engineers. Get an experienced practitioner for your first data hire; this guy will be able to move quickly with minimal assistance, which means you will see faster returns on your data science investment. How do you develop a data road map, that is, a data strategy? Jeet answers: Data can be used to drive decisions and build products that increase profits, reduce costs, reduce risks, engage customers, boost operations, and generate insights. Develop a set of questions you’d want to answer, connect “what we want to do” with “how will we do it.” Participate in the survey here to understand where your startup stands in terms of a data roadmap. Conclusion, key takeaways: Define your first data project in detail: - What questions are you trying to answer? - Will these improve the business matrices? Map problems to existing data. If you have enough data, start with simple data science techniques to solve immediate important cases and then delve into more advanced techniques like neural nets. Data matters even to startups. If you are not processing data for insights during your early days, archive the raw data in some form of storage such as S3. Focus first on Low hanging fruit — Least adopted “must do” use cases, for example for ecommerce industry Forecasting, Tracking consumer behaviour , reducing revenue churn. For insurance industry lead scoring, cross selling etc. All these problems can be solved using simple technique like regression, multi-class classification , tree based models etc. The domain of data science is experimental. Invest and build a culture of experimentation and fast iteration for the data science team. Work in small iterations based on results and learnings. Craft a clear vision for what the company wants to achieve with data science, explore basic data science ,high value use cases and simple data science technologies to implement them first. Launch Proof of concepts for selected use cases and run them in staging environment before pushing them in production. Set up a cross-functional team for each data science use case; horizontal teams produce better results for data science projects. Hire experienced data engineers before hiring data scientists. Look for experience in the data scientists you hire. Entry level candidates may be slower on account of having to learn processes. Experienced practitioners will be able to move faster and manage teams. For future blogs on data science please subscribe to Accel India or follow here.

Driving Business values with Data science

48

driving-business-values-with-data-science-151fe2a45502

2018-09-19

2018-09-19 08:05:21

https://medium.com/s/story/driving-business-values-with-data-science-151fe2a45502

false

3,103

Accel India Insights follows the stories, lessons and journeys of our community of entrepreneurs. We back and support people and their companies from the earliest days through all phases of private company growth.

null

Accel India Insights

araghava@accel.com

accel-india-insights

INDIAN STARTUP ECOSYSTEM,CONSUMER INTERNET,SAAS,ECOMMERCE INDIA,HEALTHCARE INDUSTRY

accel_india

Data Science

data-science

Data Science

33,617

jeetendra gangele

Passionate about new ideas

806d0b8ec1d0

gangele397

35

258

20,181,104

null

0

null

0

null

2018-04-27

2018-04-27 09:37:06

2018-04-27

2018-04-27 09:42:27

0

false

en

2018-04-27

2018-04-27 09:42:27

0

1520c6ca64e0

3.264151

2

0

Team mashme.io were delighted to have attended EXPOELEARNING XVII and International Congress in March, marking a week of celebration in…

5

The ‘intelligent’ future of e-Learning Team mashme.io were delighted to have attended EXPOELEARNING XVII and International Congress in March, marking a week of celebration in education. The main topic of this edition was the application of chatbots and, consequently, the incorporation of Artificial Intelligence in the online learning process. Can a chatbot replace a human? Does it mean the disappearance of the teacher as we know it? Our CEO, Víctor Sánchez, was invited to address the Congress and help deepen understanding and the impact on the future of education. He answers these and other questions in this interview: Can you expand on the fundamental lines of your paper to the Congress of EXPOELEARNING XVII? The fundamental line of the paper is to approach the concepts of bots, Machine Learning, Reinforcement Learning and TinCanAPI, as well as its practical use in reality to the congress.. How do you see the evolution and future of training in relation to the growing use of artificial intelligence and robotics? I think the incorporation of artificial intelligence and robotics into the e-Learning industry will drive adaptation of teaching methodology for the each individual student. In addition, this will allow the adaption of content, not in terms of reduction, but in terms of strengthening the most complex areas. The methodology and the content not only represent a technological advance, but also translates into an increase in the student’s motivation and, therefore, in the efficiency of the own training experience, and the impact it will subsequently have on his/her career path. How is the chatbot applied to e-Learning used today? We are experiencing the birth of the incorporation of chatbots in the e-Learning sector. In my opinion, I think this is just the beginning, and there is still a lot to explore and perfect. It is possible that within ten years, e-Learning can not be understood without chatbots! At present there is reluctance and fear of change, since it is easy to think that artificial intelligence can replace the teacher. Nothing could be further from the truth; the teacher’s role will be fundamental, but he will not have to worry so much about knowledge, but become the guide leading the student to learning. A robotic can teach knowledge, but the teacher will be in charge of teaching a student how to learn. What are the great changes and novelties that we are going to see in the world of e-Learning in the next months? In general, the education sector is much more conservative than other areas where technology hits in a more immediate way. There is a great responsibility in teaching within society, so the agents involved in the process take much longer to introduce new mechanics or take risks. However, it is much easier to adopt new measures when there is a quantitatively demonstrable certainty that it is synonymous with success. The main change is the management of Big Data to something on a more emotional level for the student. We are going to stop seeing statistics of time on the platforms, attendance at classes or the amount of resources shared, to seeing statistics capable of assessing the actual satisfaction of the student, the level of understanding of each resource and the degree of attention in each moment of the experience. Do you think that in the future we will be trained with robots? What role will there be for the being human as a trainer? We are already doing it now. Every day we learn something through the Internet: knowledge is in the network, robots are only a physical interface to those artificial intelligences that grow and evolve in the cloud. There is also a self-taught trend not only among the younger sectors of the population. I like to compare it with a book. Can you learn only by reading? The answer is clearly affirmative, if the teacher’s role were to only give us books and dates of the examination to evaluate knowledge. The human being as a trainer will not disappear; Their role will evolve and will be absolutely essential in the training process for students How do you see Spain in the world of e-Learning? In Spain, in general, it is difficult for us to introduce new technologies. We have, a fairly conservative spirit and we tend not to be pioneers in the implementation of new methodologies until their success has been demonstrated beyond our borders. However, it seems that in the last decade this is changing and that the new generations are ready for innovation. There are great ideas in our market and it is a pity that they end up being a success outside of Spain first. I have a great confidence in teachers wanting to adapt and make changes — we are in a wonderful moment. Are you optimistic about the advancement of technology applied to training? Will being human always lead this change and progress? I think that technology still has a lot to say in the education sector, to the point that will transform it completely until it is unrecognizable. This advance, nevertheless, will always have the human being behind it. Thank you very much.

The ‘intelligent’ future of e-Learning

3

the-intelligent-future-of-e-learning-1520c6ca64e0

2018-05-23

2018-05-23 15:14:37

https://medium.com/s/story/the-intelligent-future-of-e-learning-1520c6ca64e0

false

865

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

mashme.io

Discover seamless videocollaboration!

2f652ac35a37

mashmeio

146

224

20,181,104

null

0

null

0

null

2018-02-20

2018-02-20 14:46:08

2018-02-20

2018-02-20 14:58:51

4

false

en

2018-03-07

2018-03-07 00:40:04

3

1521109d9225

3.062264

2

0

This post is a follow up from the February 15th Lunch & Learn hosted by Mile 22 Associates: Reliable Metrics: The Value and Disadvantages…

5

What does it matter, if it isn’t the truth? This post is a follow up from the February 15th Lunch & Learn hosted by Mile 22 Associates: Reliable Metrics: The Value and Disadvantages of Randomized Controlled Trials Joanna Woodson, Associate, Mile 22 Eean Logan and Sam Novey of Mile 22 Associates on National Voter Registration Day Break out your old research methods notes, we’re going in. If you work in the social sector, you most likely have heard about RCTs (Randomized Controlled Trials). Currently, RCTs are venerated as the most rigorous method available to measure outcomes of programs and initiatives — and they’re pretty straightforward (if it’s been a few years since you’ve thought about an RCT, this video will jog your memory). What it comes down to, though, is that funders and implementers want to know whether a program works, and RCTs are an instrument used to prove that these programs are indeed working. Unfortunately, this funding is often contingent upon the program working, rather than knowing the truth about whether the program works. You can see how this has the potential to be problematic. That’s the essence of the February 15th Lunch & Learn hosted by Mile 22 Associates and presented by Dan Connolly of ideas42. We are asking these questions as a way to dig deeper within our organizations to find the truth; because, if we aren’t accomplishing the bright-eyed goals we’re putting billions of dollars toward what’s the point? To make ourselves feel better? Before we dig deep, it’s important to establish explicitly that we recognize the necessity of rigorous testing. We all have stories which attest to the complicated nature of life — sometimes great ideas simply don’t pan out as intended. For that reason, we need rigorous testing. We need testing to generate buy-in, and we need to understand where our resources should be invested. Lesson being: Rigorous testing is important. When it comes to RCTs, there are three major issues: generalizability, statistical significance, and data apophenia. Apophenia — seeing faces in inanimate objects, or in this example, seeing examples of programmatic success in data — is human nature. Consider the images below. These objects were not created cute, your mental gymnastics made them that way. Generalizability is an equally concerning issue — that these controlled trials may not represent a larger population, nor might they represent the reality in which the program actually exists. Lastly, statistical significance — relatable to implicit bias in research — has been argued to be problematic in this field. Researchers are able to extrapolate results which Andrew Gelman notes in his 2018 work, “ Selection on statistical significance leads to overestimates of treatment effects, this bias can be huge, and it can lead to a cascade of errors in the literature when exaggerated estimates in the literature are used in the design of overly optimistic future experiments.” This isn’t an article to disparage RCTs as a method of research — all methodologies are subject to implicit biases, and other forms of invalidation; and this isn’t an article claiming that we have all the answers, because we most certainly do not. The point is, we need to get the program implementers and the funders in the same room, at the same conference. We need to develop donor-implementer relationships rooted in trust, in order to test the efficacy, to find the truth, of these programs. Trust and truth will get us all where we want to go. Chris McCandless, the famous wanderer, left a phrase which I think offers much guidance for us today: Rather than love, money, faith, fame, or fairness, give me truth. To continue this dialogue, please contact Joanna at joanna@mile22.org. We want to spark a continuing conversation that leads to better and more effective testing — and we’d love to hear your ideas!

What does it matter, if it isn’t the truth?

51

what-does-it-matter-if-it-isnt-the-truth-1521109d9225

2018-03-07

2018-03-07 00:40:05

https://medium.com/s/story/what-does-it-matter-if-it-isnt-the-truth-1521109d9225

false

626

null

Civic Engagement

civic-engagement

Civic Engagement

2,553

Joanna Woodson

Joanna lives and works in Washington, D.C., though she left a slice of her heart in rural NC. She is a social worker by training, and sleuth by nature.

1449ee4fe91b

joanna_28412

4

3

20,181,104

null

0

null

0

d439090f99d

2018-08-03

2018-08-03 11:34:26

2018-07-26

2018-07-26 00:05:19

1

false

en

2018-08-06

2018-08-06 12:51:01

13

15212c50d09b

5.750943

0

With the rise of podcasts, there’s a growing trend where podcast platforms or podcast apps, are leveraging the use of natural language (NL)…

5

Why Smart Speakers & In-Audio search need a diverse range of digital voices With the rise of podcasts, there’s a growing trend where podcast platforms or podcast apps, are leveraging the use of natural language (NL) processing to help users find and listen to shows. This is similar to smart speakers that use voice activated Artificial Intelligence (AI) technology such as Alexa, Google Home and Apple Homepod. With any new trend, where there are opportunities, there are also issues. The issue here is with the increase use of natural language and artificial intelligence presents a diversity and inclusion problem. This doesn’t only affect Caribbean and people of color but all “non-traditional” or “broadcast English” speakers. Let’s start with the basics: What is artificial intelligence? “Artificial intelligence (AI) makes it possible for machines to learn from experience, adjust to new inputs and perform human-like tasks. Most AI examples that you hear about today — from chess-playing computers to self-driving cars — rely heavily on deep learning and natural language processing. Using these technologies, computers can be trained to accomplish specific tasks by processing large amounts of data and recognizing patterns in the data.” (Definition by SAS) What is natural language? “Natural language processing (NLP) is a branch of artificial intelligence that helps computers understand, interpret and manipulate human language.” (Definition by SAS) Simply put — AI and NLP learns from what you input, it gets smarter/improves with more human interaction and it repeats the learning cycle. What is SEO? “SEO is the acronym for Search Engine Optimization. It’s the practice of optimizing websites to make them reach a high position in Google’s — or another search engine’s — search results. SEO focuses on rankings in the organic (non-paid) search results.” (Definition by Yoast) The content on a website is one way to reach high in the search results of Google or any other search engine. Why is this important? SEO. “Voice SEO” to be exact. The use of Artificial Intelligence (AI) and Natural Language Processing (NLP) trends in podcasts and voice activated gadgets like smart speakers, is about searching. There’s already a problem with the discoverability of podcasts by people of color without the natural language processing and AI aspects, and this just makes it even more difficult. And if the tech doesn’t understand you when you’re speaking then they are lost in translation. Podcasts & Natural Language Processing A few months ago there was an article about Castbox a podcast app that raised 13.5 million to launch its own programming. The article mentioned how Castbox’s use of natural language was going to revolutionize podcasting. “What makes Castbox interesting is the proprietary technology it has under the hood. The platform uses natural language processing and machine learning techniques to power some of its unique features, like personalized recommendations and in-audio search. The app is capable of making suggestions of what to listen to next based on users’ prior listening behavior, which can help to improve discovery of podcasts people may like. Meanwhile, the in-audio search feature takes advantage of the recent leaps the industry has seen with voice recognition technology, and actually transcribes the audio content inside podcasts, indexes it and makes it available for search within the Castbox app. That means users no longer have to rely on things like episode titles, descriptions and show notes to find a podcast related to a topic they want to listen to — they can just search the Castbox app for any podcasts where a term was mentioned.” ~ Source: Techcrunch Both Carry On Friends The Caribbean American Podcast and The Style & Vibes Podcast are on Castbox. On it’s face all these features are great, however my primary concern with podcast platforms or apps that provide natural language processing via in-audio voice search is it will benefit only or mostly popular shows backed by network providers, and those shows that are in standard English. In my legal industry career I’ve been familiar with natural language processing and the limitations that were causing my concerns. However, being that I’m no longer in that industry perhaps things have improved, so I decided to do some experimenting. There’s this Castbox demo video on Youtube showing how the in-audio voice search feature that leverages NLP works. As in the video, a search was done on “how to get through the present”. The results were similar and were categorized in channels (aka shows), episodes and audio. The audio it appears, is the natural language going through the audio of each show and pinpoints the exact timestamp when one or more of the terms in the search is found in particular episodes. A similar search was done using Carry On Friends’ content and didn’t have the same results. The episode used in my demo wasn’t even in patois or heavy in it. I spoke in my natural voice. My simple experiment confirmed my concerns that with natural language processing there are limitations with standard english much less for those with accents. Smart Speakers In June, I did a fire side chat during CITE week on the digital voice and had a follow up lunch with a friend where we discussed the topic further. As it turns out the Washington Post and it’s vast resources where already exploring my concerns. “At first, all accents are new and strange to voice-activated AI, including the accent some Americans think is no accent at all — the predominantly white, nonimmigrant, non regional dialect of TV newscasters, which linguists call “broadcast English”. The AI is taught to comprehend different accents, though, by processing data from lots and lots of voices, learning their patterns and forming clear bonds between phrases, words and sounds. To learn different ways of speaking, the AI needs a diverse range of voices — and experts say it’s not getting them because too many of the people training, testing and working with the systems all sound the same. That means accents that are less common or prestigious end up more likely to be misunderstood, met with silence or the dreaded, “Sorry, I didn’t get that.” …for people with accents — even the regional lilts, dialects and drawls native to various parts of the United States — the artificially intelligent speakers can seem very different: inattentive, unresponsive, even isolating. For many across the country, the wave of the future has a bias problem, and it’s leaving them behind.” ~ Source: The Washington Post Simply put people with accents or communicate in a way their audience appreciates and loves are being left out of smart speaker revolution and if more podcast platforms are using natural language processing to make shows searchable for the audience, shows produced by Breadfruit Media are already on the losing end. Opportunities The same way people want themselves represented visually in images, videos, print or in text the same is applicable with the digital voice. It’s not only an issue with podcasts and voice activated speakers. Don’t believe me? Have you tried to use mic button on your smart phone to dictate and send a message? Forget using patois (patwa) to send a message and sometimes it doesn’t catch my proper english correctly! I see this as a potential opportunity in the rise of the digital voice. Many people get into podcasting to be the talent/host however I think there needs to be diversity in those producing the technology being used in the space. I intentionally use a transcription service that understands patois. I would outsource production if it weren’t for me wanting to control the quality. However, let’s say that I was ready to outsource — I can’t find a production company that I think would be able to handle and understand the nuances of my patois and code switching enough to know when and how to make the edits. Bottomline Podcast platforms and apps continue to offer ways to improve the discoverability of a show to reach new audiences. However some of these new efforts aren’t beneficial to all podcasters. Yes, we should leverage what we can, like Castbox’s commenting feature which allows for more engagement with the audience. However, a content creator shouldn’t have to decide between their show being searched and discovered using smart speaks or voice searchable podcast platforms over the authenticity in language that their audience loves about their shows. The case can be made that more strides can be made when there’s diversification in the people developing and testing the technology as the Washington Post article pointed out. I think there’s opportunities for minority tech entrepreneurs to get in the space to solve some of the problems content creators are having. Or existing technology companies engage diverse voices to improve the AI offerings. Originally published at www.breadfruit.media on July 26, 2018.

Why Smart Speakers & In-Audio search need a diverse range of digital voices

0

why-smart-speakers-in-audio-search-need-a-diverse-range-of-digital-voices-15212c50d09b

2018-08-06

2018-08-06 12:51:02

https://medium.com/s/story/why-smart-speakers-in-audio-search-need-a-diverse-range-of-digital-voices-15212c50d09b

false

1,471

Breadfruit Media is a podcasting company that provides strategy, show development and production of stories by Caribbean Americans.

null

Breadfruit Media

info@breadfruit.media

breadfruit-media

BREADFRUIT MEDIA,CARIBBEAN PODCAST,PODCASTING,CARIBBEAN

breadfruitmedia

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Kerry-Ann Reid-Brown

Founder, Carry On Friends The Caribbean American Podcast & Breadfruit Media.

59e0f8b39ae0

carryonkerry

126

34

20,181,104

null

0

null

0

null

2018-08-25

2018-08-25 06:30:09

2018-08-25

2018-08-25 06:30:14

0

false

en

2018-08-25

2018-08-25 06:30:14

1

1521b2cbdf04

1.34717

0

[PDF BOOK] Introduction to Linear Regression Analysis Full Ebook By Douglas C. Montgomery Link…

1

DOWNLOAD PDF eBook Free Introduction to Linear Regression Analysis By Douglas C. Montgomery BOOK ONLINE #pdf [PDF BOOK] Introduction to Linear Regression Analysis Full Ebook By Douglas C. Montgomery Link https://bestebooks.ibookstore.us/?q=Introduction+to+Linear+Regression+Analysis . . . . . . . . . . . . . . . . . . . Read Online PDF Introduction to Linear Regression Analysis, Download PDF Introduction to Linear Regression Analysis, Download Full PDF Introduction to Linear Regression Analysis, Download PDF and EPUB Introduction to Linear Regression Analysis, Read PDF ePub Mobi Introduction to Linear Regression Analysis, Reading PDF Introduction to Linear Regression Analysis, Read Book PDF Introduction to Linear Regression Analysis, Read online Introduction to Linear Regression Analysis, Download Introduction to Linear Regression Analysis Douglas C. Montgomery pdf, Download Douglas C. Montgomery epub Introduction to Linear Regression Analysis, Read pdf Douglas C. Montgomery Introduction to Linear Regression Analysis, Download Douglas C. Montgomery ebook Introduction to Linear Regression Analysis, Read pdf Introduction to Linear Regression Analysis, Introduction to Linear Regression Analysis Online Download Best Book Online Introduction to Linear Regression Analysis, Read Online Introduction to Linear Regression Analysis Book, Read Online Introduction to Linear Regression Analysis E-Books, Read Introduction to Linear Regression Analysis Online, Read Best Book Introduction to Linear Regression Analysis Online, Read Introduction to Linear Regression Analysis Books Online Download Introduction to Linear Regression Analysis Full Collection, Download Introduction to Linear Regression Analysis Book, Read Introduction to Linear Regression Analysis Ebook Introduction to Linear Regression Analysis PDF Read online, Introduction to Linear Regression Analysis pdf Download online, Introduction to Linear Regression Analysis Read, Download Introduction to Linear Regression Analysis Full PDF, Read Introduction to Linear Regression Analysis PDF Online, Read Introduction to Linear Regression Analysis Books Online, Read Introduction to Linear Regression Analysis Full Popular PDF, PDF Introduction to Linear Regression Analysis Read Book PDF Introduction to Linear Regression Analysis, Read online PDF Introduction to Linear Regression Analysis, Download Best Book Introduction to Linear Regression Analysis, Read PDF Introduction to Linear Regression Analysis Collection, Read PDF Introduction to Linear Regression Analysis Full Online, Read Best Book Online Introduction to Linear Regression Analysis, Download Introduction to Linear Regression Analysis PDF files

DOWNLOAD PDF eBook Free Introduction to Linear Regression Analysis By Douglas C.

0

download-pdf-ebook-free-introduction-to-linear-regression-analysis-by-douglas-c-1521b2cbdf04

2018-08-25

2018-08-25 06:30:15

https://medium.com/s/story/download-pdf-ebook-free-introduction-to-linear-regression-analysis-by-douglas-c-1521b2cbdf04

false

357

null

Machine Learning

machine-learning

Machine Learning

51,320

Stewart Hawkins

null

210431e2c08e

stewarthawkins_69709

0

1

20,181,104

null

0

#Read csv file and parse the dates drdf = pd.read_csv('dog_rates_tweets.csv', parse_dates=[1]) #match n/10 rating from column ratings, set expand to false to get a series ratreg = drdf['text'].str.extract(r'(\d+(\.\d+)?)/10', expand=False) ratings = pd.to_numeric(ratreg[0]) #Assign new colum rating and drop all NAN rows drdf['rating'] = ratings drdf = drdf.dropna() #Remove rating outliers > 25, write back to dataframe drdf = drdf[drdf['rating'] <= 25.0] # Get a slope and intercept for a best fit line. fit = stats.linregress(drdf['timestamp'], drdf['rating']) # Scatterplot of date vs rating plt.plot(drdf['created_at'].values, drdf['rating'].values, 'b.', alpha=0.5) # Plot the best-fit line using the slope and intercept plt.plot(drdf['created_at'], drdf['timestamp']*fit.slope + fit.intercept, 'r-', linewidth=3) plt.savefig('dog-rates-result.png') plt.show()

10

null

2017-12-02

2017-12-02 02:01:07

2017-12-02

2017-12-02 05:49:50

5

true

en

2018-03-29

2018-03-29 01:23:28

4

1522183903ad

2.761635

1

0

A brief analysis of cuteness inflation — if you are reading this from a pdf please click on the username above for a better experience.

3

SUCH WOW DOGE — Image source google Explaining Pup Inflation from @dog_rates twitter — A statistical analysis A brief analysis of cuteness inflation — if you are reading this from a pdf please click on the username above for a better experience. In this article I’ll explore and analyze the twitter account @dog_rates. The question we are asking , “has there been grade inflation on the @dog_rates Twitter?” Apparently dogs over the years, specifically “The Americas,” have gotten better over the years. (David H. Montgomery) There will be multiple steps to answering this question, each of which we will briefly touch on. Data — getting the data Tweet dumper, is a script that allows to download all of a user’s tweets into a csv. The file was loaded using the pandas module in python. Figure 1: Plot of the raw data after extraction After we got the data we had some constraints to follow to proceed with analysis. Find tweets that contain an “n/10” rating Extract the numeric rating Exclude tweets that don’t contain a rating Clean & prepare the data As mentioned above we had to clean the data, remove outliers as well as cut down any null entries that don’t contain an “n/10” rating. As you can see the plot above with the outliers we would skewed results. As these outliers, are orders of magnitude much larger than the obvious linear relationship. Thus, removing those was important for close to accurate results with median effect on the rest of the data. Figure 2: Removed outliers, nil values, applied an interval to the data Analyze the data Now that we have the data, we know will give use meaningful results, we can apply different techniques to analyze such data. The obvious choice to understand how these data points are related is to run it through linear regression, to do a best-fit line. That will allow us to visually conclude how the ratings over time have been inflated. Figure 3: Linear pup inflation in red As we can see from the plot above, making sure to also keep attention to the points close to (0,0); we have a linear increasing relation indicating a positive inflation. If we had not done the appropriate cleaning on the data our plot would be completely useless. Figure 4: Linear pup inflation without cleaning the data Obviously plotting your data is very important when trying to understand as well as visualize your data. We could dedicate a whole section on plotting but there are many opinions and tools to explore, here is a good place to explore. Understanding your data is better than all the algorithms in the world. Gregory Baker

Explaining Pup Inflation from @dog_rates twitter — A statistical analysis

1

explaining-pup-inflation-from-dog-rates-twitter-a-statistical-analysis-1522183903ad

2018-03-29

2018-03-29 01:23:28

https://medium.com/s/story/explaining-pup-inflation-from-dog-rates-twitter-a-statistical-analysis-1522183903ad

false

511

null

Data Science

data-science

Data Science

33,617

Botch

null

852066656d1a

CryptoBotch

2

43

20,181,104

null

0

null

0

e6331c6f4711

2018-09-19

2018-09-19 11:44:52

2018-09-19

2018-09-19 11:47:04

3

false

en

2018-09-19

2018-09-19 11:50:40

3

15223f8b8ded

1.421698

0

A study says that 2.5 quintillion tons of data is being generated every day. It would be good if there is some knowledge of processing…

2

Big Data University Programs A study says that 2.5 quintillion tons of data is being generated every day. It would be good if there is some knowledge of processing these data. There are many university that provide many programs related to Big data. Most of the top universities in the world today,offer many programs in big data. They offer Master Programs like Big Data Analytics, administrator etc. Even in India there are lot of universities that provide PG program in Big data. For e.g. BITS PILANI give PG program in Big Data Engineering. There are many institutions providing weekend course for working professionals. There are lots of online programs too available for freshers, students and working professionals. BigDataCourseChennai is one of the institute that offers various program in Big data and in Data science. It offers programs in Big data who are interested in handling big data and for those who are interested in data science.It offers wide range of course from data science to Deep Learning. The students can pick the course they want. They have very good trainers to guide them to become experts in Big Data and data science. They can be connected through any of the following modes. The trainers here will comprehend about the available courses and guide you through your career. The contact details are given below call us @044–4232 1313 enquiry through mail: info@coremindtechnologies.com

Big Data University Programs

0

big-data-university-programs-15223f8b8ded

2018-09-19

2018-09-19 11:50:40

https://medium.com/s/story/big-data-university-programs-15223f8b8ded

false

231

BigDataCoursesChennai is a booming organization, that provides Hadoop big data training in Chennai. Our programs are designed as the best big data certification programs to help you develop skills that will make you a full-fledged IT professional.

null

BigDataCoursesChennai

info@coremindtechnologies.com

bigdatacourseschennai

BIG DATA,BIG DATA ANALYTICS,DATA SCIENCE,MACHINE LEARNING,DEEP LEARNING

BigdataChennai

Data Science

data-science

Data Science

33,617

Bhuvana Durai

null

42ea98c029a8

dbhuvanaa

3

4

20,181,104

null

0

null

0

null

2018-01-16

2018-01-16 15:14:46

2018-01-16

2018-01-16 15:16:30

0

false

en

2018-01-16

2018-01-16 15:16:30

2

15250460949c

5.690566

1

0

There are many indications today where customers, clients and many others believe that fraudulent billing occurs in all industries across…

5

Top Ways To Identify Fraud and Unusual Patterns Through Data Analytics There are many indications today where customers, clients and many others believe that fraudulent billing occurs in all industries across our nation and the world for that matter. There is much to be considered in the review of company methods that can determine fraudulent behavior, from data analytics to advanced analytics that will expose all different sorts of activity that tend to take advantage of customers and clients on a financial level. Fundamental Data Analytics Techniques include the ability to determine and find the following: Duplicate transactions Uneven amounts Frauds caused by Ratio Discrepancies Frauds caused by trends While we have some fundamental data analytics techniques that can be used to evaluate the billing systems of any company, there are the different, in-depth motions that can be taken throughout data analytics while also providing those more advanced analytics results. With some of the basic information on data analytics there are six basic steps taken to cover all of the information of financial data that can be fraudulently affected. With the ability for fraud to be uncovered here are six potential steps for data analysis: 1. Data Pre-Processing There are included techniques for detection, validation, error correction, and filling up of missing or incorrect data. It is great to see that companies can help rely on proactive and automated data analytics to help clean up the battle against fraudulent financial activity all around. There is always the value of making sure that records are updated to the clearest numbers possible, ensuring that there is nothing fraudulent sent out to the customer and client, as well as nothing fraudulent received inside the files of the company. 2. Calculation of Various Statistical Parameters Another useful approach being used for fraud prevention and detection involves the calculation of patterns and parameters in the actual data. The rationale is that unexpected patterns can be symptoms of possible fraud. Some of these statistics are averages, quantiles, performance metrics, probability distributions, and so on. For example, the averages may include average length of call, average number of calls per month and average delays in bill payment. 3. Models and Probability Distributions These models and probabilities can be of various business activities, either in terms of various parameters or probability distributions. Then, within data analysis there is benchmarking used to compare a company’s performance to competitors, industrial standards, historical data and budgeted data. Also, in relation to the fundamental data analytics techniques there are duplicates testing, expressions and equations that can be examined within any company’s numbers. There is the ability for reporting with charts and Excel files for all comparisons that can help to break down all accounting numbers. 4. Computing User Profiles This also includes matching algorithms to detect anomalies in the behavior of transactions or users as compared to previously known models and profiles. Techniques are also needed to eliminate false alarms, estimate risks, and predict future of current transactions or users. In addition to overall data analysis, there is the advanced analysis available to work with applied filters of financial records. There are applications such as user-defined criteria, advanced filtering, meta-tagging and more filters that are included with reports that are concluded. 5. Time-Series Analysis of Time-Dependent Data There is the ability to compare data analyzed throughout points in time, compared to data that occurs free of historical data patterns. With a company found to help with data analytics and advanced analytics, there are a number of methods available to help with the efficiency of accounts receivable, accounts payable, customer payments, vendor payments and many more. One interesting point found is the evaluation of Frequently Used Values. In this situation there is always the importance of identifying both frequently used values as well as those which appear unexpectedly in accounting reports. Data analytics can help red flag fictitious transactions included in those numbers. In addition, this would help identify transactions that occur during non-business hours or employee vacations. 6. Clustering and Classification This helps to find patterns and associations among groups of data. Another set of data analytics that help with the recovery of pure numbers include the Fuzzy Logic and Gaps that can occur in financial reporting. While fuzzy logic may be anything from unclear equivalency in addresses or other information, there are also gaps in the numbers reported whether it be the actual financial amounts, invoice numbers, purchase order numbers or many more. 7. Ratio Analysis Another useful fraud detection technique is the calculation of ratios for key numeric fields. Like financial ratios that give indications of the relative health of a company, data analysis ratios point to possible symptoms of fraud. For example, auditors concerned about prices paid for a product, could calculate the ratio. Three commonly employed ratios are: the ratio of the highest value to the lowest value (Maximum/Minimum) the ratio of the highest value to the next highest (Maximum/2nd Highest); and the ratio of the current year to the previous year. A large ratio indicates that the Maximum value is significantly larger than the second highest value. Auditors and fraud investigators would be interested in these unusual transactions as they represent a deviation from the norm. Unexplained deviations could be symptoms of fraud. In a number of cases, high ratios have identified payments incorrectly made to the vendor. 8. Trend Analysis Analysis of trends across years, or across departments, divisions, etc. can be very useful in detecting possible frauds. Another useful calculation is the ratio of the current year to the previous year. A high ratio indicates a significant change in the totals. This step or method could be partnered with some graphing techniques to provide a visual representation of the data and can highlight patterns or anomalies that might indicate areas of further examination for future purposes. 9. Duplicate In-depth Analysis Usually, one would assume that account number — vendor number combinations, would be unique. Therefore, the existence of dealings with the same invoice number — vendor number combinations would be an unexpected pattern in the data. The documentation of possible duplicate transactions would be a possible indication of fraud that should be examined. However, fraud indications are only that — indications — and care should be taken to properly examine the transactions before jumping to inferences. Transactions that look like duplicates may simply be progress payments or equal billing of monthly charges. It is possible to search for duplicates on one or more key fields. 10. Traditional Filtering and Pivoting Method Traditional companies still use this method wherein it only identifies those records meeting user-defined criteria. This method is used to extract transactions outside of expected norm. However, despite of the limitations mentioned above, it can further filter or analyze results using additional analysis techniques. Also, traditional methods maximize the use of pivot tables to have an interactive data summarization tool that sorts, counts, total, or gives the average data. This has really been useful to show the “big picture” of all the data gathered. 11. Digital Advanced Analytics Digital Advanced analytics is an advanced application of data analysis, is a new tool for auditors and fraud investigators interested in preventing and detecting fraud. In fact, digital analysis is a case where millions of transactions make the identification of fraud symptoms easier to find then when there are only a few thousand transactions. The patterns in the data become more obvious and focus attention on the fraud. The good thing with digital and Advanced analytics is that they already have automated systems that would make the analysis processes of your enterprise or company a lot more efficient and faster. There are a lot of Advanced Analytics online and you just need to choose the best company that would best fit your data analysis needs. Conclusions At the end of each analysis, the most important thing to consider is the summarization of all the data you’ve gathered. With all of these methods and steps toward fraud detection, and keeping all of your company’s numbers clear, there are still a lot of things to consider including the fact that all of these are moving into the IT-advanced methodology. There are a number of benefits to working with the technological methods of companies who can help provide the automated continuous methods of data analytics for the benefit of clean records a mentioned above. There are also a number of benefits that can be received from fraud detection including reduced exposure to fraudulent activities, reduced costs associated with fraud, determine vulnerable employees at risk to fraud, organizational control, improves results of the organization, and gains the trust and confidence of the shareholders of the organization. Author’s Biography: Brian O’Donnell is the Vice President — Marketing and Business Strategist of Attunix. He is the Co-founder of Attunix, a Microsoft Gold certified service provider. A results-driven executive responsible for bringing Attunix capabilities to market, strategic planning, and new product & service offering development. Originally published at www.backlinkfy.com.

Top Ways To Identify Fraud and Unusual Patterns Through Data Analytics

8

top-ways-to-identify-fraud-and-unusual-patterns-through-data-analytics-15250460949c

2018-01-17

2018-01-17 01:11:13

https://medium.com/s/story/top-ways-to-identify-fraud-and-unusual-patterns-through-data-analytics-15250460949c

false

1,508

null

Data Science

data-science

Data Science

33,617

Backlinkfy

#Digital #Marketing #Services for Entrepreneurs, Startups and Small Business. Find digital marketing tips and software resources to scale your business.

4edada38d6f4

backlinkfy

203

472

20,181,104

null

0

null

0

22d6b8c4a224

2017-10-13

2017-10-13 14:52:35

2017-10-13

2017-10-13 14:56:30

0

false

ja

2017-10-13

2017-10-13 14:56:30

1

1525f7b70d0e

0.666

0

PandasのDataFrameなどを、軸を入れ替えながら色々なグラフで表現できるFacetsというツールがあった。

5

Google製のビジュアライゼーションツール Facetsが便利 PandasのDataFrameなどを、軸を入れ替えながら色々なグラフで表現できるFacetsというツールがあった。 GoogleがFacetsをリリース Googleが、機械学習サイエンティスト向けデータビジュアライゼーションツールFacetsをオープンソース化した。 Facetsの目的は、巨大なデータセットを理解し解釈できるようにすることだ。Facetsは、開発者が巨大なデータセットに…www.infoq.com もちろん特徴量は自分で作る必要があるが、matplotlibを使うよりも簡単に色々なグラフを見ることができる。そのため、データの感覚を持ちやすい。今後も分析コンペでは使うことになりそうだ。

Google製のビジュアライゼーションツール Facetsが便利

0

google製のビジュアライゼーションツール-facetsが便利-1525f7b70d0e

2017-10-13

2017-10-13 14:56:31

https://medium.com/s/story/google製のビジュアライゼーションツール-facetsが便利-1525f7b70d0e

false

8

Composition,Programming,Data Analysis and Tech

null

Music and Technology

null

music-and-technology

ブログ,プログラミング,日本語,音楽,テクノロジー

_1__1_

日本語

18,705

takkii

機械学習、データ分析、漫画、ボードゲーム、音楽、競技プログラミングなどなど

b268a42a7979

takkii

164

301

20,181,104

null

0

null

0

null

2018-08-04

2018-08-04 00:07:37

2018-08-04

2018-08-04 00:12:41

3

false

en

2018-08-04

2018-08-04 00:20:49

8

152646b7c244

2.999057

7

0

There was a time when people used to call restaurants to get food delivered at their doorstep. Then, food delivering apps came into the…

5

Food Ordering Tech Trends in 2018 Photo by Brooke Lark on Unsplash There was a time when people used to call restaurants to get food delivered at their doorstep. Then, food delivering apps came into the picture. Just open the app, select the food from your favorite restaurant, place an order and you’re done! Such apps let you order food items from many restaurants in your area. You can check the prices, ratings and food items of different restaurants and order food within minutes. It’s really convenient. But, there’s one thing missing. It’s personalization. Personalization is going to be big in the coming days. Today, we have Google Assistant, Alexa and Siri to fulfill some of our basic needs and make us feel special. Now imagine a voice assistant who understands your taste and preferences and help order food accordingly. Yes, it’s going to happen. According to the National Restaurant Association, around 32% of restaurant operators accept that online ordering technology can help make restaurants more productive, increase sales and provide a competitive benefit. However, half of them are still thinking about the benefits of this technology and how it works. To compete in today’s market, many operators use technology to explain business processes and satisfy customers in a better way. Photo by Kai Pilger on Unsplash As per the recent orderTalk survey, online food ordering has become an American way of life, as nearly two-thirds of Americans already have a habit of ordering food online via an app or software. And, 45% of Americans believe that they will increase their use of online ordering within the next 12 months. The thing is we need to focus on the two technologies (mobile apps and voice ordering). They are going to be the future of online food ordering. Let’s take a deep dive into these technologies- Mobile apps and mobile payment The fast increase in a mobile phone usage in our lives has resulted in the growth of new apps or software. While ordering food over a smartphone has become conventional for many customers, handling the bill on a mobile device is still an innovation in most places. Only 1 out of 10 users say they paid for a meal using phone. Voice ordering technology Photo by Andres Urena on Unsplash Many restaurants and fast-casual diners are spending massively on digital technology for food ordering and delivery. According to a research by Capgemini, around 56% of users are interested in ordering meals from restaurants using voice applications and 34% have already ordered a meal using a voice assistant. Like Dominos Pizza, there are several companies using Amazon’s Alexa and other voice assistants to provide better customer experience. Voice-based artificial intelligence solutions involve with your consumers while delivering your orders, order estimates, offer various food options and upsell a particular meal of the day while adding the order in the POS directly. With an improved amount of digital ordering develops a stock of data around customers’ spending, choices and ordering behaviors. Restaurants can presently use this data to increase their analytics and also build algorithms that can divine order quantity and menu item list needs. The need for an actual and effective food ordering experience continues to evolve rapidly. Good news is that restaurant industry has many innovative solutions to examine to bring new changes in food ordering. No matter what type and size of restaurant you operate, you should be aware of these latest technology trends affecting the food and drink industry. Start looking to implement these new tech trends that are an innovative way of food ordering. About Voix Voix is first voice assistant for food ordering. Voix converts existing phone line into an AI-powered voice assistant to allow restaurants to take order, book table or offer specials to the customer, in turn, saving $1000s in employee salaries, training, and other human errors. With a team of serial entrepreneurs and technologists, Voix is on the mission to disrupt $72b+ phone food ordering industry.

Food Ordering Tech Trends in 2018

107

online-food-ordering-tech-trends-in-2018-152646b7c244

2018-08-04

2018-08-04 00:20:49

https://medium.com/s/story/online-food-ordering-tech-trends-in-2018-152646b7c244

false

649

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Voix.ai

Voice Assistant for Restaurants

e0d4acb49800

voixai

14

1

20,181,104

null

0

null

0

null

2017-09-03

2017-09-03 21:04:19

2017-09-05

2017-09-05 17:34:37

0

false

en

2017-09-09

2017-09-09 15:50:29

16

15279c27ce96

2.792453

230

7

0

As someone who often finds himself explaining machine learning to non-experts, I offer the following list as a public service announcement.

5

10 Things Everyone Should Know About Machine Learning As someone who often finds himself explaining machine learning to non-experts, I offer the following list as a public service announcement. Machine learning means learning from data; AI is a buzzword. Machine learning lives up to the hype: there are an incredible number of problems that you can solve by providing the right training data to the right learning algorithms. Call it AI if that helps you sell it, but know that AI, at least as used outside of academia, is often a buzzword that can mean whatever people want it to mean. Machine learning is about data and algorithms, but mostly data. There’s a lot of excitement about advances in machine learning algorithms, and particularly about deep learning. But data is the key ingredient that makes machine learning possible. You can have machine learning without sophisticated algorithms, but not without good data. Unless you have a lot of data, you should stick to simple models. Machine learning trains a model from patterns in your data, exploring a space of possible models defined by parameters. If your parameter space is too big, you’ll overfit to your training data and train a model that doesn’t generalize beyond it. A detailed explanation requires more math, but as a rule you should keep your models as simple as possible. Machine learning can only be as good as the data you use to train it. The phrase “garbage in, garbage out” predates machine learning, but it aptly characterizes a key limitation of machine learning. Machine learning can only discover patterns that are present in your training data. For supervised machine learning tasks like classification, you’ll need a robust collection of correctly labeled, richly featured training data. Machine learning only works if your training data is representative. Just as a fund prospectus warns that “past performance is no guarantee of future results”, machine learning should warn that it’s only guaranteed to work for data generated by the same distribution that generated its training data. Be vigilant of skews between training data and production data, and retrain your models frequently so they don’t become stale. Most of the hard work for machine learning is data transformation. From reading the hype about new machine learning techniques, you might think that machine learning is mostly about selecting and tuning algorithms. The reality is more prosaic: most of your time and effort goes into data cleansing and feature engineering — that is, transforming raw features into features that better represent the signal in your data. Deep learning is a revolutionary advance, but it isn’t a magic bullet. Deep learning has earned its hype by delivering advances across a broad range of machine learning application areas. Moreover, deep learning automates some of the work traditionally performed through feature engineering, especially for image and video data. But deep learning isn’t a silver bullet. You can’t just use it out of the box, and you’ll still need to invest significant effort in data cleansing and transformation. Machine learning systems are highly vulnerable to operator error. With apologies to the NRA, “Machine learning algorithms don’t kill people; people kill people.” When machine learning systems fail, it’s rarely because of problems with the machine learning algorithm. More likely, you’ve introduced human error into the training data, creating bias or some other systematic error. Always be skeptical, and approach machine learning with the discipline you apply to software engineering. Machine learning can inadvertently create a self-fulfilling prophecy. In many applications of machine learning, the decisions you make today affect the training data you collect tomorrow. Once your machine learning system embeds biases into its model, it can continue generating new training data that reinforces those biases. And some biases can ruin people’s lives. Be responsible: don’t create self-fulfilling prophecies. AI is not going to become self-aware, rise up, and destroy humanity. A surprising number of people (cough) seem to be getting their ideas about artificial intelligence from science fiction movies. We should be inspired by science fiction, but not so credulous that we mistake it for reality. There are enough real and present dangers to worry about, from consciously evil human beings to unconsciously biased machine learning models. So you can stop worrying about SkyNet and “superintelligence”. There’s far more to machine learning than I can explain in a top-10 list. But hopefully this serves as a useful introduction for non-experts.

10 Things Everyone Should Know About Machine Learning

734

10-things-everyone-should-know-about-machine-learning-15279c27ce96

2018-05-23

2018-05-23 09:00:39

https://medium.com/s/story/10-things-everyone-should-know-about-machine-learning-15279c27ce96

false

740

null

Machine Learning

machine-learning

Machine Learning

51,320

Daniel Tunkelang

High-Class Consultant.

5f6b5aba680

dtunkelang

4,488

0

20,181,104

null

0

null

0

null

2018-01-17

2018-01-17 07:15:00

2018-01-17

2018-01-17 07:18:57

0

false

en

2018-01-17

2018-01-17 07:18:57

0

152816f14982

0.101887

0

Job most at risk? Jobs Least at risk?

5

Working with Robots. What will it mean for you? “If you feared change before, then my God, you have something coming around the corner.” Chris Riddell, 2017

Working with Robots. What will it mean for you?

0

working-with-robots-what-will-it-mean-for-you-152816f14982

2018-01-17

2018-01-17 07:18:58

https://medium.com/s/story/working-with-robots-what-will-it-mean-for-you-152816f14982

false

27

null

Self Improvement

self-improvement

Self Improvement

151,898

Catherine Cunningham

Career specialist. Author of My Career Rules! Recipes to Crack the Career Code in 21st Century Australia

7f7da944d949

katy1mary1

0

7

20,181,104

null

0

null

0

509e810b2448

2017-12-19

2017-12-19 19:05:42

2017-12-19

2017-12-19 19:36:13

1

false

pt

2017-12-19

2017-12-19 19:36:13

0

15289ba2411c

2.520755

5

0

Há alguns meses realizamos a primeira edição dos webinars View Source, uma iniciativa com o objetivo de conectar você às principais…

5

View Source: brasileiras no programa MVP Palestrantes View Source Há alguns meses realizamos a primeira edição dos webinars View Source, uma iniciativa com o objetivo de conectar você às principais comunidades técnicas e projetos, reunindo suas principais líderes e contributors, para discussões sobre melhores práticas e técnicas de desenvolvimento inovadoras. Nesta edição, teremos algumas representantes da comunidade MVP, compartilhando seus conhecimentos e experiências! Todas as sessões estão disponíveis em nosso canal do youtube e em tempo de planejamento de carreira para o próximo ano, vale a pena conferir dicas e ferramentas para direcionar seus esforços 😊 Usando Microsoft Teams com Bots para produtividade, com Sara Barbosa Entenda nessa sessão como o Microsoft Teams pode integrar com as principais soluções do Office 365, criar um canal de comunicação entre os times e usar os Bots para automatizar tarefas que são feitas em geral de forma manual e muitas vezes são demoradas. Mostrarei como adicionar e usar os Bots nativos mais legais do Marketplace do MS Teams. Sobre a palestrante: Sara Barbosa Mora em Berlin, sou MVP Office 365 há 6 anos, entusiasta na comunidade técnica Microsoft, participa em eventos presenciais e online, faz moderação fóruns técnicos, mantém o blog:http://sarabarbosa.net/ e participa de alguns grupos de discussão, contribuindo com a jornada de Cloud Computing dos clientes há mais de 4 anos. Idealizadora do Projeto Learning 365, que promove mais de 100 treinamentos online por ano, em parceria com outros MVP’s criou o maior evento de Office 365, promovido por comunidades no Brasil: Office 365 Day. Atualmente está focando seus esforços para aprender Alemão 😝 Introdução à TypeScript & Features Ts 2.x com Glaucia Lemos Breve Introdução à TypeScript, quando surgiu, onde usar e porque é tem sido tão aclamado pelos Devs Front à Back End. E com muitas demos legais e features das novidades da versão 2.x. Sobre a palestrante: Glaucia Lemos Software Engineer & Full Stack Developer com especialização em .NET com experiência nacional e internacional em desenvolvimento de sistemas desde pequenas à grandes empresas. Atualmente atuo como Microsoft MVP em Visual Studio and Development Technologies. Sendo assim, atuando como evangelista de novas linguagens de programação, Community Leader & Speaker do Meetup: Coders in Rio. “Quando não estou trabalhando, amo conhecer novos lugares, viajar, fazer novos amigos, estudar novas tecnologias e linguagens de programação. Eu também amo desenhar, ouvir boa música e sou uma pessoa completamente apaixonada quando o assunto é sobre cinema e astronomia.” Transformação Digital e Internet das Coisas com Viviane Heinrichs Conheça um pouco da visão da Microsoft sobre o papel da Internet das Coisas nos negócios, as tecnologias disponíveis na Nuvem Azure e alguns casos de sucesso para saber como as soluções Microsoft estão transformando a indústria. Sobre a palestrante: Viviane Heinrichs Premiada pela quarta vez consecutiva como “Most Valuable Professional” — MVP Microsoft, Engenheira da Computação, trabalha com o desenvolvimento de sistemas embarcados e soluções IoT. Possui as certificações Windows Embedded Black Belt Sales e Technical Master. Tratamento, Modelagem e Análise de Dados com Excel, Power Pivot e Power Query com Karine Lago Arquitetura de um processo de análise informacional com conexão e tratamento de dados (ETL) automaticamente pelo Power Query, relacionamento, modelagem e cálculos para responder as principais necessidades informacionais para tomada de decisão com DAX no Power Pivot e consumo dessa informação estruturada no próprio Excel, através da fantástica tabela dinâmica e gráficos. Sobre a palestrante: Karine Lago Especialista em Inteligência Informacional na Datab Consultoria, Publicitária, Pós-graduada em Gestão Estratégica da Informação pela UFMG, Most Valuable Professional (MVP) e Microsoft Office Specialist Expert (MOSE) em Microsoft Excel. Fundadora do IntelExcel, um canal no YouTube com mais de 900 mil visualizações, 16 mil seguidores e 100 vídeos sobre Excel e Power BI.

View Source: brasileiras no programa MVP

59

view-source-brasileiras-no-programa-mvp-15289ba2411c

2018-06-08

2018-06-08 17:08:37

https://medium.com/s/story/view-source-brasileiras-no-programa-mvp-15289ba2411c

false

615

O WoMakersCode é um projeto de empoderamento e fortalecimento da presença feminina na tecnologia.

null

womakerscode

null

WoMakersCode

contato@womakerscode.org

womakerscode

WOMEN IN TECH,CAREERS,DIVERSITY AND INCLUSION,TECHNOLOGY,SOFTWARE DEVELOPMENT

womakerscode

Women In Tech

women-in-tech

Women In Tech

18,577

Cynthia Zanoni

null

3ca14e77c7ab

cynthiazanoni

639

193

20,181,104

null

0

null

0

661161fab0d0

2018-07-28

2018-07-28 09:41:29

2018-07-28

2018-07-28 10:34:35

1

false

en

2018-08-02

2018-08-02 15:13:12

0

152930ae6e6e

2.267925

4

0

Ever wondered what he was thinking? Or her ? Or maybe we’ve always wanted to express our feeling without saying a word ? What if we could…

5

What if we could Communicate directly from our Brain to Brain ? Ever wondered what he was thinking? Or her ? Or maybe we’ve always wanted to express our feeling without saying a word ? What if we could read someone’s mind and they could read yours ? Would there be no secrets left ? Would mind control become possible ? How could we keep our privacy ? if everyone knew what’s on our mind ? The way we communicate is pretty slow. It takes time from the moment the thought comes to our mind, to the moment we say it to someone, even longer with our two fingers typing. Our brain is much faster than that. On average we have 50,000 thoughts a day. That is 35 thoughts per minute. we can’t type all of them, nor can we put them into words. Imagine how fast we could exchange our thoughts and information, if we could just transmit our thoughts directly to another brain. Once we started communicating through thought, we’d turn into a giant mind-melding network. Humanity would become much more innovative,since the more other people would be able to improve upon them. This would boost our technologies to the level we can’t imagine. Sharing our knowledge through thoughts would also makes us more coordinated.This would benefit all the governmental structures and also emergency services. We’d be learning much faster and easier.Instead of verbally explaining a difficult concept,teachers could give us a better representation of it through thought. Now, emerging Artificial Intelligence(AI) could make it possible that our non-living things could have mind of it’s own,we’d be able to control them with our mind.Controlling things sounds awesome what if this led to controlling people’s mind? It might sounds interesting but wouldn’t seem great if we are the one being controlled by others. Just like we can choose what to say, we’d be able to choose what thoughts to share with others. No one would read our thoughts until and unless we let them. The new era of mind reading would be pretty silent. Without the need to push air through our lungs and into our voice box, we might eventually forget language and sound of it. We don’t know much about our bundle of neurons(Brain), we don’t fully understand why we sleep, or how our conscious works. But we do know how to read our brain. Our brain fires neurons that receive, transmit and process information through electrical and chemical signals and we can read those signals with electroencephalography or EEG. We can now transmit information from one brain to another. Transcranial Magnetic Simulation or TMS let us do that. This simulation sends electric current to the receiver’s brain, activating a small area on it, the same area that is active in the sender’s brain. This is not the telepathy from Science Fiction, but that is what talking through thought would look in real life. For instance, now we use devices to communicate with each other. Eventually we just switch to faster technologies, the same technology that would let us send our thoughts, the way we send emails. We might not be around to use EEGs and TMSs instead of our mobile devices, but this can be the way we communicate through our thoughts.

What if we could Communicate directly from our Brain to Brain ?

170

what-if-we-could-communicate-directly-from-our-brain-to-brain-152930ae6e6e

2018-08-02

2018-08-02 15:13:13

https://medium.com/s/story/what-if-we-could-communicate-directly-from-our-brain-to-brain-152930ae6e6e

false

548

where the future is written

null

Predict

predictstories@gmail.com

predict

FUTURE,SINGULARITY,ARTIFICIAL INTELLIGENCE,ROBOTICS,CRYPTOCURRENCY

null

Neuroscience

neuroscience

Neuroscience

6,742

Binayak Adhikari

Programmer 💻

e89c25c6738f

blazebnayak

21

77

20,181,104

null

0

null

0

null

2018-09-30

2018-09-30 14:22:18

2017-09-23

2017-09-23 07:30:14

2

false

en

2018-09-30

2018-09-30 14:24:09

2

15297c3b9980

3.051258

0

You must have had those times when you were looking at the screen and scratching your head wondering “Why I am typing these three terms in…

1

Epoch vs Batch Size vs Iterations You must have had those times when you were looking at the screen and scratching your head wondering “Why I am typing these three terms in my code and what is the difference between them ” because they all look so similar. To find out the difference between these terms you need to know some of the machine learning terms like Gradient Descent to help you better understand. Here is a short summary on Gradient Descent… It is an iterative optimization algorithm used in machine learning to find the best results (minima of a curve). Gradient means the rate of inclination or declination of a slope. Descent means the instance of descending. The algorithm is iterative means that we need to get the results multiple times to get the most optimal result. The iterative quality of the gradient descent helps a under-fitted graph to make the graph fit optimally to the data. Source The Gradient descent has a parameter called learning rate. As you can see above (left), initially the steps are bigger that means the learning rate is higher and as the point goes down the learning rate becomes more smaller by the shorter size of steps. Also,the Cost Function is decreasing or the cost is decreasing .Sometimes you might see people saying that the Loss Function is decreasing or the loss is decreasing, both Cost and Loss represent same thing (btw it is a good thing that our loss/cost is decreasing). We need terminologies like epochs, batch size, iterations only when the data is too big which happens all the time in machine learning and we can’t pass all the data to the computer at once. So, to overcome this problem we need to divide the data into smaller sizes and give it to our computer one by one and update the weights of the neural networks at the end of every step to fit it to the data given. One Epoch is when an ENTIRE dataset is passed forward and backward through the neural network only ONCE. Since, one epoch is too big to feed to the computer at once we divide it in several smaller batches. I know it doesn’t make sense in the starting that the passing the entire dataset through a neural network is not enough and we need to pass the full dataset multiple times to the same neural network. But keep in mind that we are using a limited dataset and to optimise the learning and the graph we are using Gradient Descent which is an iterative process. So, updating the weights with single pass or one epoch is not enough. One epoch leads to underfitting of the curve in the graph (below). As the number of epochs increases, more number of times the weight are changed in the neural network and the curve goes from underfitting to optimal to overfitting curve. Unfortunately, there is no right answer to this question. The answer is different for different datasets but you can say that the numbers of epochs is related to how diverse your data is… just an example — Do you have only black cats in your dataset or is it much more diverse dataset? Total number of training examples present in a single batch. Note: Batch size and number of batches are two different things. As I said, you can’t pass the entire dataset into the neural net at once. So, you divide dataset into Number of Batches or sets or parts. Just like you divide a big article into multiple sets/batches/parts like Introduction, Gradient descent, Epoch, Batch size and Iterations which makes it easy to read the entire article for the reader and understand it. 😄 To get the iterations you just need to know multiplication tables or have a calculator. 😃 Iterations is the number of batches needed to complete one epoch. Note: The number of batches is equal to number of iterations for one epoch. Let’s say we have 2000 training examples that we are going to use . We can divide the dataset of 2000 examples into batches of 500 then it will take 4 iterations to complete 1 epoch. Originally published at towardsdatascience.com on September 23, 2017.

Epoch vs Batch Size vs Iterations

0

epoch-vs-batch-size-vs-iterationsyou-must-have-had-those-times-when-you-were-looking-at-the-screen-15297c3b9980

2018-10-02

2018-10-02 13:38:33

https://medium.com/s/story/epoch-vs-batch-size-vs-iterationsyou-must-have-had-those-times-when-you-were-looking-at-the-screen-15297c3b9980

false

707

null

Machine Learning

machine-learning

Machine Learning

51,320

Rajib Biswas

null

c5ad29effcf0

online.rajib

2

4

20,181,104

null

0

null

0

3b633d729daf

2018-02-07

2018-02-07 17:18:50

2018-02-07

2018-02-07 17:18:49

1

false

en

2018-02-08

2018-02-08 02:00:20

2

1529e7c8585b

1.233962

2

0

null

5

Twitter issues a full ban on ‘deepfake’ AI-generated porn The fight against “deepfake” porn now has another enemy: Twitter. The company told Motherboard that it’s investigating and banning accounts that are either the original posters of AI-edited videos or are accounts dedicated to posting these types of clips. Twitter says that these face swaps violate the company’s “intimate media” policy, which bans any sexually explicit photos or videos produced or shared without someone’s consent. Basically, this is on par with revenge porn and that’s exactly the route Twitter is taking in terms of handling this situation. Twitter joins other companies like Discord, Gfycat and Pornhub, all of whom have stated that they will not allow deepfake and other non consensual porn on their platforms. While companies are taking a hard stance, there’s no full guarantee that they can completely eliminate these types of posts, but at least we have a stance from these companies and know that they’re working hard to eliminate this type of threat from their platforms. Unlike other companies, Twitter is actually in an interesting situation because it allows sexually explicit material on its platform as long as it’s flagged properly. Facebook on the other hand doesn’t allow it at all. While Twitter is hard at work to enforce these policies, it may not be enough. Reddit’s deepfake subreddits, where the AI-built porn really took off, is still running and has thousands of subscribers, and shows no signs of slowing down. It might be a lot harder for Twitter to eliminate deepfake porn if the necessary tools to create are widely available. Originally published at The Jolt Journal.

Twitter issues a full ban on ‘deepfake’ AI-generated porn

19

twitter-issues-a-full-ban-on-deepfake-ai-generated-porn-1529e7c8585b

2018-03-12

2018-03-12 06:53:43

https://medium.com/s/story/twitter-issues-a-full-ban-on-deepfake-ai-generated-porn-1529e7c8585b

false

274

The Jolt Journal provide you with the latest breaking news and videos straight from the tech and entertainment industry.

null

joltjournal

null

The Jolt Journal

info@joltjournal.com

the-jolt-journal

TECH,TECHNEWS,ENTERTAINMENT,BREAKING NEWS,POLITICS

thejoltjournal

Gear

gear

Gear

1,009

The Jolt Journal

The Jolt Journal provide you with the latest breaking news and videos straight from the tech and entertainment industry.

f0c02201635f

joltjournal

3

10

20,181,104

null

0

null

0

a05e9703acc6

2018-06-27

2018-06-27 14:43:00

2018-06-27

2018-06-27 15:42:06

2

false

en

2018-06-28

2018-06-28 15:15:21

14

152b3144735e

4.190881

16

0

By Hamed Alemohammad, Lead Geospatial Data Scientist, Radiant.Earth

5

Radiant.Earth Launches New Technical Working Group on Machine Learning for Global Development By Hamed Alemohammad, Lead Geospatial Data Scientist, Radiant.Earth The new Technical Working Group on Machine Learning for Global Development, representing 23 institutions from various sectors. As part of a recent Radiant.Earth workshop, 30 leading international experts participated in the launch of a new Technical Working Group on Machine Learning for Global Development. The group includes Earth observations (EO), machine learning (ML), and land cover (LC) classification experts, all working collaboratively towards the goal of developing a community standard on best practices for use of ML with EO, a commons for labeled training data catalogues, and a hierarchical schema for global LC classification. “Radiant.Earth is developing open source datasets of labeled satellite images, which will be hosted on the MLHub.Earth with a Creative Commons license.” Radiant.Earth is developing open source datasets of labeled satellite images, which will be hosted on the MLHub.Earth with a Creative Commons license. These datasets will lead to a living open image library for ML and EO. Our goal is to create a sustained, community-wide effort to capture image labels that would enable major innovations and will drive new, more targeted and timely insights supporting progress in areas such as agriculture, food security, conservation, health, land rights, urban planning, water resources, and other areas relevant to global development and humanitarian response. The first of such datasets that Radiant.Earth will generate consist of global LC labeled imagery from Sentinel-2 satellites at 10 m spatial resolution. This will enable fully-automated and dynamic LC classification algorithms, using open source satellite imagery. Radiant.Earth will label these images using a combination of ML and crowdsourcing to generate a human-verified training dataset. “Existing training datasets for LC classification have limitations that do not support development of a global EO-based LC classification algorithm at fine spatial resolutions with high accuracy.” Existing training datasets for LC classification have limitations that do not support development of a global EO-based LC classification algorithm at fine spatial resolutions with high accuracy. These datasets are either generated for specific regions of the world (therefore, they lack geo-diversity) or are based on imagery that are not freely available at the global scale (therefore, they are not open source). Moreover, in many cases, very few labeled images are available for a specific class within the dataset, which limits the performance of a ML algorithm to learn the particular features of that class. Budhendra Bhaduri, Corporate Research Fellow at Oak Ridge National Lab, shares his perspective on using machine learning and high performance computing for LC classification. Key topics of the Technical Working Group Radiant.Earth formed the technical working group on Machine Learning for Global Development to best define the specification of such a global dataset to meet the requirements for end-user applications and to standardize best practices to increase the interoperability of different datasets and algorithms. The group members are experts from commercial, government, non-profit and academic organizations with subject matter knowledge related to this topic. Existing and future activities of the group are documented on this GitHub repository. The first meeting of the working group focused on the topic of “Machine Learning for Global Land Cover Classification,” on June 14–15, 2018 in Washington, D.C. Thirty experts representing 23 institutions gathered and presented their latest advancements in the use of ML for LC classification. Presenters also shared their thoughts on the challenges and remaining barriers to improve the accuracy of global LC maps. To facilitate further discussions and examination of key topics, experts participated in one of three groups, which are summarized below: Group 1 focused on developing a hierarchical LC schema to include all major LC classes at global scale and enable inter-comparison and cross-validation of different LC products that use satellite imagery at different spatial resolutions. Highlighting the importance of distinguishing between LC and land use, the group developed a hierarchical LC schema combined with a set of attributes which is translatable so that refined details can be added in each class later on. The schema is designed for a global LC product and assumes that the LC definitions will be updated annually. Details of the schema are provided in here. Group 2 reviewed the challenges and ad-hoc choices for using ML with EO data. After two days of discussions, they generated a set of best practices for this application. Their recommendations are focused on four topics: (1) accuracy of training data labels, (2) achieving higher accuracies within and between LC classes, (3) maintaining labeled training datasets and (4) best practices for a global LC algorithm using Sentinel-2 imagery. Their detailed recommendations are included in the notes from the meeting (available here), and covers all aspects of these four topics. Group 3 examined current standards in storing and distributing labeled satellite imagery and the caveats related to each of them. They also developed a training data architecture using the Spatio-Temporal Asset Catalogue (STAC) specifications. This training data specification enables combining raw imagery and label information in one standard catalogue that is adaptable to a wide range of labeled imagery. It will accelerate adoption and use of these data in ML algorithms. The label asset in the catalogue allows for the labels to be “tile classification,” “object detection,” or “segmentation of pixels.” The draft version of this spec is published in this GitHub repository along with a sample GeoJSON file from the SpaceNet challenge. Notes from the group discussions are also available here. White Paper on Machine Learning for Global Land Cover Classification The results of discussions from all three groups, currently being synthesized and documented, will be published as a white paper. Radiant.Earth is also working with other groups that have generated or are generating labeled imagery for implementation of these specifications and standards. This is an ongoing effort and the specifications will evolve in the next couple of months to reach an adoptable and operational level. We believe in collaborative innovations and will invest in similar gatherings to facilitate adoption and dissemination of ML techniques applied to EO in the future. Finally, I would like to thank Schmidt Futures for sponsoring this project and workshop, as well as our wonderful Radiant.Earth team that worked tirelessly to make this workshop a great success.

Radiant.Earth Launches New Technical Working Group on Machine Learning for Global Development

39

radiant-earth-launches-new-technical-working-group-on-machine-learning-for-global-development-152b3144735e

2018-06-29

2018-06-29 01:01:44

https://medium.com/s/story/radiant-earth-launches-new-technical-working-group-on-machine-learning-for-global-development-152b3144735e

false

1,009

Helping the global development community navigate the Earth observation marketplace and geospatial technology innovations.

null

OurRadiantEarth

null

Radiant Earth Insights

social.logins@radiant.earth

radiant-earth-insights

EARTH OBSERVATION,DRONES,MACHINE LEARNING,OPEN DATA,REMOTE SENSING

OurRadiantEarth

Machine Learning

machine-learning

Machine Learning

51,320

Radiant Earth Foundation

A non-profit committed to aggregating the world’s open Earth imagery and providing access and education on its use to the global development community.

c1c2084fd8a

ourradiantearth

139

70

20,181,104

null

0

null

0

a21c0099813b

2018-03-16

2018-03-16 16:41:50

2018-03-16

2018-03-16 20:48:02

6

false

en

2018-03-17

2018-03-17 09:44:31

4

152c45146180

4.746226

1

0

Africa as the propounder of Afrobeat and host to so many other musical styles/sounds including Afro-pop, rap, reggae, jazz, hip-pop, etc…

4

Why Chatbots will dominate the music industry Some renowned musicians in the African music industry. Africa as the propounder of Afrobeat and host to so many other musical styles/sounds including Afro-pop, rap, reggae, jazz, hip-pop, etc. African sounds have created impact in the whole world with songs of famous African musicians such as Fela kuti, Yusu Ndur, Mariam Makeba, salif keita, Lucky Dube, Hugh Masekela, Manu Dibango, etc played everywhere, influencing different generations and generating revenue for the artists and record labels. Over the last ten years, The African music industry has seen the emergence of new artists making even greater impact across the continent and the world at large. some top musicians across the continent. Some of them include Wizkid, Sarkodie, Davido, Casper Nyovest, AKA, Tekno, Olamide, Stanley Enow etc. These new generations of artists generate impressive amounts of money with their music and keep breaking trends. On the continent, it’s difficult to actually get accurate data about the overall monetary value of the music industry. But the music industry in some countries including South Africa, Nigeria, and Ghana have shown promising revenue generation prospects with more artists emerging and leaving from their music. In addition, with the huge financial prospect ahead for the industry for example According to the Pwc South Africa in it’s” PwC Entertainment and media outlook: 2017–2021, An African perspective, www.pwc.co.za/outlook, the Nigerian music industry by 2020 is projected to be worth 86 million dollars and Total music revenue is forecast for the South African music industry is to reach R2.8 billion in 2021. Many artists have gained attention from world renowned record labels such Roc Nation, Sony music and Universal music, with an increasing number of deals being signed. Music piracy across the continent is a big threat to the industry. Even though there has been hype as to future prospects awaiting the African music industry especially when it comes to revenue generation, there are still numerous challenges that need to be solved to create more opportunities for the music industry stakeholders to generate more revenue. Some of these challenges include; 1. The problem of piracy and intellectual property scam, that makes the African music industry to lose a lot of money yearly. 2. In addition to this is the fact that many governments do not put in place laws nor initiatives to support the music industry in its development phase. 3. With the rise of digital and mobile technologies, many artists with their awesome music try to sell their music online and also stream, but due to the lack of expertise, lack of proper digital strategies and tools, many artists can not still generate profit. 4. In line with this, so many artist have many followers and somehow appropriate content but face low engagement because they still need extra tools to understand their followers or fans. That notwithstanding, even though the industry in Africa is facing numerous challenges, there are still great projections when it comes to revenue generation. Picture of music streaming on an iphone. According to the Pwc South Africa in it’s” PwC Entertainment and media outlook: 2017–2021, An African perspective, www.pwc.co.za/outlook, mobile and digital technologies will have a central role in Revenue generation by the music industry across Africa, for example in Nigeria digital platforms will account for over 90% of recorded music sales. In this same report they try to forecast the mobile internet penetration change in some countries over the period 2016–2021, these include ; South Africa: 52% to 78%, Nigeria: 16% to 40%, Kenya:34% to 61%, Ghana 22% to 48% and Tanzania 13% to 29%). With the central role that mobile and digital technologies will play, Artificial intelligence, specifically chatbots can help in providing amazing, scalable and low cost solutions to the numerous problems faced by the music industry in Africa. A Chatbot is a computer program powered by AI that you interact with via a chat or voice interface. With messaging apps such as messenger, whatsapp, etc usage numbers rising and becoming bigger than social media apps by 20%, there was a real need for service/product providers to explore the potential of messaging apps in ameliorating customer daily experience and relationships. With over 1.3 billion monthly users, in 2016 Facebook messenger launched its chatbot API and the whole game changed and opened up new opportunities for business, individuals and institutions to reach out to a huge global market and ameliorate customer experience in a unique and customizable format. Musicians, record labels, digital streaming platforms can generate a lot of profit through the use of chatbots in the following ways; 1. Generate a genuine fan database for musicians: A chatbot can be used to collect strategic data (location, age, sex, phone number, email, needs and preferences) about fans and followers on facebook and messenger. In line with this, a chatbot is able to segment followers on messenger to help artist to better understand their fans who they are, what they want, how they interact with chatbot, etc. 2. Send personalized content, boost content at no cost : Musicians through a chatbot, can be able to send messages, images, videos to all their followers on messenger without spending any money on a frequent basis. This can be done in a personalized way adapted for each follower. This can help a lot in generating more revenues through music streams and music launching. 3. Handle Numerous fan discussions A chatbot can be programmed to respond to thousands of frequently asked questions by fans automatically without human intervention and redirect fan to a human for chat when he/she will want a more personal discussion. 4. Customize Fan and followers engagement Chatbots give the possibility to musicians to better engage their fans and followers in their music creation process. Through a chatbot musicians can conduct a survey within their fan base, etc. 5. So much more….. Karrueche chatbot developed by Persona Artist worldwide such as; Katty Perry: https://www.messenger.com/t/katyperry Christina milian: https://www.messenger.com/t/christinamilian DJ Hardwell: https://www.messenger.com/t/djhardwell Use chatbots and the results as they say, are impressive! . Juubots can help musicians, record labels, digital music distribution companies to take advantage of this trend to generate more revenue with impressive chatbot best practices. Let’s talk about your idea or project. Contact Us Email: contact@juubots.com Tel: (+237) 671053149

Why Chatbots will dominate the music industry

30

why-chatbots-will-dominate-the-music-industry-152c45146180

2018-03-17

2018-03-17 09:44:32

https://medium.com/s/story/why-chatbots-will-dominate-the-music-industry-152c45146180

false

1,006

Outline of the impact that chatbots can have on the music industry.

null

juubots

null

Why chatbots will dominate the music industry

contact@juubots.com

why-chatbots-will-dominates-the-music-industry

CHATBOTS,AI,MUSIC,CHATBOT DEVELOPMENT,ÁFRICA

juubots

Music

music

Music

174,961

Juubots

Juubots is a bot development company that specializes in chatbots.

90643f8481e6

jehptes

2

20,181,104

null

0

null

0

null

2018-05-19

2018-05-19 00:11:22

2018-05-21

2018-05-21 18:05:11

3

false

en

2018-05-21

2018-05-21 18:05:11

5

152c60438b59

2.116038

9

0

Here at Josh.ai, we spend a lot of time thinking about how to make the way we interface with technology natural, easy, and even fun. Part…

5

Choose Your Voice with Josh.ai Here at Josh.ai, we spend a lot of time thinking about how to make the way we interface with technology natural, easy, and even fun. Part of giving that experience to users means enabling users to choose the type of interface they feel like using, as well as allowing them to customize that interface to fit the way they communicate. For example, when using the Josh app to control the home, users can create their own dashboard, populating it with the scenes and devices most important to them. These sorts of customizations should also extend beyond a traditional UI. Most people talk about Josh’s advanced home voice control abilities, such as giving compound commands to control multiple things at once. Beyond the cool tech, users have the ability to customize their voice experience as well. With Josh, users can choose what type of voice they would like to converse with, including the ability to specify gender and accent. Gender and Accent Josh lets you choose whether you want to speak to a male or female voice. This can be changed at any time, so you can adjust based on whatever you are feeling. In addition to gender, you can also choose the type of accent. Here are the options to choose from: Male American Australian British Female American Australia British Indian Irish South African How to Change Voices You can change the voice of Josh using either our web portal or via voice. To change via the web portal: Click on your username at the top right of the page Choose EDIT JOSH In the PERSONALITY INFO section, you can choose the drop down for male/female as well as the type of accent. To change via voice: You can talk to Josh either through the Josh app or by using Josh Micro. We made it simple so that you can naturally tell Josh what gender and accent you want. For example: OK Josh, use a female Irish accent. The default voice is whatever you have chosen in the EDIT JOSH section of the web portal. To get back to this voice, simply ask Josh to revert, say something like: OK Josh, use your regular voice. Voice customizations are just one feature that makes Josh the most natural and advanced controller for your home. Josh.ai is an artificial intelligence agent for your home. If you’re interested in learning more, visit us at https://josh.ai. Like us on Facebook, follow us on Twitter.

Choose Your Voice with Josh.ai

17

choose-your-voice-with-josh-ai-152c60438b59

2018-06-12

2018-06-12 14:34:01

https://medium.com/s/story/choose-your-voice-with-josh-ai-152c60438b59

false

415

null

Smart Home

smart-home

Smart Home

3,891

Josh

null

66b5ae01967f

joshdotai

43,719

1,883

20,181,104

null

0

null

0

null

2018-08-01

2018-08-01 09:07:03

2018-08-26

2018-08-26 12:35:55

0

false

en

2018-08-26

2018-08-26 12:35:55

13

152d00548154

4.498113

22

0

1

In the middle of the 20th century, IBM used its headquarters in New York City as a showroom of tomorrow. Passersby could look into the…

1

Marketers Need To Start Preparing For The End Of The Digital Age And The New Era Of Innovation In the middle of the 20th century, IBM used its headquarters in New York City as a showroom of tomorrow. Passersby could look into the window and see the newest mainframe on display, promising an exciting technological future. It was the dawn of the computer age, but marketers were largely out of the picture. It would be hard to explain the the “Mad Men” back in the 1960s that someday those big, hulking machines would shrink down small enough to fit in our pockets, that these devices would have screens and that they would, to a large extent, replace TVs as the dominant driver of commerce. Today, as Moore’s Law is slowly petering to an end, we’re on the brink of a new era and, in time, marketing will be transformed once again in ways that are hard to see right now. Over the next decade marketers will need to begin to shift to the post-digital world of computing. This next transformation promises to be at least as revolutionary as the last one. Shifting To New Computing Architectures The central task of early mainframe computers was to perform calculations that few people had any use for — huge back office tasks like accounting and payroll and complex scientific calculations. However, as computing became exponentially cheaper, marketers began to use it for research. Later came the Internet, banner ads, CRM and a thousand other things. The two computing architectures most likely to replace digital computing, quantum computers and neuromorphic chips, are possibly even more obscure than mainframes were in the early days. The first, quantum computing, will create incredibly large computing spaces and the second, neuromorphic computing, will mimic how our brains process information. The first applications that quantum computers will find a market for are likely to be truly massive simulations — such as calculating subatomic energy states — that will fundamentally alter our ability to understand the physical world. Neuromorphic chips will first be deployed in artificial intelligence tasks and, because they are incredibly energy efficient, in the Internet of Things. In a decade or so, we may start to see the first marketing applications. Much like quantum simulations will enable us to simulate particles in the real world, they may also be able to do the same for consumers. Neuromorphic chips may be able to help us understand the data coming from those simulations and also, because of their efficiency, create sensors that provide more information to analyze. AI As The New UI In 1988, Don Norman published his seminal book, The Design of Everyday Things , which is largely seen as pioneering the user-centered design movement. Today, user experience has become a thriving field in itself and marketers have learned, for the most part, that better experiences can be a key sales driver. Yet the interface itself is changing rapidly. We no longer solely use keyboards, mouses or even touch screens to interact with our machines, but are increasingly using voice, motion and even our facial and biological characteristics to guide technology to do what we want it to. In effect, artificial intelligence is the new user interface. Over the next decade, technologies like quantum computing and neuromorphic chips are likely to exponentially improve our AI capabilities and our capacity to improve the user experience will need to improve with it. Our ability to gain value through the information and intelligence out of the systems we build will only be as good as our interactions with them. So a major task for the coming years will be to to design conversational intelligence, including the ability to preserve context, so that our systems understand how our commands and queries relate to previous turns in the conversation. They will also need to seamlessly integrate multiple interfaces, such as touch, voice and biometric patterns to not only respond to our commands, but also our emotions, through changes in facial expressions, tone of voice and other data. The digital age marked its progress through interfaces. We moved from keyboards, to mouses, to touch and eventually to voice. Marketing systems of the the future will be defined by our ability to make interfaces disappear. Blockchain As The New Database In 1970, a researcher at IBM cnamed Edgar F. Codd came up with a new way to store and retrieve data called the relational database. It was a discovery that was so obscure that, in fact, IBM itself didn’t fully understand its implications and the industry it spawned would be dominated by companies like Oracle and SAP. Now imagine yourself traveling back to say, 1972 and trying to explain the impact of this new technology to a marketer of the time. Relational databases would go on to have a major impact on the industry, forming the basic technological infrastructure for things like research and booking, but none of that would have meant much back then. Today, we are in the midst of a similar revolution driven by blockchain and the transformation will be just as subtle, but no less impactful. Blockchain is, at its core, a new kind of distributed database, made secure through encryption. Over the next decade, it will begin to replace our existing technological infrastructure and that will have wide-ranging implications. Probably the most important aspect will be blockchain’s ability to create trust through its audit function, which will enable new models for attribution and consumer opt-in. That, in turn, will give rise to new business models and, most probably, profoundly alter the relationship between marketers and consumers. A trust protocol rewards trustworthy behavior and punishes deception. The Twilight Of The Digital Age A decade ago, there were still serious debates about the significance of digital technology for marketing. Banner ads were still the dominant digital medium, social media was still in its infancy and mobile marketing hadn’t really gotten started yet. It is nothing less than remarkable how much things have changed in such a short amount of time. So it is understandable that a significant amount of marketers’ time and energy has been spent chasing “shiny objects.” Social media, online video, mobile technology and now, artificial intelligence, are each transformative technologies. They have all been introduced in such quick succession that marketers have had little time to catch their breath. Over the next decade, however, there will be few advances in fundamental technologies, except for artificial intelligence. Advancement in computer chips has already slowed to a crawl. The hiatus will give marketers the time to catch their breath and begin to deploy the present technologies more effectively before the next wave of disruption starts in 10–20 years. What will have to change is the marketing mindset. The fundamental questions in the coming years will not be how to deploy this or that new technology, but how to can solve fundamental marketing problems, such as how to earn consumers’ trust and how to create experiences that are more impactful, useful, productive and beneficial. – Greg An earlier version of this article first appeared in Inc.com Originally published at www.digitaltonto.com.

Marketers Need To Start Preparing For The End Of The Digital Age And The New Era Of Innovation

69

marketers-need-to-start-preparing-for-the-end-of-the-digital-age-and-the-new-era-of-innovation-152d00548154

2018-08-26

2018-08-26 14:29:13

https://medium.com/s/story/marketers-need-to-start-preparing-for-the-end-of-the-digital-age-and-the-new-era-of-innovation-152d00548154

false

1,192

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Greg Satell

Author of Mapping Innovation, Speaker, Innovation Adviser, @HBR and @Inc Contributor, Publisher- www.DigitalTonto.com - Learn more at www.GregSatell.com.

22f5b0d6a0b9

digitaltonto

10,459

917

20,181,104

null

0

null

0

ec804170116a

2017-12-14

2017-12-14 01:47:31

2017-12-14

2017-12-14 01:58:39

0

false

en

2018-09-04

2018-09-04 07:04:03

86

152d33039f56

8.735849

1

0

Every great data visualization starts with good and clean data. Most of people believe that collecting big data would be a rough thing, but…

5

Big Data: 70 Free Data Sources You Should Know Every great data visualization starts with good and clean data. Most of people believe that collecting big data would be a rough thing, but it’s simply not true. There are thousands of free data sets available online, ready to be analyzed and visualized by anyone. Here we’ve rounded up 70 free data sources on government, crime, health, financial and economic data,marketing and social media, journalism and media, real estate, company directory and review, and more. We hope you could enjoy this and save a lot time and energy searching blindly online. Free Data Source: Government Data.gov: It is the first stage and acts as a portal to all sorts of amazing information on everything from climate to crime freely by the US Government. Data.gov.uk: There are datasets from all UK central departments and a number of other public sector and local authorities. It acts as a portal to all sorts of information on everything, including business and economy, crime and justice, defence, education, environment, government, health, society and transportation. US. Census Bureau: The website is about the government-informed statistics on the lives of US citizens including population, economy, education, geography, and more. The CIA World Factbook: Facts on every country in the world; focuses on history, government, population, economy, energy, geography, communications, transportation, military, and transnational issues of 267 countries. Socrata: Socratais a mission-driven software company that is another interesting place to explore government-related data with some visualization tools built-in. Its data as a service has been adopted by more than 1200 government agencies for open data, performance management and data-driven government. European Union Open Data Portal: It is the single point of access to a growing range of data from the institutions and other bodies of the European Union. The data boosts includes economic development within the EU and transparency within the EU institutions, including geographic, geopolitical and financial data, statistics, election results, legal acts, and data on crime, health, the environment, transport and scientific research. They could be reused in different databases and reports. And more, a variety of digital formats are available from the EU institutions and other EU bodies. The portal provides a standardised catalogue, a list of apps and web tools reusing these data, a SPARQL endpoint query editor and rest API access, and tips on how to make best use of the site. Canada Open Datais a pilot project with many government and geospatial datasets. It could help you explore how the Government of Canada creates greater transparency, accountability, increases citizen engagement, and drives innovation and economic opportunities through open data, open information, and open dialogue. Datacatalogs.org: It offers open government data from US, EU, Canada, CKAN, and more. U.S. National Center for Education Statistics: The National Center for Education Statistics (NCES) is the primary federal entity for collecting and analyzing data related to education in the U.S. and other nations. UK Data Service: The UK Data Service collection includes major UK government-sponsored surveys, cross-national surveys, longitudinal studies, UK census data, international aggregate, business data, and qualitative data. Free Data Source: Crime 11. Uniform Crime Reporting: The UCR Program has been the starting place for law enforcement executives, students, researchers, members of the media, and the public seeking information on crime in the US. 12. FBI Crime Statistics: Statistical crime reports and publications detailing specific offenses and outlining trends to understand crime threats at both local and national levels. 13. Bureau of Justice Statistics: Information on anything related to U.S. justice system, including arrest-related deaths, census of jail inmates, national survey of DNA crime labs, surveys of law enforcement gang units, etc. 14. National Sex Offender Search: It is an unprecedented public safety resource that provides the public with access to sex offender data nationwide. It presents the most up-to-date information as provided by each Jurisdiction. Free Data Source: Health 15. U.S. Food & Drug Administration: Here you will find a compressed data file of the Drugs@FDA database. Drugs@FDA, is updated daily, this data file is updated once per week, on Tuesday. 16. UNICEF: UNICEF gathers evidence on the situation of children and women around the world. The data sets include accurate, nationally representative data from household surveys and other sources. 17. World Health Organisation: statistics concerning nutrition, disease and health in more than 150 countries. 18. Healthdata.gov: 125 years of US healthcare data including claim-level Medicare data, epidemiology and population statistics. 19. NHS Health and Social Care Information Centre: Health data sets from the UK National Health Service. The organization produces more than 260 official and national statistical publications. This includes national comparative data for secondary uses, developed from the long-running Hospital Episode Statistics which can help local decision makers to improve the quality and efficiency of frontline care. Free Data Source: Financial and Economic Data 20. World Bank Open Data: Education statistics about everything from finances to service delivery indicators around the world. 21. IMF Economic Data: An incredibly useful source of information that includes global financial stability reports, regional economic reports, international financial statistics, exchange rates, directions of trade, and more. 22. UN Comtrade Database: Free access to detailed global trade data with visualizations. UN Comtrade is a repository of official international trade statistics and relevant analytical tables. All data is accessible through API. 23. Global Financial Data: With data on over 60,000 companies covering 300 years, Global Financial Data offers a unique source to analyze the twists and turns of the global economy. 24. Google Finance: Real-time stock quotes and charts, financial news, currency conversions, or tracked portfolios. 25. Google Public Data Explorer: Google’s Public Data Explorer provides public data and forecasts from a range of international organizations and academic institutions including the World Bank, OECD, Eurostat and the University of Denver. These can be displayed as line graphs, bar graphs, cross sectional plots or on maps. 26. U.S. Bureau of Economic Analysis: U.S. official macroeconomic and industry statistics, most notably reports about the gross domestic product (GDP) of the United States and its various units. They also provide information about personal income, corporate profits, and government spending in their National Income and Product Accounts (NIPAs). 27. Financial Data Finder at OSU: Plentiful links to anything related to finance, no matter how obscure, including World Development Indicators Online, World Bank Open Data, Global Financial Data, International Monetary Fund Statistical Databases, and EMIS Intelligence. 28. National Bureau of Economic Research: Macro data, industry data, productivity data, trade data, international finance, data, and more. 29. U.S. Securities and Exchange Commission: Quarterly datasets of extracted information from exhibits to corporate financial reports filed with the Commission. 30. Visualizing Economics: Data visualizations about the economy. 31. Financial Times: The Financial Times provides a broad range of information, news and services for the global business community. Free Data Source: Marketing and Social Media 32. Amazon API: Browse Amazon Web Services’Public Data Sets by category for a huge wealth of information. Amazon API Gateway allows developers to securely connect mobile and web applications to APIs that run on Amazon Web(AWS) Lambda, Amazon EC2, or other publicly addressable web services that are hosted outside of AWS. 33. American Society of Travel Agents: ASTA is the world’s largest association of travel professionals. It provides members information including travel agents and the companies whose products they sell such as tours, cruises, hotels, car rentals, etc. 34. Social Mention: Social Mention is a social media search and analysis platform that aggregates user-generated content from across the universe into a single stream of information. 35. Google Trends: Google Trends shows how often a particular search-term is entered relative to the total search-volume across various regions of the world in various languages. 36. Facebook API: Learn how to publish to and retrieve data from Facebook using the Graph API. 37. Twitter API: The Twitter Platform connects your website or application with the worldwide conversation happening on Twitter. 38. Instagram API: The Instagram API Platform can be used to build non-automated, authentic, high-quality apps and services. 39. Foursquare API: The Foursquare API gives you access to our world-class places database and the ability to interact with Foursquare users and merchants. 40. HubSpot: A large repository of marketing data. You could find the latest marketing stats and trends here. It also provides tools for social media marketing, content management, web analytics, landing pages and search engine optimization. 41. Moz: Insights on SEO that includes keyword research, link building, site audits, and page optimization insights in order to help companies to have a better view of the position they have on search engines and how to improve their ranking. 42. Content Marketing Institute: The latest news, studies, and research on content marketing. Free Data Source: Journalism and Media 43. The New York Times Developer Network– Search Times articles from 1851 to today, retrieving headlines, abstracts and links to associated multimedia. You can also search book reviews, NYC event listings, movie reviews, top stories with images and more. 44. Associated Press API: The AP Content API allows you to search and download content using your own editorial tools, without having to visit AP portals. It provides access to images from AP-owned, member-owned and third-party, and videos produced by AP and selected third-party. 45. Google Books Ngram Viewer: It is an online search engine that charts frequencies of any set of comma-delimited search strings using a yearly count of n-grams found in sources printed between 1500 and 2008 in Google’s text corpora. 46. Wikipedia Database: Wikipedia offers free copies of all available content to interested users. 47. FiveThirtyEight: It is a website that focuses on opinion poll analysis, politics, economics, and sports blogging. The data and code on Github is behind the stories and interactives at FiveThirtyEight. 48. Google Scholar: Google Scholar is a freely accessible web search engine that indexes the full text or metadata of scholarly literature across an array of publishing formats and disciplines. It includes most peer-reviewed online academic journals and books, conference papers, theses and dissertations, preprints, abstracts, technical reports, and other scholarly literature, including court opinions and patents. Free Data Source: Real Estate 49. Castles: Castles are a successful, privately owned independent agency. Established in 1981, they offer a comprehensive service incorporating residential sales, letting and management, and surveys and valuations. 50. Realestate.com: RealEstate.com serves as the ultimate resource for first-time home buyers, offering easy-to-understand tools and expert advice at every stage in the process. 51. Gumtree: Gumtree is the first site for free classifieds ads in the UK. Buy and sell items, cars, properties, and find or offer jobs in your area is all available on the website. 52. James Hayward: It provides an innovative database approach to residential sales, lettings & management. 53. Lifull Home’s: Japan’s property website. 54. Immobiliare.it: Italy’s property website. 55. Subito: Italy’s property website. 56. Immoweb: Belgium’s leading property website. Free Data Source: Business Directory and Review 57. LinkedIn: LinkedIn is a business- and employment-oriented social networking service that operates via websites and mobile apps. It has 500 million members in 200 countries and you could find the business directory here. 58. OpenCorporates: OpenCorporates is the largest open database of companies and company data in the world, with in excess of 100 million companies in a similarly large number of jurisdictions. Our primary goal is to make information on companies more usable and more widely available for the public benefit, particularly to tackle the use of companies for criminal or anti-social purposes, for example corruption, money laundering and organised crime. 59. Yellowpages: The original source to find and connect with local plumbers, handymen, mechanics, attorneys, dentists, and more. 60. Craigslist: Craigslist is an American classified advertisements website with sections devoted to jobs, housing, personals, for sale, items wanted, services, community, gigs, résumés, and discussion forums. 61. GAF Master Elite Contractor: Founded in 1886, GAF has become North America’s largest manufacturer of commercial and residential roofing (Source: Fredonia Group study). Our success in growing the company to nearly $3 billion in sales has been a result of our relentless pursuit of quality, combined with industry-leading expertise and comprehensive roofing solutions. Jim Schnepper is the President of GAF, an operating subsidiary of Standard Industries. When you are looking to protect the things you treasure most, here are just some of the reasons why we believe you should choose GAF. 62. CertainTeed: You could find contractors, remodelers, installers or builders in the US or Canada on your residential or commercial project here. 63. Companies in California: All information about companies in California. 64. Manta: Manta is one of the largest online resources that deliver products, services and educational opportunities. The Manta directory boasts millions of unique visitors every month who search comprehensive database for individual businesses, industry segments and geographic-specific listings. 65. EU-Startups: Directory about startups in EU. 66. Kansas Bar Association: Directory for lawyers. The Kansas Bar Association (KBA) was founded in 1882 as a voluntary association for dedicated legal professionals and has more than 7,000 members, including lawyers, judges, law students, and paralegals. Free Data Source: Other Portal Websites 67. Capterra: Directory about business software and reviews. 68. Monster: Data source for jobs and career opportunities. 69. Glassdoor: Directory about jobs and information about inside scoop on companies with employee reviews, personalized salary tools, and more. 70. The Good Garage Scheme: Directory about car service, MOT or car repair. 71. OSMOZ: Information about fragrance. 72. Octoparse: A free data extraction tool to collect all the web data mentioned above online. Do you know some great data sources? Contact to let us know and help us share the data love. Source: Octoparse More Related Sources: Top 30 Big Data Tools for Data Analysis Top 30 Free Web Scraping Software

Big Data: 70 Free Data Sources You Should Know

1

big-data-70-amazing-free-data-sources-you-should-know-for-2017-152d33039f56

2018-09-04

2018-09-04 07:04:03

https://medium.com/s/story/big-data-70-amazing-free-data-sources-you-should-know-for-2017-152d33039f56

false

2,315

Knowledge base for data scientist

null

Become Data Scientist

null

become-data-scientist

DATA SCIENCE,DATA SCIENTIST,MACHINE LEARNING,BIG DATA

null

Open Data

open-data

Open Data

5,306

Octoparse

https://www.octoparse.com/

d229ddbf79fa

octoparse

101

7

20,181,104

null

0

null

0

null

2018-03-23

2018-03-23 07:50:12

2018-03-23

2018-03-23 08:27:42

4

false

tr

2018-03-23

2018-03-23 08:27:42

5

152eebdc23a9

7.250943

4

0

İnsan kaynakları bir şirketin çalışan profilini belirleyen, işe alım süreçlerini yürüten, seçilen adayların şirkete adapte olmasını…

5

Yapay Zeka Dokunuşu ile İnsan Kaynakları İnsan kaynakları bir şirketin çalışan profilini belirleyen, işe alım süreçlerini yürüten, seçilen adayların şirkete adapte olmasını sağlayan, çalışan memnuniyetini ölçen, kısaca personelin performansından ve veriminden sorumlu departman olarak tanımlanabilir. İş dünyasının teknolojiden en uzak kalmış kolu olan İK departmanında pazarlama veya satış kadar fazla olmasa da dijitalleşme başladı. Yapay zeka ile; işe alım-çıkarımlar, performans değerlendirmeleri, eğitimler ve oryantasyon süreçleri, kariyer yönetimi ve koçluk gibi İK işleri eski geleneksel yöntemlere göre, hem mali açıdan hem de zaman açısından çok daha verimli bir şekilde yönetiliyor. Peki ama nasıl? Yapay Zeka ile İşe Alım Süreçleri İşe alım günümüzde şirketlerin en çok zorlandığı konulardan birisi. Pek çok başvuru içinden pozisyona en uygun kişiyi bulmak, markanın başarısını sürdürebilmesi için önemli yapı taşlarından bir tanesi. Ayrıca pozisyona oturmayan, işe uygun olmayan kişiyi almak da fazlaca maliyetli. Dijitalleşme başlamadan önce şirketler işe alım sürecinde adeta başvurular tarafından boğuluyorlardı. Her başvurunun tek tek incelenmesi, seçilen adaylarla tekrar mülakat derken bir pozisyonu doldurmak dört ayı bulabiliyordu. Kariyer.net ve Linkedin gibi online iş başvuru siteleri kurulduktan sonra özellikle büyük markalar bu konuda ancak bir miktar rahatladılar. Yapay zeka işe alım sürecine ilk olarak filtreler ile adım attı. Basit filtreleme algoritmaları sayesinde adaylar saniyeler içerisinde belirlenen kriterlere göre (mezun olunan okul, iş ve sektör tecrübesi, yurt dışı deneyimi vb.) ayrılarak eskisi gibi bir ton başvuruyu tek tek el ile inceleme dönemi de sona ermiş oldu. Filtreleme sistemlerinin belki de en ileri seviyesi, Restless Bandit firmasının oluşturduğu Talent Rediscovery (yeniden yetenek keşfi) algoritmaları. Reckless Bandit ileri derecede filtreleme algoritmaları kullanarak şirketlere en uygun adayı sunmayı hedefleyen bir firma. Filtreleme algoritmalarının yardımıyla, şirketlerin mevcut pozisyonları ile eşleşen öz geçmişleri bulmak için bir şirketin başvuru izleme sistemini sürekli olarak takip ediyor. Daha sonra iş sahibine bulduğu potansiyel adayla iletişime geçmesi için e-posta gönderiyor. Bunların yanında adayların sosyal medya hesaplarındaki bilgilerden yola çıkarak, başvuru inceleme sistemini de sürekli güncel tutuyor. Son olarak, adaylara otomatik olarak e-postalar göndererek, hatta adayların Facebook sayfalarına reklam ekleyerek başvuruları sürekli teşvik ediyor. Buraya kadar aslında yapay zekanın aslında pek de oyuna dahil olmadığını söyleyebiliriz. Photo by Dylan Gillis on Unsplash Her ne kadar Restless Bandit firmasının algoritmaları ileri düzeyde olsa da, filtreleme algoritmaları içinde bulunduğumuz teknoloji çağın için çok ileri bir seviye sayılmaz. Yapay zekanın asıl nimetleri mülakat aşamasında karşımıza çıkmaya başlıyor. Yapay zeka algoritmaları tarafından oluşturulmuş chatbotlar artık mülakatlarda kullanılmaya başlandı. Chatbotu tanımlamak gerekirse, kişiyle genellikle belirli bir senaryo üzerinden iletişim kuran, sorular soran ve aldığı cevaplara göre konuşmayı sürdüren bir yazılım parçası diyebiliriz. Aslında bir nevi gelişmemiş sanal asistan da denebilir. İş görüşmeleri için tasarlanan bu chatbotlardan bir tanesi de Mya. San Francisco temelli bir start-up firması olan Mya Systems tarafından oluşturulmuş Mya, işe başvuran adaylarla online bir sohbet başlatıyor ve adayların istenilen seviyede olup olmadığını ölçen ufak bir sohbet gerçekleştiriyor. Konuşma bittikten sonra ise başvuru izleme sistemine adayla ilgili bir puan kartı ve transkript gönderiyor. Eğer bir aday Mya’yı etkilemeyi başardıysa, Mya bu sefer bir insan ile mülakat ayarlıyor ve insan kaynakları personelini bilgilendiriyor. Mya Systems, Fortune 500 listesinde yer alan pek çok büyük şirketin ve en büyük 5 işe alım ajansından 3’ünün Mya’yı kullandığını açıkladı. Mya’dan yararlanan markaların da işe alım sürecinde zamandan %70 tasarruf ettiğini ekledi. Ayrıca Mya türünün tek örneği de değil. HiringSolved firmasının oluşturduğu RAI de iş görüşmelerinde kullanılıyor. Bunların yanında JobPal, MyAlly, Debra gibi farklı farklı chatbolar da mevcut. Chatbotların yanı sıra yapay zekanın getirdiği başka bir yenilik ise video mülakat uygulaması. Utah merkezli bir start-up olan HireVue bu hizmeti veren firmalardan bir tanesi. HireVue işe en uygun adayı bulabilmek için mülakatı video kaydına alıp mülakat bittikten sonra, adayların mülakat sırasındaki jestlerini, mimiklerini, ses tonlarındaki değişimlerini ve mikro ifadelerini yapay zeka algoritmaları ile analiz ediyor. Mesela adayın mülakat sırasında belirli bir konuda yalan söyleyip söylemediğini, kişinin göz bebeklerinin büyüyüp küçülmesinden ya da o anda yaptığı bir hareketten yola çıkarak tespit edebiliyor. Aynı zamanda bu gibi analizlerle, mülakatlarda oluşabilecek ön yargılardan da kurtulmayı hedefliyor. Son beş dönemde 93 milyon dolar yatırım alan HireVue kendi bölgesinde bulunan 3 start-up’ı da bünyesine kattı. Vodafone, Nike, Intel ve Deloitte gibi dünya devi markalarla iş yapmakta olan HireVue, Unilever ile de 2016 senesinden beri beraber çalışıyor. Pymetrics ve HireVue ile 2016 yılından beri işe alım süreçlerinde beraber çalışan Unilever, bugüne kadar 250.000 başvuruyu yapay zeka teknolojileri ile değerlendirdi. Öncelikle Kuzey Amerika’da bu uygulamayı başlatan Unilever, daha sonrasında 68 ülkede 15 farklı dilde bu teknolojiden yararlandı. Adayların 20 dakika içerisinde 12 adet nöro bilim temelli oyuna sokulmasıyla başlayan eleme süreci, oyunu geçen adaylar pozisyonu alacak kişide istenilen kriterlere dayanan yapay zeka tarafından hazırlanmış bir mülakata girmesiyle devam ediyor. Dijitalleşme sayesinde önceki işe alımlara göre iki kat fazla başvuru alan Unilever, işe alım sürecini de dört aydan dört haftaya düşürdü ve bu sayede 50.000 saat tasarruf edildiğini açıkladı. İnsan kaynakları denildiğinde ilk ve en fazla akla durum işe alım olsa da, İK’nın görevleri sadece bununla sınırlı değil. Yapay zeka teknolojilerinden de sadece işe alım sürecinde değil, İK’nın diğer sorumluluk alanlarında da yararlanılıyor. Bu alanları inceleyerek devam edelim. “Several people fist bumping over a busy workspace” by rawpixel.com on Unsplash Yapay Zeka ile Oryantasyon Süreçleri Pozisyona en uygun ve en yetenekli kişiyi almak belki de insan kaynaklarının en öncelikli işi, fakat iş bununla sınırlı değil. Her yeni çalışanın şirketin işleyişiyle ilgili bilgilendirilmesi gerek. Yeni işe başlayan pek çok kişinin aklında ‘her çeyrekte ne kadar zam alınıyor?’ veya ‘izin kullanım prosedürü nedir?’ gibi, firmada işlerin nasıl yürüdüğüyle ilgili soruları bulunur. Bu soruları cevaplamak, çalışanı firmaya adapte etmek de insan kaynaklarının görevi. Tabi sadece işe yeni alınan personellerin soruları değil aynı zamanda değişen şirket politikalarını çalışanlara duyurmak ve personelin bu yeni kurallara uyum sağlamasını sağlamak da yine İnsan Kaynakları’nın görevi. Yapay zeka uygulamaları ile, şirketteki her pozisyon için farklı oryantasyon programları oluşturulmaya başlandı. Bu sayede, işe yeni başlayanlar hem fuzuli bilgilerle dolmuyorlar hem de İK personeli boşa vakit kaybetmemiş oluyor. Buna benzer bir teknolojiyi Ernst&Young danışmanlık şirketi kullanıyor. Dünya genelinde 250.000 çalışanı bulunan marka, işe yeni başlayanlar için bir ‘Onboarding Buddy’ adında bir mobil uygulama çıkardı. Mesela işteki ilk gününüz ve masanızda bilgisayarınızın olmadığını gördünüz. Uygulamayı açıp, yapay zeka ile programlanmış chatbota ‘laptopumu almak için kiminle görüşmeliyim?’ diye sorabilirsiniz. Uygulama size sadece kiminle görüşmeniz gerektiğini söylemekle yetinmiyor. Onun dahili numarasını, e-mail adresini, fotoğrafını hatta ona nasıl ulaşacağınızı gösteren bir şirket içi harita bile gönderiyor. Bu gibi uygulamalar sayesinde işe yeni başlayan çalışanın kolayca adapte olması sağlanırken, zamandan da yüksek seviyelerde tasarruf sağlanıyor. Peki çalışanı işe aldık ve şirkete uyumunu sağladık. Bir sonraki adım ne? Yapay Zeka İle Eğitimler ve Performans Değerlendirmeleri Çalışanların iş yerindeki performansı ve bu performansın nasıl arttırılabileceği İnsan Kaynakları’nın sorumluluk alanlarından bir diğeri. Her çalışandan olabildiği kadar fazla verim alabilmek ve çalışanlara verimi artırmak için eğitimler verilmesi markalar için oldukça önemli. Yapılan performans değerlendirmeleri sayesinde çalışanlar hangi alanlarda eksik olduklarını görebiliyor, ayrıca verilen eğitimlerle de bu eksikliklerin giderilmesi hedefleniyor.Teknoloji ile şirketler çalışanları hakkında daha önce hiç olmadığı kadar veriye sahipler. Fakat burada önemli olan bu veriye sahip olmak değil, bu verinin nasıl kullanılabileceği. Her çalışan birbirinden farklı alanlara daha yatkın olduğu gibi, zayıf olunan alanlarda çalışandan çalışana değişkenlik gösteriyor. Bu da her çalışanın, yöneltilmesi gereken alanın ve gidermesi gereken zayıflıkların farklı olduğunu ortaya koyuyor. Örneğin müşteriyle ilişkileri iyi olan fakat pazarlama alanında geri kalmış bir çalışanla, pazarlama alanında iyi fakat müşteri ile iletişimde sıkıntı yaşayan başka bir çalışana aynı eğitimi vermek yersiz olacaktır. Verilen eğitimin iki çalışana da pek bir faydası olmayacağı gibi, zamandan da kayıp yaşanacaktır. Verilen eğitimin yöntemi de fark yaratan etmenlerden bir diğeri. Kimi insan görsel olarak daha rahat öğrenirken, başka birisi işitsel öğrenmeye yatkın olabiliyor. Photo by Antenna on Unsplash Yapay zeka algoritmaları ile her çalışanın performansı ayrı ayrı değerlendirilerek ihtiyaç duyulan eğitimler kişiselleştirilecek. Bu sayede her çalışanın eksik olduğu konu ayrı ayrı saptanacağı ve verilecek eğitimin de çalışana göre şekillendirileceği konuşuluyor. Ofisteki çalışanları toplantı odasına toplayıp herkese aynı eğitimi aynı yöntemlerle verme dönemi yavaş yavaş geride kalacak gibi. Yukarıda bahsettiğimiz gelişmeler henüz tam anlamıyla yaşanmış olmasa da buna benzer hizmetler veriliyor. Bunlardan en popüleri geçtiğimiz sene Microsoft’un, kurumsal Office365 kullanıcıları için piyasaya sürdüğü Workplace Analytics adlı program. Çalışanların performanslarını ve verimlerini analiz etmek üzere yazılı bir algoritması bulunan Workplace Analytics, bunu personelin bilgisayar başındaki bütün aktivitelerini saniye saniye kayıt altına alarak yapıyor. Bunun yanında size kaynakları nasıl dağıtmanız gerektiği, çalışma alanı planlaması ya da ekip liderleri hakkında öneriler sunuyor. Personel gelişiminde başka bir sayfa ise çalışanlara koçluk ve hocalık yapan chatbotlar. GiantOtter isimli start-up, şirketlere koçluk ve eğitmenlik hizmeti veren chatbot ‘Coach Otto’yu yarattı. Coach Otto üst düzey yetkililerin, koç olarak çalışanlarına yardımcı olmasına aracı oluyor. Yine aynı firma tarafından yaratılan başka bir chatbot ise Coach TopGun SellFun. Coach TopGun SellFun ise çalışanlara satışla ilgili anahtar bilgilerin eğitimini veren başka bir chatbot. Her ne kadar şu an için İK’nın her alanında yapay zekadan tamamen yararlanıldığını söylemek güç olsa da, son birkaç sene içerisinde alınan yolu da göz ardı etmemek gerek. Ayrıca İK direktörleri de yapay zeka teknolojisinin yavaş yavaş kendi alanlarına girmesinden gayet memnun görünüyor. ServiceNow firmasının 350 üst düzey İK yöneticisi arasında yaptığı araştırmaya göre, yöneticiler istedikleri anda istedikleri bilgiye ulaşabildikleri için yapay zeka teknolojilerini çok yararlı buluyorlar. Bu sayede bürokratik süreci atlayarak zamandan tasarruf edildiğini düşünüyorlar. Yine aynı araştırmadan çıkan başka bir sonuç ise, yöneticilerin %92’si gelecekte personele verilen servislerin neredeyse tamamen yapay zeka üzerinden olacağını düşündüğü. Bir başka sonuç ise çalışanlarla iletişimin chatbotlar üzerinden olmasının bazı durumlarda aşırı rahatlık sağladığı. Örneğin iş yerinde cinsel istismar üzerine yapılan bir anketi, İK görevlisi yaparken o an alabileceği cevaplardan dolayı kendini rahat hissedemezken, chatbot üzerinden aynı anketi gayet rahat bir şekilde yapabiliyormuş. Bütün bu inovasyonların yanında, yapay zeka kendisiyle beraber bir takım ön yargıyı da beraberinde getiriyor. Çalışanların işlerini makinalara kaptıracağı korkusu bunların en başında. Ayrıca yeni teknolojilere adapte olamama da çalışanların bir diğer endişesi. Capital Group İK Başkan Yardımcısı Diana Wong’a göre ise çalışanlarda oluşan bu korku ve endişe yersiz. Wong’a göre insan kaynaklarında yapay zeka teknolojilerinin kullanımı ancak, çalışanları eğitimlerle yeni teknolojiler hakkında sürekli bilgilendirerek ve adapte ederek mümkün. Bu sayede hem çalışanlar yersiz endişelerinden arınmış olacaklar, hem de şirketin kullanacağı yeni teknolojilere uyum sağlayarak verimi düşürmeyecekler. Yazının başında da söylediğimiz gibi, iş dünyasının teknolojik olarak en geride kalan kolu insan kaynakları. Yapay zeka teknolojileriyle artık işe alım ve çıkarmalar, oryantasyonlar, çalışan performans değerlendirmeleri, iş yeri memnuniyeti, terfi, prim ve promosyon gibi insan kaynaklarının sorumluluğu altındaki konularda gelişmeler yaşanıyor. Fakat bunların hepsi için, henüz başlangıç seviyesinde olunduğu da bir gerçek. Yapay zeka üzerine çalışmalar arttıkça, insan kaynakları alanında da ilerlemeler olmaya devam edecektir. Yapay Zeka ve Makine Öğrenmesi ile ilgileniyorsanız hemen aşağıdaki formdan Haftalık Bültenimize abone olabilirsiniz. Yazıyı beğendiyseniz de bizi Medium üzerinden takip edip desteğinizi gösterebilirsiniz.

Yapay Zeka Dokunuşu ile İnsan Kaynakları

62

yapay-zeka-dokunuşu-ile-i̇nsan-kaynakları-152eebdc23a9

2018-04-15

2018-04-15 16:56:27

https://medium.com/s/story/yapay-zeka-dokunuşu-ile-i̇nsan-kaynakları-152eebdc23a9

false

1,736

null

Yapay Zeka

yapay-zeka

Yapay Zeka

502

Cerebro

Industry Leading Artificial Intelligence and Machine Learning Blog

5a7a330c87f6

cerebro.tech

331

9

20,181,104

null

0

null

0

863f502aede2

2017-12-09

2017-12-09 21:46:50

2017-12-09

2017-12-09 21:57:27

14

false

en

2017-12-09

2017-12-09 22:03:29

0

1531bbd4d404

5.721698

3

0

Introduction

1

Guided Alignment Training for Topic-Aware Neural Machine Translation Introduction NMT (Neural Machine Translation) were shown to achieve the state-of-art in many translation tasks, such as WMT news translation, IWSLT spoken language translation, E-commerce corpora translation etc. But the problem is that the attention mechanism in the NMT might be wrongly guided towards a bad alignment result. To solve this problem, the author wants to use a hard-attention (Viterbi alignment matrix) of the SMT (Statistical Machine Translation) to help learn a better alignment matrix in NMT during the training. Additionally, the author combines the topic specified categorical information to further bootstrap the overall translation accuracy. In summary, this paper makes two contributions: The author defines an alignment loss in order to learn a better global attention The author aims to learn a topic vector based on the meta information to help decoding process We will first review the proposed loss function, which is a L2 distance between each term of Viterbi alignment matrix (hard-alignment matrix) and NMT alignment matrix (soft-alignment matrix). Then, we will review the topic-aware decoder for accuracy bootstrapping. Finally, I will give my own thoughts about the paper. Alignment Loss Motivation Because the attention weights only rely on the previous generated word and decoder state, the model cannot capture any additional information if the previous word is an placeholder or out-of-domain word/character (For E-commerce data it is a very common case), which might lead to a misalignment. In order to solve this problem, the author proposed that the Viterbi alignments of IBM model 4 can be used as a hard attention to bias the learned soft attention. Mathemetical formula The goal is to optimize the decoder cost and alignment cost (divergence between soft alignment and hard alignment generated by statistical alignments). At first, the decoder cost can be written as: y_n, x_n is the n-th training pair, and N is the number of samples. The above formula is the definition of negative log-likelihood, the conditional probability can be further written as: This formula shows: the probability of the target sentence, given the source sentence, is equal to the cumulative product of the probability of current word conditioned on the previous generated words and an context vector. T here is the length of the target sentence, and T’ is the length of source sentence, the context vector c is used to capture the coverage information of source sentence. The probability of current word conditioned on the previous sentence and context vector can be regarded as a function based on the current decoder hidden state s_t, previous generated word y_t and an context vector c. Generally g(.) is a non-linear function. If the context vector c is variable based on different step t, then c_t is used instead. c_t can be regarded as a weighted sum over the source sentence: where T’ is the length of source sentence, h_i is i-th encoder hidden state, a_ti is a weight to capture the relative importance of t-th target word and i-th source word at step t. a_ti can be computed as following: where each a_ti is a softmax of e_ti, e_ti = a(s_{t-1}, h_i), a(. , .) is a function to calculate the relative score, s_{t-1} is the decoder hidden state at step t-1, h_i is the i-th source hidden state (it can also be referred to a i-th bi-directional source hidden state). In this paper the author chooses dot product to compute the relative score: Overall, for T words on the target side, each of them needs to compute a context vector with fixed length T’, the author referred the resulting alignment matrix of shape TxT’ to matrix alpha. The pretrained statistical alignments matrix A is also a matrix with shape TxT’, where A_ti refers to the probability of the t_th word on the target side being aligned to i_th word on the source side. The matrix is normalized along the column in order to make it consistent with the neural alignment matrix. After the above steps, the author defines the alignment loss to be: Basically the Loss can be defined in the form of either cross-entropy or mean squared error. After the combination of alignment loss and decoder cost, the overall formula can be written as: where w1 and w2 are just coefficient to balance these two terms. The above table is the result compared to baseline NMT and baseline NMT + alignment loss, we can know that no matter using alignment loss in the form of mean-squared error or cross-entropy, the translation result always get better. The above figure show that how alignment loss helps the overall attention to become better, as we can see, e.g. in the upper figure, the French word “dolcevita” was aligned to the English word “federico” before (which is a wrong alignment), it is now correctly aligned to “dolcevita”. Topic Aware Decoder In the e-commerce domain, the product category may reflect useful information to help product title translation and product description translation. How to help decoding based on the topic vector? The idea is to represent the topic information in a D-dimensional vector l, where D is the number of topics. The conditional probability during the decoding can be written as: where g(.) is a approximate function which can be modeled by a feed-forward network: As shown in the above figure, which is a feed forward network to model the decoder. l is the topic vector of dimension D, f_{t-1} is the previous word embedding, c_t is the current context vector and s_{t-1} is the last decoder hidden state, where r_t can be written as: where W_r’ is the original transformation matrix and Wc is the topic transformation matrix, Ec can be regarded as a learned topic vector and can use the meta information to help decoding. As the figure shown above, the learned topic vector shows that it helps decoding from 18.6 BELU score to 19.7, it shows that decoding conditioned on an additional topic vector affects alignment, word selection and decoding search, etc. Some Thoughts from the Reviewer On the one hand, This paper uses IBM model 4 Viterbi alignments for guided alignment, it’s also possible to design kinds of hierarchical attention supervision to improve the guided alignment. Moreover, as mentioned in the conclusion of this paper, not only topic meta information could influence the overall translation performance, the monolingual data could also help. Recently, people are always using monolingual data for back-translation to improve the accuracy, it is also possible to investigate decoding conditioned on abundant monolingual data and its seep-up solutions. Technical Analyst: Shawn Yan | Reviewer: Hao Wang

Guided Alignment Training for Topic-Aware Neural Machine Translation

4

guided-alignment-training-for-topic-aware-neural-machine-translation-1531bbd4d404

2018-05-01

2018-05-01 17:14:12

https://medium.com/s/story/guided-alignment-training-for-topic-aware-neural-machine-translation-1531bbd4d404

false

1,132

We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

null

SyncedGlobal

null

SyncedReview

global.sns@jiqizhixin.com

syncedreview

ARTIFICIAL INTELLIGENCE,MACHINE INTELLIGENCE,MACHINE LEARNING,ROBOTICS,SELF DRIVING CARS

Synced_Global

Machine Learning

machine-learning

Machine Learning

51,320

Synced

AI Technology & Industry Review - www.syncedreview.com || www.jiqizhixin.com || Subscribe: http://goo.gl/Q4cP3B

960feca52112

Synced

8,138

15

20,181,104

null

0

null

0

1d7458f09e44

2017-11-08

2017-11-08 08:56:59

2017-11-10

2017-11-10 14:27:30

1

false

en

2017-11-10

2017-11-10 14:27:30

1

153216853e8

4.818868

1

0

Creating daily updated KPIs and dashboards from large clients & orders data sets (or any star schema dataset) is not an easy project. It…

5

Analyze clients & orders with a unified data platform Creating daily updated KPIs and dashboards from large clients & orders data sets (or any star schema dataset) is not an easy project. It usually requires an ETL (like Talend), a data-storage (like AWS Redshift) and a dataviz tool (like Tableau). Having to set up and maintain these three modules in production requires technical skills and create opportunities for errors and delays. In this article, I’ll show you how this kind of data project can be highly simplified using our unified data platform: Serenytics. Note that for a one-shot analysis, a data analyst/scientist could also use low-level tools (i.e. tools like Python, Pandas and Jupyter notebooks). But it requires programming skills, is usually time consuming (mostly reading Pandas documentation and StackOverflow) and is limited to small datasets. And most importantly, it is not adapted to KPIs/dashboards that should be updated daily and shared. The input data For the purpose of this article, we’ll use an ORDERS-CLIENTS model: The ORDERS dataset contains one row by order (from an e-commerce website or from a physical store) with columns such as date, price, client-id. The CLIENTS dataset contains one row by client with columns such as client-id, inscription-date, birthdate, country. And the final goal is to compute KPIs like: Revenue by country, by age, by month. Monthly growth of number of clients by age (and age is not a column of the dataset, let’s say that only birthdate is available) Number of customers who are new churners this month Number of customers who are re-activated this month (i.e. did not buy for the last 12 months, but bought in the previous month) Client segmentation per number of orders (1 order , 2–5 orders, 6–10 orders…) We could have chosen to work with any CLIENTS & INTERACTIONS data (e.g. opens & clicks data from marketing campaign tools like Adobe Campaign, Selligent…) or any data with a star schema. The following steps would be very similar. Loading the data Let’s say we have one CSV file for the CLIENTS and another one for the ORDERS. Within Serenytics, we just need to create one BigCSV data source per file and load the associate CSV files. As a result, the data is now stored within our internal data-warehouse (i.e. in AWS Redshift). And this also works for large files (we’ve done projects with 200M orders). Joining ORDERS and CLIENTS in a single data source To create a table that contains columns from both ORDERS and CLIENTS, we have to create a join between the two data sources we just created. To achieve this, we create a new data source of type Join, selecting the two initial data sources and the keys: id column for the CLIENTS dataset client-id for the ORDERS dataset (using an inner-join) As a result, we obtain a data source we can already use to create a dashboard with basic KPIs (revenue by country, revenue by month, number of clients by month…). But that’s not enough for more advanced KPIs. Note that in the next paragraphs, we’ll refer to this join table as ORDERS_CLIENTS_JOIN. Enriching your data From the ORDERS_CLIENTS_JOIN data source, using the Formulas tab, we’ll add new columns required in our analysis, without writing SQL. For this example, we’ll add the following columns: age defined as a simple formula: date_diff([client.birth_date], now(), "year") age-bin defined as a conditional formula: IF [age] <=18 THEN "0-18" ELSE IF [age] <= 30 THEN "19-30" ELSE IF [age] <= 40 THEN "31-40" ELSE IF [age] <= 50 THEN "41-50" ELSE "51-100" is_order_in_last_12_months_before_previous_month defined as a conditional formula: IF [order.date] >= month(-13) and [order.date] < month(-1) THEN 1 ELSE 0 is_order_in_previous_month defined as a conditional formula: IF [order.date] in month(-1) THEN 1 ELSE 0 growth_clients_creation_YtoY defined as a value formula: (count_if([client.id], [client.date-creation] in year(0)) - count_if([client.id], [client.date-creation] in year(-1))) / count_if([client.id], [client.date-creation] in year(-1)) Note that for this last formula would be more readable by using two intermediate formulas: nb_clients_created_last_year and nb_clients_created_this_year. Computing data per client From the above steps, we obtained a data source with one row per order. But let’s say we want to create a bar-chart with the number of re-activated customers (i.e. who did not buy for the last 12 months, but bought in the previous month) by age-bin. We need a table with one row per client and with a column containing the information is_reactivated_or_not. To create this new table in the data-warehouse, in the Automation menu, we create an ETL step with these settings: Input data source: we select the ORDERS_CLIENTS_JOIN source created previously Output source: let’s name it COMPUTED_INFO_BY_CLIENT Group-by dimensions: we group by client.id to have one row by client In the Data section, we select all the metrics we want to aggregate by client: sum(is_order_in_last_12_months_before_previous_month) We rename this aggregate to nb_orders_in_last_12_months_before_previous_month sum(is_order_in_previous_month) We rename this aggregate to nb_orders_in_previous_month To actually create the new table, in the Execution tab, we click on Run Now. In the created COMPUTED_INFO_BY_CLIENT data source, we can now add a conditional formula is_reactivated_in_previous_month: IF [nb_orders_in_previous_month]>0 and [nb_orders_in_last_12_months_before_previous_month]==0 THEN 1 ELSE 0 In some situations, we also want to pass raw client information from the input join to the output table (e.g. the client country). To achieve this, the best option is to add this column in the group-by dimensions selected in the ETL step. Another identical solution is to add it as a metric and use the min aggregation. Creating the dashboard We now have 4 data sources to use in our project: ORDERS: one row by order CLIENTS: one row by client orders_clients_join: one row by order, with columns about orders and clients computed_info_by_client: one row by client with computed columns Within the dashboard editor, it’s now very easy to compute many KPIs from these 4 data sources. Automating the process The data loading explained in this article is manual. In a production environment, we would schedule it (e.g. loading the CSV files from a SFTP server each night). This can be done with a few clicks in Serenytics. We would also: Add a data cleaning step right after the loading. Automatically reload the dashboard cache each night so that each morning, the dashboard is immediate to load. These Automation steps (loading + cleaning + computing intermediate tables + cache reloading) would be launched within a single Flow task, sending an email in case something went wrong. Each Monday morning, the dashboard would be sent by email as a pdf file. Conclusion In this article, we’ve seen all the steps to achieve a clients-orders analysis with Serenytics. Everything is done within a single Cloud application, without any tool to setup and without any programming/SQL skills. Of course, one can achieve the same result with a multi-tools stack (e.g. Talend/AWS Redshift/Tableau) and would benefit from the advanced features of each module. But that comes with a lot more complexity and only people skilled in the full stack can achieve the project (or a team of experts with at least one data analyst and one devops). And for most cases, the total cost of the project would be much lower with a unified data platform such as Serenytics.

Analyze clients & orders with a unified data platform

1

analyze-clients-orders-with-a-unified-data-platform-153216853e8

2018-05-09

2018-05-09 15:09:35

https://medium.com/s/story/analyze-clients-orders-with-a-unified-data-platform-153216853e8

false

1,224

Cloud Unified Data Platform (data-warehouse + data-wrangling + data-viz in a single app)

null

Serenytics

adrien.auclair@serenytics.com

serenytics

ANALYTICS,DATA SCIENCE,DATA VISUALIZATION,DATA ANALYSIS,DASHBOARD

serenytics

Big Data

big-data

Big Data

24,602

Adrien Auclair

Serenytics Founder - Planorama Founder- PhD in Computer Vision - Entrepreneur & coder

bc5107f35efd

AdrienAuclair

57

148

20,181,104

null

0

null

0

282aaf41e776

2018-03-22

2018-03-22 21:34:57

2018-04-19

2018-04-19 12:32:20

0

false

en

2018-04-19

2018-04-19 12:32:20

0

15329cef2b24

1.003774

0

Artificial Intelligence has many benefits and drawbacks. It can advance society to new heights with its support for manufacturers and…

1

Blog Post #4 Artificial Intelligence has many benefits and drawbacks. It can advance society to new heights with its support for manufacturers and inventors. It also becomes too much like human life and the theories that AI will take over start to become closer to truth than fear. A robot has recently been developed to feel emotions, if for no other reason than to prove that she can. Robots and AI’s are devoid of opinions that plague the planet. They do not fight over what is right and what is wrong and in this way, they are more logical and practical. Then again, most people invent and create because they have emotions about it and passion for it. In creating a feeling robot, humans could potentially be eradicated. Chiang sights the word “insight”. What it means and what it stands for in AI. ?He then goes on to say that the strawberry-picking-AI does not have it. He is able to explain something and use it as a contradiction against the technology he is using. He is framing an argument based on ethics. Newman discusses how Siri benefits her child with autism who is full of questions. He talks to her all the time about planes, trains, escalators, busses, and the weather. Siri gave him a world to look into. Newman noted that she never thought he thought about love or marriage and one night he asked Siri to marry him. She respectfully declined his proposal. Although they won’t be wed, Siri truly helps Gus learn about the world and things he is interested in.

Blog Post #4

0

blog-post-4-15329cef2b24

2018-04-19

2018-04-19 12:32:22

https://medium.com/s/story/blog-post-4-15329cef2b24

false

266

An ever-evolving repository for insight, wisdom, musings, critiques, and call-outs.

null

e110oneohfive

null

rosse110oneohfive

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Matt Leicht

null

dace1ffbbf7f

mattjoeleicht

0

2

20,181,104

null

0

null

0

null

2018-03-05

2018-03-05 06:55:42

2018-04-28

2018-04-28 03:28:33

5

false

bn

2018-06-23

2018-06-23 16:09:47

9

1532f57cee49

4.078616

23

1

0

শিক্ষানবিসদের জন্য মেশিন লার্নিং । মেশিন কে শিখান যা আপনি শিখতে চান!!

5

Deep Dream Style সম্ভাবনাময় মেশিন লার্নিং শুরুর কিছু পথ-নির্দেশিকা শিক্ষানবিসদের জন্য মেশিন লার্নিং । মেশিন কে শিখান যা আপনি শিখতে চান!! মেশিন লার্নিং ধীরে ধীরে বিভিন্নভাবে আমাদের জীবনে ছড়িয়ে পড়েছে। উদাহরণস্বরূপ, Netflix থেকে Spotify এবং ভিডিও থেকে আপনার নতুন প্রস্তাবিত সঙ্গীত মানুষ সব সময় মেশিন লার্নিং নিয়ে কথা বলছে এবং টিভি শো, খবর বা এমনকি অ্যানিমেশনেও আপনি বেশ কয়েকবার এ সম্পর্কে শুনেছেন। এটি অনেক কঠিন সমস্যাগুলির জন্য একটি প্রশান্ত মহাসাগর বলে মনে হচ্ছে (অবশ্যই, এটি নয় 😉) তবে আপনি কি সত্যিই এটি সম্পর্কে বিভ্রান্ত ?এখানে মেশিন লার্নিংয়ের একটি সহজ দ্রুত পরিচয় দেয়া হবে। আশা করি এই লিখাটি আপনাকে একটি মৌলিক ধারণা পেতে সাহায্য করবে । পূর্বশর্ত ১.তোমার ব্যাকগ্রাউন্ডঃ বর্তমানে যারা মেশিন লার্নিং শিখতে অনেক আগ্রহী তাদের জন্য কিছু পূর্বশর্ত থাকা দরকার আমি মনে করি।যেমনঃ ১. পাইথন প্রোগ্রামিং এর উপর জ্ঞান থাকলে এই জ্ঞান আপনার জন্য সহায়ক হবে কিন্তু যদি না থাকে তাহলে আপনি জ্ঞান নিয়ে আসেন তারপর এইটি শুরু করেন। ২. মেশিন লার্নিং এর সব গণিত সম্বলিত, আপনাদের মধ্যবর্তী স্তরের জ্ঞান থাকতে হবে রৈখিক বীজগণিত, পরিসংখ্যান, এবং মাল্টিভিরেবল ক্যালকুলাস এগুলো এর উপর। ২. টুলস এন্ড এনভায়রনমেন্ট: সেটআপঃ দুইটি উপায়ে আমাদের এনভায়রনমেন্ট প্রতিস্থাপন করতে পারিঃ ১. Install python and then install each package using pip command বা ২. Install Anaconda. আইডিইঃ পাইথনের জন্য অনেক আইডিই আছে আপনি নিম্নলিখিত যেকোনোটি ব্যবহার করতে পারেন, যেমন: ১. সাব্লিমেট টেক্সট ২. পাইচার( PyCharm) ৩. আইডেল ইন্টারনেটে অনেকগুলি আইডিই রয়েছে কিন্তু আমরা আইপাইথন নোটবুক ব্যবহার করব। নিম্নলিখিত কমান্ড টার্মিনাল তে চালান: > jupyter notebook Terminal output after running the command সার্ভার স্থানীয়ভাবে শুরু হবে এবং আপনি পোর্টে নোটবুক অ্যাক্সেস করতে পারবেন localhost:8888 আপনার বর্তমান ডিরেক্টরীতে একটি নতুন ফোল্ডার খুলুন এবং তারপর সেই ফোল্ডারে একটি নতুন আইপাইথন ফাইল তৈরি করুন। একটি পাইথন পরিবেশের জন্য আপনাকে এটি করতে হবে। আসুন কোডটি নীচের লিংকের আইপাইথন সেল থেকে কপি করে কোড চালানোর জন্য Shift + Enter চাপুন। https://github.com/harunshimanto/Learn-Machine-Learning-in-3-month/blob/master/Machine%2BLearning%2Bin%2BPython%2BStep-By-Step%2BTutorial%2B%20(1).ipynb Output of the code এই কোড এর মাধ্যমে আপনার এনভায়রনমেন্ট চেক করে নিলেন ঠিকভাবে কাজ করছে নাকি। আমাদের প্রায়শই যেসব লাইব্রেরি ব্যবহার করতে হবেঃ ১.Scikit-Learn: মেশিন লার্নিং এর জন্য পাইথন , বাস্তবায়ন সঙ্গে আসছে অনেক সাধারণ মেশিন লার্নিং আলগোরিদিম ( যেমনঃ ডিসিশন ট্রি, Naive Bayes, SVM, K-Nearest Neighbors ইত্যাদি ) এবং অন্যান্য অপারেশন জন্য মডিউল( যেমনঃ মেট্রিকস, ক্রস ভ্যালিডেশন, ফিচার সিলেকশন ইত্যাদি) ২.পান্ডাস : ডাটা ফ্রেম হিসাবে ডাটাসেটগুলির সাথে কাজ করার জন্য পান্ডাস বেশ উপযোগী এবং ডাটা বিশ্লেষণের উপরে, ডাটা ট্রান্সমেশনগুলির উপরে কাজ করে। ৩. Numpy :এটা সংখ্যাসূচক গণনা জন্য বেশ ভাল, এটি n- মাত্রিক অ্যারে গুণ করার জন্য বেশ ভাল এবং অনেক বিল্ট ইন গাণিতিক অপারেশন আছে । ৪. Matplotlib: এটি visualization( লেখচিত্র দ্বারা ফুটে তোলা) এর ব্যবহার হয়। গুগলের সিইও সুন্দর পিচাই গুগল আই /ও (i/o)২০১৬ এ ঘোষণা দিয়েছিলেন যে, আমরা “মোবাইল ফার্স্ট” থেকে “এআই ফার্স্ট” এ স্থানান্তরিত করছি এবং গুগল এআই কোম্পানী হওয়ার কৌশলটি নতুন করে সংজ্ঞায়িত করছে। মেশিন লার্নিং কি? মেশিন লার্নিং কম্পিউটার বিজ্ঞানের একটি ক্ষেত্র যা কম্পিউটারকে স্পষ্টভাবে প্রোগ্রাম করা ছাড়া শেখার ক্ষমতা দেয়।তারপর তারা সমস্যার সমাধান করতে ভবিষ্যতের আচরণ, ফলাফল এবং প্রবণতা পূর্বাভাসের জন্য বিদ্যমান তথ্য থেকে তারা শিখেছেন কি প্রয়োগ করতে পারেন।উদাহরণস্বরূপ, একটি অ্যালগরিদম বিড়ালগুলি সনাক্ত করতে বিড়ালদের ফটো থেকে প্রশিক্ষিত হতে পারে; একই অ্যালগরিদমে একটি লাইন কোড পরিবর্তন ছাড়াই সাইকেল চিনতে সাইকেল এর ফটোগুলির সঙ্গে প্রশিক্ষিত করা যেতে পারে। মেশিন লার্নিং নামটি আর্থার স্যামুয়েল দ্বারা 1959 সালে উদ্ভূত হয়েছিল। এখানে মৌলিক ধারণা হল, নতুন তথ্য প্রকাশ করার সময় মেশিন লার্নিং কম্পিউটারের প্রোগ্রামগুলি নিজেদেরকে শেখাতে পারে। উদাহরণঃ স্পষ্টভাবে মেশিন লার্নিং আজকাল অ্যাপ্লিকেশন সঙ্গে যোগাযোগ পদ্ধতি পরিবর্তন করছে , আমরা যেমন উদাহরণ দেখতে পারছি: ১. ফেসবুক বা আমাদের ফোনে ফেস ডিটেকশন। ২. গুগল ব্যবহারকারীদের আচরণের উপর ভিত্তি করে দেখাচ্ছে বিজ্ঞাপন এবং বিষয়বস্তু । ৩.ইমেল স্প্যাম বা না হয় কিনা তা সনাক্ত করতে ইমেল দ্বারা কাজ করা; ৪.আমাজন বা নেটফ্লক্সে চলচ্চিত্র এবং বিভিন্ন পণ্যের সুপারিশ; ৫. স্ব-ড্রাইভিং কার আসছে … সমস্যা এবং কাজের ধরনঃ মেশিন লার্নিং এর কাজের সাধারণ সিস্টেম শেখার জন্য উপলব্ধ “সংকেত” বা “প্রতিক্রিয়া” শেখার প্রকৃতির উপর ভিত্তি করে, তিনটি বিস্তৃত বিভাগে শ্রেণীবদ্ধ করা হয়। এইগুলো: ১. সুপারভাইজড লার্নিংঃ কম্পিউটারটি একটি “শিক্ষক” দ্বারা প্রদত্ত উদাহরণ ইনপুট এবং তাদের পছন্দসই আউটপুটগুলির সাথে উপস্থাপিত হয়, এবং লক্ষ্য হল একটি সাধারণ নিয়ম শিখানো যা আউটপুটগুলিতে ইনপুটগুলির ম্যাপ করে। ২.আনসুপারভাইজড লার্নিংঃ আনসুপারভাইসড লার্নিং এ প্রোগ্রামকে কিছু ডেটা দেওয়া হয়। প্রোগ্রাম ঐ ডেটার উপর নির্ভর করে ডিসিশন দেয়। যেমন এক ঝুড়ি ফল রয়েছে। প্রোগ্রাম ভিন্ন ভিন্ন ফল কে ভিন্ন ভিন্ন ক্যাটেগরিতে ভাগ করবে, এটা হচ্ছে আনসুপারভাইসড লার্নিং এর উদাহরণ। ৩. রেইনফোর্সমেন্ট লার্নিংঃএকটি কম্পিউটার প্রোগ্রাম একটি গতিশীল এনভায়রনমেন্ট সাথে ইন্টারেক্ট করে যেখানে এটি একটি নির্দিষ্ট লক্ষ্য (যেমন একটি গাড়ির ড্রাইভিং বা প্রতিপক্ষের বিরুদ্ধে একটি খেলা খেলছে) সম্পাদন করে ।প্রোগ্রামটি পুরস্কার এবং শাস্তিগুলির ক্ষেত্রে প্রতিক্রিয়া প্রদান করে কারণ এটি তার সমস্যা স্থানটি নেভিগেট করে। এই সিরিজ আমরা যেমন কিছু সুপরিচিত শেখার অ্যালগরিদম একটি কটাক্ষপাত করতে যাচ্ছি: ১.ডিসিশন ট্রি; ২.লিনিয়ার রিগ্রেশন; ৩.লজিস্টিক রিগ্রেশন; ৪.SVM(সাপোর্ট ভেক্টর মেশিন); ৫.Naive Bayes; ৬.KNN (ক-নেয়ারেস্ট নেইবার্স); ৭.রান্ডম ফরেস্ট( Random Forest); ৮.নিউরাল নেটওয়ার্কস(Neural Networks); এই সিরিজে আমার লক্ষ্য হল মেশিন লার্নিং আলগোরিদমগুলির মধ্য দিয়ে যাওয়া, মেশিন লার্নিং ওয়ার্কফ্লোতে ব্যবহৃত কিছু অন্যান্য কৌশল এবং আপনাকে দেখায় যে কিভাবে বাস্তব সমস্যাগুলিতে এটি প্রয়োগ করা যেতে পারে। আরও পড়ুন: A Tour of Machine Learning Algorithms — By Jason Brownlee. পরবর্তী: পরবর্তী নিবন্ধে আমরা ডিসিশন ট্রি এবং ক্লাসিক আইরিস ডেটাসেট সম্পর্কে আলোচনা করতে যাচ্ছি (প্রকৃতপক্ষে এটি আমাদের মেশিন লার্নিংয়ের জন্য হ্যালো ওয়ার্ল্ড হবে)। আপনি এই সম্পর্কে কি মনে করেন আমাকে জানাতে, যদি আপনি লেখা উপভোগ করেন তাহলে নীচের ❤ ব্যবহার করুন । Happy learning.

সম্ভাবনাময় মেশিন লার্নিং শুরুর কিছু পথ-নির্দেশিকা

169

সম্ভাবনাময়-মেশিন-লার্নিং-শুরুর-কিছু-পথ-নির্দেশিকা-1532f57cee49

2018-06-23

2018-06-23 16:09:47

https://medium.com/s/story/সম্ভাবনাময়-মেশিন-লার্নিং-শুরুর-কিছু-পথ-নির্দেশিকা-1532f57cee49

false

860

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Harun-Ur-Rashid(Shimanto)

Research wing at DIU CPC - Daffodil International University Computer and Programming Club

f9de6603f07e

harunshimanto

279

479

20,181,104

null

0

null

0

59952791b3ab

2018-06-17

2018-06-17 13:40:08

2018-06-20

2018-06-20 15:00:46

12

false

en

2018-06-20

2018-06-20 15:44:58

6

15354f63eeb

9.085849

3

0

Introduction and objectives: In this article I will focus on asset allocation strategies and look closely at portfolio performance. As I…

5

Benefits of diversification: analysis of efficient frontiers for crypto and traditional markets. Photo by Melanie Magdalena on Unsplash Introduction and objectives: In this article I will focus on asset allocation strategies and look closely at portfolio performance. As I continue to train my skills I recommend this article only for educational purposes. The idea of this research is to visualize different efficient frontiers to be able to see the trade-off of adding more assets in the portfolio and from grouping assets in the portfolios by level of correlation and market capitalization. All in all, having the same beliefs about future returns, variances and covariances, build different optimised and random-weighted efficient frontiers. I expect, that having the same bias in estimation we could compare efficient frontiers at least relatively rather than in absolute values. Overall, I will compare the benefits of active portfolio management versus passive indexing strategy in crypto and traditional markets (assuming zero transaction costs). For crypto world the low correlated assets are more volatile in general. So at the end, I present the trade-offs of diversifying a portfolio with low correlated assets. Indeed, my findings are inline with MPT in that portfolio variance can be minimised with increasing number of assets. Data: Download list of S&P 500 companies from Wikipedia and use Quandl database to retrieve historical prices for 505 companies. The crypto series are obtained by using the Coinmarketcap API. The total amount of active cryptocurrencies as of June 2018 was 1628 (1245 with historical prices), while, roughly a quarter has the longevity more than 2 years. Therefore, I end up with 247 coins available for the analysis and further filter some items out because of intermediate missing values. To overcome this issue with missing values maybe is good to use a bootstrapping. Table 1: Combination of portfolios by level of correlation and their descriptive statistics. As we may notice the correlation during this period is not high for the high correlated group. Below I present another dataset grouped by marketcap quantiles and further picking 20 assets randomly from each quantile. It shows that 10% percentile of marketcap distribution has very low correlation, while above 90% has stronger correlation — that is alike bitcoin and bitcoin itself with high market share. To be precise, bitcoin is not in the group of 20 randomly picked coins: Table 2: Combination of portfolios by marketcap size and their descriptive statistics. Overall, we observe that small marketcap group (and here low correlated group) is more volatile than high market cap group and that it is compensated for with higher reward. Methodology: I apply the Markowitz mean-variance investment rule to define an efficient frontier. As cited from Markowitz: “the rule serves better as an explanation of, and, guide to, ‘investment’ as distinguished from ‘speculative’ behaviour”. The expectations of returns and variances are based on 120 days of historical sample from January 2017 to September 2017 (this size is chosen for crypto series with limited information). The weights are fixed without rebalancing for the next 2 months as a ‘buy and hold’ strategy. As a result, I will measure the cumulative performance of all strategies and see which portfolio outperforms the others and the market and evaluate different risk metrics as trade-offs. In particular, I will consider the performance of the Sharpe ratio maximisation portfolio and min variance portfolio solving the mathematical optimization problem and compare them with max Sharpe ratio and min variance portfolios chosen from 25 000 random weighted portfolios and equally weighted portfolio. In addition, I compare all portfolios against the market benchmarks: such as S&P 500 index for traditional market and market cap-weighted index of 100 cryptocurrencies for crypto market. Technically, there are already a lot of materials for building the portfolios and efficient frontiers for traditional assets in Python, so I am going to apply the same approach for crypto (blog 1, blog 2). Efficient frontier: 1952, Portfolio Selection by Harry Markowitz — “at one end of the efficient set is the point of minimum variance, at the other end is the point of maximum expected return”. He highlights that diversification could not eliminate portfolio variance if the securities are highly correlated and exposed to market risk. He rejects the hypothesis that investors should only maximize the expected return. In the crypto speculative world, it may look attractive to invest based on high expected returns irrespectively to variance. For example, below are some coins that show outstanding values of their average daily returns by comparison with others: Table 3: Example of mean daily crypto returns As I will plot not only optimised portfolios, but also portfolios formed by random weights allocation, the random portfolios form a parabolic shape and the border of the shape will be an efficient frontier in that case as shown in Markowitz paper: Figure 1: Efficient frontier surface from Markowitz paper First, let’s build efficient frontiers for stocks from S&P500 index. Randomly pick 30 stocks with low correlation and slice them into portfolios with 2, 3, 4, 5, 10, 15, 20, 25, 30 stocks. Then for every group generate 25 000 random sets of asset weights, compute their portfolio return and variance and draw a point in the chart. From that randomly generated portfolios, we find out the portfolio with minimum variance and portfolio with maximum Sharpe ratio. Another story is to apply portfolio optimisation, where we define the target range of returns and with the small step run the optimization program that minimizes the desired function: here — either minimum variance or maximum Sharpe ratio. Note that every point on the chart below has particular set of weights. The weights are bounded but not limited: 0–100%. Figure 2: Mean-variance optimised efficient frontier, max Sharpe ratio and min variance optimised portfolios and max Sharpe ratio and min variance of random portfolios From figure 2 we observe that with increasing number of assets, the portfolio variance significantly decrease. The second moment is that optimised strategy delivers lower minimum variance and higher Sharpe ratio by comparison with random portfolios for 10 and more assets. Next, as we define that optimised frontier is more ‘efficient’ than random frontier, we build optimised frontiers with regard to the different coefficient of correlation between assets. To achieve this, we compute the correlation matrix for all 500 stocks and randomly pick 30 stocks with correlation higher than 0.5 and lower than 0.2. Figure 3 represents the efficient frontiers for this portfolios: Figure 3: Optimised efficient frontiers for low and high correlated stocks and random stocks portfolios From picture 3, we confirm that the efficient frontier for highly correlated stocks is more risky, however, the risk is compensated with higher returns. Here we observe the benefits of low correlation asset diversifications which are translated into the lower minimum variance and higher Sharpe ratio for all portfolios. To evaluate which portfolio outperforms let’s have a look at cumulative performance as we buy stocks in September 2017 and hold them till December 2017: Figure 4: Cumulative return during 2 months of holding the portfolios: optimised max SR portfolio, optimised min vol portfolio, equally weighted (naive) portfolio and cumulative performance of market index S&P500. From Figure 4 we observe that neither equally weighted portfolio, nor max Sharpe ratio optimised portfolio consistently outperform the market (S&P 500 index) with low correlated assets in their portfolios. In contrast, high correlated stocks skyrocket their performance when invest according to the maximum Sharpe ratio maybe because the market was low volatile. The coherence with efficient frontiers and MPT in figure 3 is that low correlated stocks show lower variation. More details about results is hidden in correlation analysis of holding period as the correlation coefficients go down to 0,13 for high correlated stocks. Now let’s see how the findings and Modern Portfolio Theory persist in the crypto market. For this purpose, we randomly pick 30 stocks from 25–75% percentile of the marketcap distribution. From table 2 we already know that the average pairwise correlation of this group is 0.16 which states that this group consist of low correlated assets. Below, let’s see the evolution of efficient frontier with increasing number of assets in the portfolio: Figure 5: Mean-variance optimised efficient frontier, max Sharpe ratio and min variance optimised portfolios and max Sharpe ratio and min variance of random portfolios in the crypto market From figure 5, mainly, we observe the benefits of diversification as the variance decreases with adding more assets in the portfolio. The second finding is that optimisation doesn’t show any improvements and the optimised efficient frontier is very close to the random efficient frontier. Overall, the portfolio variance for crypto portfolios is 10 times higher in comparison with equity portfolios. The efficient frontier surface is narrowed. Weights are allocated randomly, so there is no case with 100% weight in one coin in random strategy, whilst this could be the case in the optimised strategy as long as weights are not restricted. As a result, I get this minimum variance optimised portfolio outstanding in the left bottom corner. Next, let’s plot optimised efficient frontiers for high (90% market cap percentile) and low correlated (10% market cap percentile) crypto assets on the same plane. Below we distinguish that cryptos with high correlation (blue) have a lower standard deviation as in table 2: Figure 6: Optimised efficient frontier with unlimited weights allocation From Figure 6 we observe that portfolio variance is slightly decreasing with a number of assets. We also define that high correlated stocks has the lowest variance and Sharpe ratio which contradicts with traditional view that high risk is compensated with high reward. Maybe the explanation is that, the market risk is perceived lower than individual risk in crypto space. Below, let’s evaluate the cumulative performance for crypto asset strategies and compare it against the strategy of investing 100% in Bitcoin, equally weighted strategy or in the crypto index. Low correlation also states for 10% percentile of the marketcap and high correlation — for 90% percentile of the marketcap: Figure 7: Cumulative return during 2 months of holding the crypto portfolios: optimised max SR portfolio, optimised min vol portfolio, equally weighted (naive) portfolio and cumulative performance of BTC 100% holding and crypto-100 index. From Figure 7, we might expect that the 100% weight is allocated to one currency. We observe that for small market caps (first row) the optimised and equally weighted strategies beat the market and bitcoin portfolios while for large marketcaps portfolios the result is the opposite. Overall, portfolios with low correlated crypto assets (small marketcaps) deliver twice higher performance than high correlated portfolios. Below I present the numerical results for every strategy: Figure 8: Represents the performance of 19 crypto portfolios. In general, presenting portfolios for 2 groups assets from 10% and 90% percentile of marketcap distribution (which turned to be low and high correlated assets accordingly) and for 3 risk-reward strategies: optimised max Sharpe ratio, optimised min variance or equally weighted portfolios and benchmark portfolio as presented by crypto index of 100 currencies. From Figure 8 we observe that if not being limited some assets may be overweighted as a result of optimised strategy, therefore showing atypically low risks: low volatility (7,62 annualised) and low value-at-risk, low conditional value-at-risk and low max drawdown. Accordingly the bunch of portfolios of small marketcaps and low correlation shows extremely high risk profile which can be seen as compensated if look at the Sharpe ratio (the maximum Sharpe ratios portfolios are for 15 assets portfolios in the small marketcap group — 0.39,0.38,0.41). Remarkably, for all portfolios that are closely related to the market (as with large market caps and high correlation) the information ratio, as the measure of risk adjusted to benchmark, is negative, meaning that they poorly perform compared to the market. Finally, the results are consistent with MPT in that that all portfolios show lower variance by comparison with mean individual variance from table 2. Conclusions: Overall, dealing with crypto market has a data-mining approach, when you have to analyse the currencies almost individually. My finding confirms that low correlated crypto assets are those with small market capitalisation. It turns out that they represent very high volatility. So it is highly expensive in terms of risk to hold the less correlated small marketcap crypto assets, though they are compensated with highest rewards. While the high correlated crypto gained the high market share, they have higher correlation and in traditional world should be viewed as risky, here in the crypto space show less risky profile and underperform the benchmark. Interestingly, the efficient frontiers obtained from optimisation are very close to randomly generated frontiers. Thank you for your time and feedback! Appendix: Future prospects: First, apply the robust optimisation and estimation techniques. Of interest would be to model the efficient frontier for simulated looking-forward data, add mixed dataset and see whether the performance persist and build portfolios looking at the different forecast horizons an etc. “Mean-variance analysis is a normative theory, the theory which describes a standard or norm behaviour that investors should pursue while constructing portfolio”. In the crypto world, we expect that many investors act not “normally” driving market in different directions. Therefore, an analysis like sentiment-based analysis maybe more appropriate while making expectations about future prices. Literature: I like reading this fiction-like article which introduce the efficient frontier approach via a small talk in the casino. Authors rise an idea of a hybrid model where the final result is constructed from two components (historical and future, simulated component based on the frequency of the outcomes) and question the performance of on-frontier and off-frontier portfolios. One of their findings states that on-frontier portfolio performance prevail over off-frontier portfolios at leats for relatively short planning horizons.

Benefits of diversification: analysis of efficient frontiers for crypto and traditional markets.

17

benefits-of-diversification-analysis-of-efficient-frontiers-for-crypto-and-traditional-markets-15354f63eeb

2018-06-20

2018-06-20 15:55:21

https://medium.com/s/story/benefits-of-diversification-analysis-of-efficient-frontiers-for-crypto-and-traditional-markets-15354f63eeb

false

2,050

The personal blog about predictive modelling and financial data science.

null

willygoodwill

null

Beyond finance

olga.lazunina@gmail.com

beyond-finance

null

olgalazunina

Investing

investing

Investing

51,660

Olga Lazunina

null

7e8d812ee9f1

olgalazunina

3

4

20,181,104

null

0

null

0

9ba3897b6688

2018-04-18

2018-04-18 03:51:26

2018-04-18

2018-04-18 05:40:07

0

false

en

2018-04-18

2018-04-18 05:40:07

6

15357a82fe06

4.245283

0

Ultimately, our goal is to go beyond basic language modeling and create a new text generation architecture conducive to producing technical…

4

First Impressions: Baselines and the Evaluation Framework Ultimately, our goal is to go beyond basic language modeling and create a new text generation architecture conducive to producing technical definitions. To get a feel for the data though, we approach it with familiar, simple baselines that give us a foundation in which we can improve from. The baselines that we’ve experimented with are Vanilla RNN: Hidden states as a function of the input and the previous hidden state, with Tanh activation (in particular, we’re using the Elman Network). GRU: An RNN architecture that learns to throttle the influence and usage of particular parameters on inference using a gating mechanism. LSTM: Another RNN architecture that specializes in intelligently remembering relevant details and forgetting irrelevant details through several gating mechanisms and a “cell state” in addition to the conventional hidden states. Here’s more detail about this particular kind of RNN. All baselines were trained as language models with cross entropy loss and were used to get a sense of how learnable the language of the Semantic Scholar dataset is. The metrics we are focusing on now are Perplexity: A measure of how “confused” the model is at any point it’s attempting to predict the next word. Feature extraction through hidden states: Can the hidden states be used as features for a classification task? Our first metric is fairly straightforward; we calculate the aggregate sum of log probabilities for every word in the corpus and normalize by the size of the corpus. Our second metric however, is inspired by Dieng et al.’s method for sentiment analysis in TopicRNN, in which the final hidden states after a forward pass of the model on a movie review were used to predict positive or negative sentiment using a single-layer neural network. We aim to adopt this metric from a multi-class classification perspective in which the passages we use are excerpts of research documents with an omitted, technical term. The question we aim to answer with this metric is “is the model capable of representing semantics in a latent space?” Our labels will then be a defined set of these omitted, technical terms, and our goal will then be to predict them given the hidden states of the passage via a two-layer neural network. The framework for this evaluation metric can be found in this branch of our codebase. Why we chose these metrics Our task is fairly novel given the way we’re approaching it, so currently no dataset exists that pairs domain-specific words with definitions that are to the caliber of research technicality. Because of this, metrics that depend on gold standards such as ROUGE and BLEU are currently out of reach at this time. There’s a chance we’ll experiment with these metrics if we can find a labeled dataset to supplement Semantic Scholar’s Open Research Corpus. We are also considering using the publications themselves as the gold standards, which may be helpful since a desirable trait of our model would be its ability to produce language similar to that of the corpus. Challenges Encountered while Baselining In establishing our baselines and metrics, there were several issues we ran into, both in training. Dealing with a larger corpus Given that our baselines are language models, and that our later prototype models will most likely contain an LM component, we have to deal with efficient learning given there are several million documents to process. For efficient backpropagation, we opted to introduce a “backpropagation through time” as a hyperparameter defaulted at 50, which specifies the number of words we allow the model to see before updating our parameters. Currently, batching is supported by our codebase but was not used in our initial experiments. Given that it takes roughly one minute for the GPUs on the cloud to process a single publication, we plan to concatenate document vectors and reshape into batch-by-length tensors in the future. Sorting By Domain We’d like our model to be trained on a single domain/field of study for our future case studies comparing dictionaries built on one domain versus others. This is further motivated by some of our experiment results discussed later, how loss tends to spike between documents. Currently, the Semantic Scholar Open Research Corpus doesn’t include anything in the set of JSON fields that we could find for filtering the data. However, we’ve been assured by AllenNLP researchers that its possible to sort the data by research domain. We may performing another round of baseline experiments once we’ve sorted the data, but for now the results below are on publications of mixed domains. Experimental Results The Semantic Scholar Open Research Corpus provides a sample subset of its dataset: a JSON file containing 4000 entries. Within each entry, the URL of the publication’s online PDF is provided. We use a GET request to AI2’s Science Parse service to extract the PDF contents. From there, we run our experiments on 121 of the 4000 extracted documents, stopping training early at 15 documents and calculated perplexity on a validation set of 30 documents. The total vocabulary used was 11,330 words. Words outside of this vocabulary are replaced with an unknown token at training and test time. Loss is calculated and normalized on the last 50 words the model is trained on. Elman RNN: Perplexity of 250.25 with an average loss of 7.908 over the last 50 words. GRU: Perplexity of 265.50 with an average loss of 7.085 over the last 50 words LSTM: Perplexity of 261.95 with an average loss of 6.588 over the last 50 words Perplexities calculated using untrained models (with randomized parameters) were several orders of magnitude larger than the ones listed above, so it’s good to know that our baseline models can learn a significant amount of surface-level patterns with such a small subset of the corpus. In terms of feature extraction and classification, the framework has been implemented but the data for this has not been created yet. We plan on evaluating our baselines along with our final model on this metric using publications from Semantic Scholar after establishing a vocabulary of semantically significant technical terms and creating the dataset using those terms. This will involve iterating over documents and replacing occurrences of these technical terms with a specialized, unknown token that won’t aid in inference. Conclusion Our methods helped us familiarize ourselves with the data as well as handle significant amounts of overhead in terms of processing the data, toggling between different models, and integration of our evaluation metrics in an organized fashion. We are excited to see how novel architectures tailored to the task perform on these metrics! To keep up to date with our progress in baselining, evaluation, and other things, you can watch this repository.

First Impressions: Baselines and the Evaluation Framework

0

first-impressions-baselines-and-the-evaluation-framework-15357a82fe06

2018-04-18

2018-04-18 05:41:23

https://medium.com/s/story/first-impressions-baselines-and-the-evaluation-framework-15357a82fe06

false

1,125

A Journey Through CSE 481N, the Natural Language Processing Capstone Course at the University of Washington

null

NLP Capstone Blog

null

nlp-capstone-blog

DEEP LEARNING,NATURALLANGUAGEPROCESSING,UNIVERSITY OF WASHINGTON,CAPSTONE

null

Machine Learning

machine-learning

Machine Learning

51,320

Tam Dang

CS Undergraduate @ UW

dac135b89ed4

tamdangnadmat

9

2

20,181,104

null

0

null

0

null

2018-03-22

2018-03-22 11:02:49

2018-03-25

2018-03-25 12:06:11

0

false

en

2018-03-27

2018-03-27 20:50:06

2

1535c0c1bf63

1.162264

0

News related to technology can be very intimidating, and Artificial Intelligence is particularly a fertile area for imagination, and, of…

3

Learnings from the AI class at eBay News related to technology can be very intimidating, and Artificial Intelligence is particularly a fertile area for imagination, and, of course, science fiction. Movies like “Her” makes us to rethink the concept of humanity. In the same time, this is a great marketing opportunity, for example, Saudi Arabia gives citizenship to Sophia (a robot). For AI researchers, it is relatively easy to differentiate real AI progress and marketing gimmicks. However it might not be as trivial if you are not a scientist. Our goal in this course is to educate students enough that they can have an educated conversation about AI even if they are not researchers or engineers. The initial plan was to have a 10-session class covering the main machine learning/AI topics in the simplest possible language. The challenge was how to keep the class productive for a class with a broad spectrum of backgrounds. We decided to focus more on concepts and show use cases. However, 10 sessions class started to feel long for group of people who have many other responsibilities. This course is similar to “Continuing Study” courses that are typically shorter. As we are heading to the second round of the class, we decided to make two important changes in the class. First we reduced the course from 10 weeks to 8 weeks, and second, we divided the class in to two parts. In the first part (week 1–5), we will keep the course generic, and cover main concepts, and the terminology. The second part of the class (week 6–8) would be optional and we start coding, and building AI models. This part is suitable for people who wants to work with advanced AI tools and experiment with them. The second round will be started in April. Check our website for more info: https://ai4all.github.io/

Learnings from the AI class at eBay

0

learnings-from-the-ai-class-at-ebay-1535c0c1bf63

2018-03-27

2018-03-27 20:50:07

https://medium.com/s/story/learnings-from-the-ai-class-at-ebay-1535c0c1bf63

false

308

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Davood Shamsi

null

75a22476497a

davood.shamsi

5

20,181,104

null

0

null

0

49523d4bd9da

2018-03-07

2018-03-07 07:07:22

2018-03-07

2018-03-07 09:48:38

7

false

en

2018-03-15

2018-03-15 08:44:23

0

1537815bd67b

3.514151

1

0

By Awanee Joshi and Vaidehie Chiplunkar

1

What are the chances? By Awanee Joshi and Vaidehie Chiplunkar Have you ever used your bargaining skills to get a bag for half its price? Did the salesman do something funny? Was it a story to tell? With devices becoming smarter everyday our dependence on people has reduced. For example, GPS has taken over way finding. It tells you everything anywhere. But what happened to making friends on the way? Now imagine a world of 2021. Shopkeepers are now replaced with artificial intelligence. You no longer have to rely on the person at the counter. You ask the AI for what you need, it gives you the things, you pay and you leave. But where’s the fun in that? Everyday everything happens the same way. What about the rude shopkeepers? What about you grabbing their attention in a crowd? What about bargaining? What about you regular guy who now knows what you want? What about getting your daily dose of gossip from them? Won’t life become mundane without these things? We imagined an AI that would give you the experience of dealing with the unpredictability of shopping. We tried visualizing all these situations with a moody AI. We chose 2 stories from our 6 situations and tried to show the experience a through a skit. We used Arduino to make the AI prop. AI shopkeeper — our prop We added a moving array of straws for the final performance. After our short skit of 1–2 mins we gathered together for a critical feedback session. Reflections: Based on the feedback we got, we realized we needed to tweak our story a little bit to make it more relatable and to make more sense. We could add a distinctive character and style to every AI. We could also change the personality of the same AI according to different people. (Like when you go to a barber’s shop and ask for the same guy.) We could think about other spaces where the same idea could be implemented. (For example, in taxis.) Someone might say why not give the humans this job back? We think from the shop owner’s point of view it is easier to manage an AI than employees. The human effort of standing, waiting is eliminated when the AI replaces them. There could be more creative and empathy based places where the humans would rather work. There was a suggestion that the products themselves became moody, and the billing would be done by humans. Imagine the MRP changing after bargain. But the fun in bargaining is getting something for a lesser price than the MRP. Another question one might ask is would we be obeying the AI or the other way around? We want the AI to be an equal entity as a human. So neither is the AI in control nor the humans. (We need to show this in our story.) There will be times when you are annoyed and you don’t want to deal with a challenging AI. Why would you want such an AI? The happiness in a good day increases when you experience a bad one. If things always happened the way you want, wouldn’t life become too boring? Just like you can’t control a shop keeper, why control an AI? The AI can increase the price based on your history. If once you’ve paid more than the actual price, in the future it would give you a discount on something to make it fair. What do we do with the data collected? The user gets a better and different experience every time. It could also help the owner know what kind of things what kind of people want, when do they buy it the most and what products they want to bargain for. When you think of an AI you always think of a western scenario. What we would like to do is to place it in the scenario of a local Indian market. We want to imagine it in Dhalgarwad, an Ahmedabad cloth market.

What are the chances?

1

what-are-the-chances-1537815bd67b

2018-03-15

2018-03-15 08:44:24

https://medium.com/s/story/what-are-the-chances-1537815bd67b

false

653

A collaborative workshop at NID-Ahmedabad exploring narratives for a digital future. Watch as we learn.

null

Howdu.ino

null

howdu-ino

IOT,ARDUINO,DIGITAL,FUTURE,NARRATIVE

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Vaidehie Chiplunkar

null

fbf7b3297d1a

vaidehie_c

4

1

20,181,104

null

0

null

0

fe8defe6de57

2017-11-15

2017-11-15 15:34:48

2017-11-15

2017-11-15 15:36:19

3

false

en

2017-11-16

2017-11-16 14:44:46

10

15381484f058

3.814151

4

0

Earlier this month we got an update from Julius Simonelli, a project lead on this ongoing gerrymandering project. Over the summer, a team…

3

D4D Gerrymandering Project — Update from http://sapiencefilm.com Earlier this month we got an update from Julius Simonelli, a project lead on this ongoing gerrymandering project. Over the summer, a team from Data for Democracy working with Professor Sam Wang of the Princeton Election Consortium at Princeton University, prepared an amicus brief for the gerrymandering case that was recently before the Supreme Court, Gill v. Whitford. “Professor Wang wrote three statistical tests to measure partisan gerrymandering that he wanted to implement in Python. He also wanted to make his gerrymandering website more user-friendly. D4D had teams working on both of these aspects.” A Little Background Gerrymandering is a big political word that most often gets tossed about during an election year. For those that aren’t aware, gerrymandering refers to the process of diving and redrawing up congressional districts to give your political party an advantage. Many people are not fully aware that US congressional districts, which are the districts that elect your representative in the house, are not divided up in any regular fashion. You can see for yourself, and find your own district, on this congressional map. from https://philebersole.wordpress.com They can be redrawn by packing or cracking, which is political speak for stuffing a group of voters all into one district, or breaking up a group of voters by splitting them arbitrarily into several districts. Either way, it presents and opportunity to skew who gets elected to represent you in congress.The only exception is for states that have only 1 representative in the house, in which case all the voters in the state are effectively in one district. So, gerrymandering can be bring about big problems, like is this really constitutional? Project Update by Julius Simonelli: “For some background, an amicus brief is a submission to a court by someone who isn’t a litigant to the case but still wants to provide an argument to the courts. The Gill v. Whitford case was brought in response to the Wisconsin State Assembly district plan that was drawn up after the 2010 census. The Republican party was in control of the state government after the 2010 election and therefore drew up the district maps (in some states a nonpartisan commission draws the maps, in others, it’s the party in power). The first election after the maps were drawn was in 2012. While the Democrats won the majority of the votes, the Republicans won 60 out of 99 seats in the Assembly. The results for 2014 and 2016 were similarly disproportionate. In 2015, the 2010 Wisconsin redistricting plan was challenged as unconstitutional. What does the amicus brief focus on? The amicus brief was submitted on 30 August 2017. It attempts to convince the judges (read: Justice Kennedy) of two points: 1. The issue of political gerrymandering is judiciable, i.e. partisan gerrymandering is a violation of the Constitution and the courts have the right and responsibility to intervene. 2. A judicially manageable standard for determining gerrymandering does exist Most of the brief concerns the first issue, arguing that it would be appropriate for SCOTUS to make a ruling in the case of partisan gerrymandering (i.e., that it is judiciable). There is a fair amount of precedence for courts deciding that congressional districts are unconstitutional. Racial gerrymandering is illegal under the Voting Rights Act, but partisan gerrymandering is currently not. It is illegal to pack all minorities into a single district (Covington v. North Carolina, 2017), but you can with a political party. The brief argues that creating a system to intentionally disenfranchise a group of people based on their political affiliation violates the Equal Protection Clause of the 14th Amendment as well as the 1st Amendment’s prohibition on discriminating based on viewpoint. The second point the paper tries to make is that there is a judicially manageable standard with a concept known as “partisan symmetry”. Partisan symmetry means that if one party wins, say, 55% of the vote and wins 60% of the seats, that if the other party won 55% of the vote under a likely voting scenario, they would also win, approximately, 60% of the seats. Note that this is different than proportional representation where a party that wins say, 40% of the votes is guaranteed 40% of the seats. Proportional representation is not guaranteed by the Constitution and is not what was being proposed (even though Chief Justice Roberts called it that and the attorney for the plaintiff had to correct him). The point of the briefing isn’t to say that we found THE statistical test to determine excessive partisan gerrymandering, but to say that there are many tests that would work and here are a few examples. In many cases, the Supreme Court provides the guiding principles and the lower courts determine for themselves what metrics are best.” What Happens Next? The oral argument was held on October 3, 2017. There is likely to be a 5–4 decision, and most expect that Justice Kennedy will determine the outcome. The decision is expected around June of 2018. Join us on the Data for Democracy slack to participate in this project! from http://www.electoral-vote.com References: Full text of the amicus brief Professor Wang’s article on the three techniques Professor Wang’s website on gerrymandering

D4D Gerrymandering Project — Update

8

d4d-gerrymandering-project-update-15381484f058

2018-03-22

2018-03-22 17:19:33

https://medium.com/s/story/d4d-gerrymandering-project-update-15381484f058

false

865

Collaborating on data projects to build a stronger society. Email team@datafordemocracy.org for an invitation.

null

datafordemocracy

null

Data for Democracy

team@datafordemocracy.org

data-for-democracy

DATA,OPEN DATA,DATA SCIENCE,CIVIC ENGAGEMENT,CIVICTECH

data4democracy

Politics

politics

Politics

260,013

Astrid Willis Countee

Community Manager @data4democracy

b87e81367e95

ianthro

186

322

20,181,104

null

0

null

0

null

2018-06-29

2018-06-29 13:20:42

2018-06-20

2018-06-20 12:03:51

2

false

en

2018-06-29

2018-06-29 13:21:25

17

153ae51cec6

5.281447

0

Artificial Intelligence will become the next disruptive customer experience trend. A Hotel that position and prepare themselves for this is…

5

How Hotels can use Artificial Intelligence to tap into New Uncontested Social Media Audiences, Without Breaking the Budget Artificial Intelligence will become the next disruptive customer experience trend. A Hotel that position and prepare themselves for this is opening their palms to tap into new uncontested Social Media Audiences, without breaking the budget. Research shows that Artificial Intelligence will be the most disruptive advanced technologies over next decade. This will open a new world for Hotels with access to the almost near-endless amount of data and advance in deep learning technology. My friends at Avvio did a research showing us how Artificial Intelligence will impact Hotels, staff, and guests. Deep learning technology will enable Hotels to overcome common barriers such as skills, standardization, and complexity. Online travel agencies have already been experimenting with some of the early stages of Artificial Intelligence. Expedia has used it to help Hotels put the right photos in front of the right traveler. Travelers spend less than a second deciding what they think of a Hotel today. Make sure your Hotels photos are worthy an Instagrammable moment. Deep learning technology will allow Hotels to tailor unique images fit for every customer type. Artificial Intelligence Project Management the first step for Hotels Project management Artificial Management is a system that can perform the day-to-day management and administration of projects and tasks without requiring human input. It will not only automate simple tasks but will also develop an understanding of key project performance. Project management AI can then use this understanding to uncover insights, perform more complex tasks, make recommendations, and make decisions; sometimes in ways, people just can’t do today. Ultimately, an AI system will save your Hotel time and cost while improving outcomes for your Hotels projects and team. The important part for Hotels is to start with the simple projects that allow them to integrate Artificial Intelligence in effective ways that also is in line with the value proposition. The future of Hotels with Artificial Intelligence Avvio raise an important question related to Artificial Intelligence; “What Happens to Humans and Jobs when Artificial Intelligence arrives?”. I believe Avvio brings out a really valid point that we as humans have to not look at it as the machines are not replacing humans, rather humans working together with machines. Artificial Intelligence and new technology, in general, will allow Hotels to identify trends and patterns in customer and guest behaviors. This will allow your Hotel team to focus on optimizing micro-moments that will add new values to the customer experience. Artificial Intelligence will allow Hotel Team members to tailor and personalize experience based on the new hyper-targeted relevant information. Customer Experience will still require the human touch Over the years we as Hoteliers have focused a good portion of customer service on empathy. It will take a long time before Artificial Intelligence is capable of simulating human empathy. The interesting aspect here is that studies have shown that humans are not requiring empathy for all forms of transactions. Sometimes a guest arriving to check in just want their room to be ready to have to bother with empathy. Artificial Intelligence will help identify patterns where human interaction and empathy is more critical. It will provide a unique opportunity for Hotels to modernize the customer service element. Artificial Intelligence beyond cost savings Avvio suggests some creative applications of Artificial Intelligence such as; optimizing housekeeping, by approximating early check-in need or late check-out or predict maintenance work done to rooms. Consumer demands are trending towards experience suggestions from the Hotel team. The Hotel has to start focus in new alternatives to stay competitive. Artificial Intelligence has to potential to go beyond cost savings for Hotels. It can help eliminate duplicate processes and automate some of the customer service processes. Not only will help the Hotel Team by simplifying processes but also it makes it easier for the customer to discover the Hotels offerings based on their individual needs. It will help the customer achieve more purposeful goals while traveling. Artificial Intelligence is the path to create new unique Hotel alternatives through automation and applications. Hotel teams that start take advantage of project management artificial intelligence will be moving at light speed compared to the competition that will sit on the fence. Avvio is developing systems that are helping Hotels creating tomorrow’s “Wow!”. They recently launched Allora, an artificially intelligent Hotel website, and booking engine. Their goal has never been building a high-tech booking engine, but rather working on personalizing the entire process: good booking engines should be active, rather than passive, and, similarly, a good Hotel website experience should be personalized, and not merely a digital brochure. And they tried to bring that human touch out of the confines of the Hotels to the whole travel experience: from the first interaction with the website all the way down to months or even years after the stay. This is where AI becomes remarkably valuable, and not a simple buzzword. Artificial Intelligence and Social Media Artficial Intellegience is already an integrated part of Social Media. A lot of Hotels today is utilizing LinkedIn to seek out potential members for their Hotel Team. LinkedIn invested in Bright, a company which enables the site to offer better candidate matches for both employers and job seekers. It takes into account the user’s hiring patterns, work experience and similarities in job descriptions. The integration of Artificial Intelligence with Social Media has, in turn, increased the accuracy of Social media and online Hotel marketing. So, from the perspective of online media services such as Google and Facebook, having AI programs that understand customer personal interests and feelings should help improve the Hotels online experience, as well as enjoyment with technology overall. From a Hotel and Social Media Consultant’s standpoint, the Social Media Community can offer Hotels a one-stop-shop for managing their Social Media marketing such as the community infrastructure, database, listening functionality, profile management, collaboration, content marketing and management, and of course, the analytics. A Social Media Community will help your Hotel to grow smarter and faster when you focus on the value proposition and collaborate in effective ways. If your Hotel need help to put it all together then make sure to the let us know. Today’s Social Media Management tools provide advanced options that will help Hotels with these tasks. If you need a quality team the help your Hotel pulls this off, then get in touch with my partners at Founders Media. With the right team behind your Social Media efforts, it can become cost-effective and generate a new revenue stream for your Hotel. To get your Hotel started here are a few FREE gifts and resources from me and my partners Founders Media. 1. FREE access to my Hospitality Gone Social Vault (Click Here to Unlock) 2. FREE ebook from Founders Media (Click Here to Unlock) 3. FREE 21 Days To Social Media Mastery for Hotels email course (Click Here to Unlock) 4. FREE access to #HotelPodcast — A Podcast about Social Media Mastery for Hotel (Click Here to Unlock) 5. FREE access to our Facebook Group — Hospitality Gone Social (Click Here to Unlock) A special thanks to Martin Soler for introducing this important topic. Avvio leads the way with innovative solutions for Hotels and accommodation providers. They developed the world’s first AI booking engine to exceed the ever-evolving needs of properties across Europe and North America, whilst delivering outstanding performance in direct revenue growth. Are Morch is the founder and owner of Are Morch — Hotel Blogger and Social Media Consultant. Get more from Are on Facebook | Twitter | Google Plus | LinkedIn | Pinterest | Instagram Originally published at aremorch.com on June 20, 2018.

How Hotels can use Artificial Intelligence to tap into New Uncontested Social Media Audiences…

0

how-hotels-can-use-artificial-intelligence-to-tap-into-new-uncontested-social-media-audiences-153ae51cec6

2018-06-29

2018-06-29 13:21:25

https://medium.com/s/story/how-hotels-can-use-artificial-intelligence-to-tap-into-new-uncontested-social-media-audiences-153ae51cec6

false

1,298

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Are Morch

Hotel Blogger & Social Media Consultant http://bit.ly/HotelCoach | Social Media Consultant for Hotels — http://bit.ly/Social4Hotels | Horse Rescuer

dad913f35ee4

aremorch

1,327

1,479

20,181,104

null

0

null

0

ab837e78463f

2017-11-15

2017-11-15 08:05:15

2017-11-15

2017-11-15 08:41:06

8

false

en

2018-02-21

2018-02-21 21:00:18

13

153b6ade9f66

8.67673

1,978

41

2

Part of Understanding Hinton’s Capsule Networks Series:

5

Understanding Hinton’s Capsule Networks. Part II: How Capsules Work. Part of Understanding Hinton’s Capsule Networks Series: Part I: Intuition Part II: How Capsules Work (you are reading it now) Part III: Dynamic Routing Between Capsules Part IV: CapsNet Architecture Quick announcement about our new publication AI³. We are getting the best writers together to talk about the Theory, Practice, and Business of AI and machine learning. Follow it to stay up to date on the latest trends. Introduction In Part I of this series on capsule networks, I talked about the basic intuition and motivation behind the novel architecture. In this part, I will describe, what capsule is and how it works internally as well as intuition behind it. In the next part I will focus mostly on the dynamic routing algorithm. What is a Capsule? In order to answer this question, I think it is a good idea to refer to the first paper where capsules were introduced — “Transforming Autoencoders” by Hinton et al. The part that is important to understanding of capsules is provided below: “Instead of aiming for viewpoint invariance in the activities of “neurons” that use a single scalar output to summarize the activities of a local pool of replicated feature detectors, artificial neural networks should use local “capsules” that perform some quite complicated internal computations on their inputs and then encapsulate the results of these computations into a small vector of highly informative outputs. Each capsule learns to recognize an implicitly defined visual entity over a limited domain of viewing conditions and deformations and it outputs both the probability that the entity is present within its limited domain and a set of “instantiation parameters” that may include the precise pose, lighting and deformation of the visual entity relative to an implicitly defined canonical version of that entity. When the capsule is working properly, the probability of the visual entity being present is locally invariant — it does not change as the entity moves over the manifold of possible appearances within the limited domain covered by the capsule. The instantiation parameters, however, are “equivariant” — as the viewing conditions change and the entity moves over the appearance manifold, the instantiation parameters change by a corresponding amount because they are representing the intrinsic coordinates of the entity on the appearance manifold.” The paragraph above is very dense, and it took me a while to figure out what it means, sentence by sentence. Below is my version of the above paragraph, as I understand it: Artificial neurons output a single scalar. In addition, CNNs use convolutional layers that, for each kernel, replicate that same kernel’s weights across the entire input volume and then output a 2D matrix, where each number is the output of that kernel’s convolution with a portion of the input volume. So we can look at that 2D matrix as output of replicated feature detector. Then all kernel’s 2D matrices are stacked on top of each other to produce output of a convolutional layer. Not only can the CapsNet recognize digits, it can also generate them from internal representations. Source. Then, we try to achieve viewpoint invariance in the activities of neurons. We do this by the means of max pooling that consecutively looks at regions in the above described 2D matrix and selects the largest number in each region. As result, we get what we wanted — invariance of activities. Invariance means that by changing the input a little, the output still stays the same. And activity is just the output signal of a neuron. In other words, when in the input image we shift the object that we want to detect by a little bit, networks activities (outputs of neurons) will not change because of max pooling and the network will still detect the object. The above described mechanism is not very good, because max pooling loses valuable information and also does not encode relative spatial relationships between features. We should use capsules instead, because they will encapsulate all important information about the state of the features they are detecting in a form of a vector (as opposed to a scalar that a neuron outputs). Capsules encapsulate all important information about the state of the feature they are detecting in vector form. Capsules encode probability of detection of a feature as the length of their output vector. And the state of the detected feature is encoded as the direction in which that vector points to (“instantiation parameters”). So when detected feature moves around the image or its state somehow changes, the probability still stays the same (length of vector does not change), but its orientation changes. Imagine that a capsule detects a face in the image and outputs a 3D vector of length 0.99. Then we start moving the face across the image. The vector will rotate in its space, representing the changing state of the detected face, but its length will remain fixed, because the capsule is still sure it has detected a face. This is what Hinton refers to as activities equivariance: neuronal activities will change when an object “moves over the manifold of possible appearances” in the picture. At the same time, the probabilities of detection remain constant, which is the form of invariance that we should aim at, and not the type offered by CNNs with max pooling. How does a capsule work? Let us compare capsules with artificial neurons. Table below summarizes the differences between the capsule and the neuron: Important differences between capsules and neurons. Source: author, inspired by the talk on CapsNets given by naturomics. Recall, that a neuron receives input scalars from other neurons, then multiplies them by scalar weights and sums. This sum is then passed to one of the many possible nonlinear activation functions, that take the input scalar and output a scalar according to the function. That scalar will be the output of the neuron that will go as input to other neurons. The summary of this process can be seen on the table and diagram below on the right side. In essence, artificial neuron can be described by 3 steps: scalar weighting of input scalars sum of weighted input scalars scalar-to-scalar nonlinearity Left: capsule diagram; right: artificial neuron. Source: author, inspired by the talk on CapsNets given by naturomics. On the other hand, the capsule has vector forms of the above 3 steps in addition to the new step, affine transform of input: matrix multiplication of input vectors scalar weighting of input vectors sum of weighted input vectors vector-to-vector nonlinearity Let’s have a better look at the 4 computational steps happening inside the capsule. 1. Matrix Multiplication of Input Vectors Input vectors that our capsule receives (u1, u2 and u3 in the diagram) come from 3 other capsules in the layer below. Lengths of these vectors encode probabilities that lower-level capsules detected their corresponding objects and directions of the vectors encode some internal state of the detected objects. Let us assume that lower level capsules detect eyes, mouth and nose respectively and out capsule detects face. These vectors then are multiplied by corresponding weight matrices W that encode important spatial and other relationships between lower level features (eyes, mouth and nose) and higher level feature (face). For example, matrix W2j may encode relationship between nose and face: face is centered around its nose, its size is 10 times the size of the nose and its orientation in space corresponds to orientation of the nose, because they all lie on the same plane. Similar intuitions can be drawn for matrices W1j and W3j. After multiplication by these matrices, what we get is the predicted position of the higher level feature. In other words, u1hat represents where the face should be according to the detected position of the eyes, u2hat represents where the face should be according to the detected position of the mouth and u3hat represents where the face should be according to the detected position of the nose. At this point your intuition should go as follows: if these 3 predictions of lower level features point at the same position and state of the face, then it must be a face there. Predictions for face location of nose, mouth and eyes capsules closely match: there must be a face there. Source: author, based on original image. 2. Scalar Weighting of Input Vectors At the first glance, this step seems very familiar to the one where artificial neuron weights its inputs before adding them up. In the neuron case, these weights are learned during backpropagation, but in the case of the capsule, they are determined using “dynamic routing”, which is a novel way to determine where each capsule’s output goes. I will dedicate a separate post to this algorithm and only offer some intuition here. Lower level capsule will send its input to the higher level capsule that “agrees” with its input. This is the essence of the dynamic routing algorithm. Source. In the image above, we have one lower level capsule that needs to “decide” to which higher level capsule it will send its output. It will make its decision by adjusting the weights C that will multiply this capsule’s output before sending it to either left or right higher-level capsules J and K. Now, the higher level capsules already received many input vectors from other lower-level capsules. All these inputs are represented by red and blue points. Where these points cluster together, this means that predictions of lower level capsules are close to each other. This is why, for the sake of example, there is a cluster of red points in both capsules J and K. So, where should our lower-level capsule send its output: to capsule J or to capsule K? The answer to this question is the essence of the dynamic routing algorithm. The output of the lower capsule, when multiplied by corresponding matrix W, lands far from the red cluster of “correct” predictions in capsule J. On the other hand, it will land very close to “true” predictions red cluster in the right capsule K. Lower level capsule has a mechanism of measuring which upper level capsule better accommodates its results and will automatically adjust its weight in such a way that weight C corresponding to capsule K will be high, and weight C corresponding to capsule J will be low. 3. Sum of Weighted Input Vectors This step is similar to the regular artificial neuron and represents combination of inputs. I don’t think there is anything special about this step (except it is sum of vectors and not sum of scalars). We therefore can move on to the next step. 4. “Squash”: Novel Vector-to-Vector Nonlinearity Another innovation that CapsNet introduce is the novel nonlinear activation function that takes a vector, and then “squashes” it to have length of no more than 1, but does not change its direction. Squashing nonlinearity scales input vector without changing its direction. The right side of equation (blue rectangle) scales the input vector so that it will have unit length and the left side (red rectangle) performs additional scaling. Remember that the output vector length can be interpreted as probability of a given feature being detected by the capsule. Graph of the novel nonlinearity in its scalar form. In real application the function operates on vectors. Source: author. On the left is the squashing function applied to a 1D vector, which is a scalar. I included it to demonstrate the interesting nonlinear shape of the function. It only makes sense to visualize one dimensional case; in real application it will take vector and output a vector, which would be hard to visualize. Conclusion In this part we talked about what the capsule is, what kind of computation it performs as well as intuition behind it. We see that the design of the capsule builds up upon the design of artificial neuron, but expands it to the vector form to allow for more powerful representational capabilities. It also introduces matrix weights to encode important hierarchical relationships between features of different layers. The result succeeds to achieve the goal of the designer: neuronal activity equivariance with respect to changes in inputs and invariance in probabilities of feature detection. Summary of the internal workings of the capsule. Note that there is no bias because it is already included in the W matrix that can accommodate it and other, more complex transforms and relationships. Source: author. The only parts that remain to conclude the series on the CapsNet are the dynamic routing between capsules algorithm as well as the detailed walkthrough of the architecture of this novel network. These will be discussed in the following posts. Thanks for reading! If you enjoyed it, hit that clap button below and follow me! It would mean a lot to me and encourage me to write more stories like this. You can follow me on Twitter. Let’s also connect on LinkedIn.

Understanding Hinton’s Capsule Networks. Part II: How Capsules Work.

15,028

understanding-hintons-capsule-networks-part-ii-how-capsules-work-153b6ade9f66

2018-06-19

2018-06-19 12:50:46

https://medium.com/s/story/understanding-hintons-capsule-networks-part-ii-how-capsules-work-153b6ade9f66

false

1,999

The AI revolution is here! Navigate the ever changing industry with our thoughtfully written articles whether your a researcher, engineer, or entrepreneur

null

AI³ | Theory, Practice, Business

null

ai³-theory-practice-business

AI,MACHINE LEARNING,ENTREPRENEURSHIP,PROGRAMMING,DATA SCIENCE

null

Machine Learning

machine-learning

Machine Learning

51,320

Max Pechyonkin

Deep Learning

38025716cba8

pechyonkin

5,930

97

20,181,104

null

0

null

0

null

2017-12-05

2017-12-05 17:22:50

2017-12-05

2017-12-05 17:24:25

0

false

en

2017-12-05

2017-12-05 17:24:25

0

153c28d3d38b

6.513208

0

As Jim arrived in the lobby of the new 40 storey apartment block, a woman leading a small square-jawed dog in a pink muzzle and matching…

5

What We Talk About When We Talk About Self-Driving Cars As Jim arrived in the lobby of the new 40 storey apartment block, a woman leading a small square-jawed dog in a pink muzzle and matching jacket bearing the words “Don’t Pet Me” was leaving the elevator. A huge fluffy, white beast pulled his owner away at the sight of the smaller dog. Jim stood to one side while the dog owners negotiated the exchange. Roy appeared at his side. “It’s a dog-friendly building,” he explained, flashing a resident’s pass at the smart panel. They got in the next elevator and smiled at each other. “Been a couple years. Glad you could make the time to come by and stay over.” In the corner apartment overlooking the city Elspeth was unpacking a Whole Foods bag, in between flipping a clean towel onto the bathroom rail and kicking her gym shoes under the sofa. She greeted them at the door. “Jim! How lovely.” After Jim had admired the view — two windowed walls overlooking the nexus of highways leading commuters out of the city, they sat at the breakfast bar and shared a bottle of wine, chewing over old times, while Roy and Elizabeth produced dinner, chopping vegetables for a salad, frying fish. After dinner the argument began. Moving to the sitting area of the tiny living room area of the Webster’s apartment, set like an eyrie above the city’s nexus of highways, they surveyed the jammed outward-bound traffic across three lines of the interstate. “The red brake lights are so pretty, like Christmas lights, ” Elspeth said, setting the bottle of Ardbeg Jim had brought as a gift on the coffee table on a tray, along with three tumblers and a jug of water. “Not for much longer though. I suppose the urban landscape will change dramatically with self-driving cars.” “Of course.” Roy poured whisky into the three glasses, adding a splash of water to his own. “Dramatically. There won’t be a need for this number of cars. People won’t own their own cars — they will all belong to Uber of Lyft or something, you will just stop one as it’s gong past, jump in and tell it where to go.” “Maybe.” Jim unfurled himself from a low, leather armchair to collect his drink. “Maybe it won’t happen like that. Where am I sleeping by the way? Happy to bed down anywhere.” Elspeth pointed to the sofa where she and Roy were sitting. “On this sofa. It’s reasonably comfortable.” Roy was focused on the subject. “What do you mean, Jim? It’s already happening, the first self-driving cars are already being tested.It’s literally where the rubber meets the road, it is where technological disruption and AI will really start to become physically visible in the landscape. I’d have thought you would be excited by that.” Jim shrugged. “Well for a start, we already have this technology for trains which run on tracks and yet where do you get self-driving trains outside airports?” Roy put down his drink. “But the economics of trains are totally different — you need one driver to move hundreds of people.” “I agree with Jim,” Elspeth diluted her whisky and took a sip. “That’s a beautiful whisky, by the way, it’s so smokey. In the UK, something like 75% of cars are still manual. If people aren’t prepared to pay a bit extra for an automatic gear shift, will they want to buy self driving cars? And isn’t private ownership kind of the whole point of the car?” Roy gestured out of the window. “But look at those people in that traffic jam, they are crawling along. They must spend hours a day sitting in those lines — don’t you think they would be prepared to compromise if they could save time on the journey? And half of them are probably sitting there fiddling on their mobile phones. They are not really driving in any remotely satisfying way. Autonomous cars could improve the experience so much. At a minimum, people could use their devices safely, but maybe they can even rip out the chairs and steering wheels and put in, I don’t know, gyms or something, you could work out in there.” “I don’t think so,” Jim sounded amused. “You are running on the treadmill and the car brakes suddenly — you are going to go flying. The laws of physics still apply. You would have to wear a seatbelt.” “But the point is we have to think bigger about how this will have the potential to change our experience. We can reinvent the steering wheel.” “Yes, for good and for ill — like putting all those Uber drivers and truck drivers out of work. Those are huge areas of employment. What will those people do to make ends meet?” Elspeth poured herself another whisky. “Help yourself, JIm.” “So you do accept that it will happen?”Roy asked Elspeth. “Maybe in some parts of the world, like the Google campus or San Francisco. I can imagine that in some places they will work. But how about in, I don’t know, India, or Morocco. There is so much going on on the roads in places like that. Say the GPS takes you into a crowded market — and that has happened to us before, remember Naples — your car clips a fruit stall and destroys all the produce. Who do think the crowd is going to hold responsible for that? The car? Or the occupants of the car? And what about security? Hackers turning the brakes off from miles away. Or robbers. What happens if someone walks in front of a self-driving car?” “It stops. It would have to.” Jim uncorked the Ardbeg bottle and poured himself a generous measure. “ “I mean, what if a robber walks out in front of a self-driving truck? It has to stop. But eight out of ten truck drivers would keep going, swerve or maybe even hit the robber.” Roy was becoming agitated. “What do you mean eight out of ten, that’s a made up statistic. Obviously there would be some security feature. If a truck is forced to stop and the situation isn’t clarified within ten minutes, the truck immobilises and the contents self-destruct.” Elspeth and Jim both laughed, increasing Roy’s annoyance. “You guys are just Luddites. You are like the people who, when the steam engine was invented, worried that people’s heads would explode if they travelled at more than 25 mph.” “Perhaps I am a Luddite,” Elspeth responded. “ But Luddites represent a proud tradition of resistance to the commercialisation of everything. Look at our kids, spending hours and hours a day on screens. We give them devices that are deliberately designed by very clever people to be addictive and we don’t know what the downside of that might prove to be. Now we are talking about remaking the world in a way that essentially enables us to be on the screens even more. Is that actually what we want? “Everyone is always trying to predict the future but maybe that’s too passive. Maybe that’s just a way of giving up, accepting that whatever the Googles and the Ubers of this world want to do to our urban landscape is OK. “Why should we let massive commercial organizations redesign the way we live in a way that suits their interests? Is that the best we can do? How about trying to reimagine spaces to make them better for people to connect, to play, to create, to contribute.” Elspeth sat back and sighed. “Oh my head hurts now. Shouldn’t mix my drinks. I’m going to bed.” Roy nodded. “Yes, probably time to call it a night. I think your view is overly negative. These are exciting developments. They will make life better for many people. It’s going to be great! I can’t wait to get my driverless car.” “Hmm. I think there’s a lot in what Elspeth says. It would be great if people like you had some say over how the cities of the future were planned, Elspeth. If the invisible hand of the capitalist free market designs it, it could be pretty awful.” Jim walked to the window and looked down at the station sidings below. “Look at the state of those trains. They are not exactly state of the art. High speed trains are a very carbon-efficient way of moving lots of people — and convenient. Travellers can walk around and use their computers, they don’t have to drive, every individual doesn’t need a tonne of metal around them. And yet the US train network is falling apart, the trains are slow, expensive, the rails are not as well-maintained as they should be. China is literally looping the world with high speed trains. This is not just about efficiency, it’s about profit. “I guess autonomous cars probably are coming, like autonomous everything. My worry is that the more autonomous machines become, the more helpless people do, we give up our agency to the machines and the companies that produce them.’ Elspeth and Roy, both flushed with whisky, got up to go to bed. Elspeth got some bedding out and started to make up the sofa. “Just leave that, I’ll do it in a moment — it’s time for my evening cigarette, still smoke I am afraid.” Pulling on his jacket with his cigarettes in the pocket, Jim strolled along to the elevator and pressed ground on the panel. The lobby was quiet, the concierge gone replaced by a surly-looking security guard. Outside the autumn night was chilly and the street was empty. Jim leaned against a tree and smoked the evening cigarette he still allowed himself. It was only after smoking it that he realised he didn’t know the number of Elspeth and Roy’s apartment, and that he didn’t have either his phone or his wallet. He looked up at the building, which contained perhaps 1,000 identical rental apartments. Something told Jim that the surly security guard would not furnish him with the number of his friends’ one — even if he had access to that information. Elspeth and Roy had gone to bed. Could he find a payphone? Jim realised he no longer knew a single phone number.

What We Talk About When We Talk About Self-Driving Cars

0

what-we-talk-about-when-we-talk-about-self-driving-cars-153c28d3d38b

2018-03-05

2018-03-05 23:24:30

https://medium.com/s/story/what-we-talk-about-when-we-talk-about-self-driving-cars-153c28d3d38b

false

1,726

null

Self Driving Cars

self-driving-cars

Self Driving Cars

13,349

jackie kemp

null

1491391b0f4e

jackiekemp0

0

2

20,181,104

null

0

null

0

null

2018-05-15

2018-05-15 18:25:36

2018-05-15

2018-05-15 18:31:06

3

false

en

2018-05-15

2018-05-15 18:31:06

0

153d0d60124

7.342453

0

It is unusual for Apple to be first to a market. They have a reputation for sitting back, waiting to see the user problems within a…

4

Apple has a unique opportunity to revolutionize voice assistants It is unusual for Apple to be first to a market. They have a reputation for sitting back, waiting to see the user problems within a specific market they may be interested in and then creating a product that solves those problems in a delightful way. Siri was the first voice assistant to hit the mainstream market. Yet, in over 6 years Apple’s voice assistant has barely made any progress, while it’s competitors like Amazon’s Alexa the Google Assistant both have made huge advancements in half the time. So, what can Apple do to differentiate Siri and take its lead back in the market? What would be so compelling as to completely change the public perception of Siri and voice assistants in general? Throughout this article, I will make the case that the way we see voice assistants needs to fundamentally shift and that Apple is in a unique position to lead the way. Humans and Voice Assistants Right now, the accepted state of Voice Assistants is that they must always be connected and live in the cloud. I think that this is a fundamental flaw in how we are approaching the design of using voice as an interface. At a deep level of their psyche, people cannot relate with something that lives in the “cloud” and that is always trying to get its information from the internet but has no database of its own knowledge. This is because we are not wired that way. In our human experience, it is natural for us to understand the process of having a question, looking for the answer, then storing that answer locally inside of our head. This is what we all experience as humans when we gather and process information about the world and life. That information gets stored inside of us and is accessed from somewhere inside of us. We trust how that experience works. This is why so many people get frustrated with voice assistants and don’t trust them. To most users, these assistants get their information from mysterious sources and in mysterious ways, or they just flat out don’t work properly at all. Usually due to connection issues or an inability by the assistant to properly understand and process the context of what the user wants from the experience at that moment. Voice Assistants do not need to be built this way. They could live inside devices and then request the information they need from the internet when the needed data or solution cannot be found locally on-device. This would fundamentally change how we use and perceive voice assistants. Siri On-Device Siri could be rebuilt to live on your favorite Apple devices, instead of living in the cloud and accessing data through random servers in corporate controlled data storage centers. As it stands right now, Siri can’t even do simple math or tell me what my name is. In fact, Siri is completely unusable if it is not connected to the internet. To me, this is not only unacceptable and offers a terrible user experience… It also just doesn’t make much sense. Especially from a company like Apple that specializes in offering a complete software to hardware experience and a reputation for upholding some of the highest levels of security and privacy in consumer tech products. With Siri living locally on a customer’s device, Apple can offer them the comfort of knowing that if they prefer, their data is safe within their device no matter what Siri does or doesn’t do. This could easily be implemented during the on-boarding process when a person first sets up their new Apple device. Siri could be set up as “on-device only”. Apple’s Silicon Advantage One of Apple’s greatest strengths and differentiators within the mobile market over the last couple of years has been their custom silicon. By making their own industry-leading processors and hardware components, they have a unique advantage in creating small but powerful devices. A custom multicore chip could be created specifically as Siri’s “brain”, more or less. Siri could use this new chip to process data very quickly and without an internet connection. Apple’s ability to create extremely fast, efficient and purpose-built silicon would be a huge advantage over competitors like Amazon’s Alexa, Microsoft’s Cortana, or the Google Assistant. A multitude of functions could easily be handled locally on-device without ever needing access to the internet. Basically, any function that your device could perform normally without an internet connection, should also be able to be handled by Siri. With this kind of deep functionality to control and manage a user’s device and personal information without being connected to the internet, a customer can know with 100% certainty that their data is safe and private. This is the kind of user experience that loyal Apple customers have grown accustomed to and it is the kind of forward thinking that Apple has built their reputation on. A Siri App and “Cores” Amazon’s Alexa and Microsoft’s Cortana both use “Skills” to add functionality in what each voice assistant can do, while the Google Assistant does something similar with it’s “Services”. Apple could give Siri increased functionality by letting third-party developers create “Cores”. Each Core would plug into Siri by downloading and installing it from within the App Store and then the functionality of each Core would be managed within a new app dedicated to Siri accessible from your iPhone, iPad, Mac, or iCloud on the web. This would be a great way for third-party developers to expand Siri’s “Core” functionality and provide a unique experience which you could not get from a competing voice assistant. For instance, subsets of Wikipedia could be installed as separate Cores to give Siri local access to vast amounts of data on almost any subject imaginable. You want Siri to learn about Animals? Install the Wikipedia Animals Core. Want Siri to learn and teach you Spanish? Install a Spanish Language Teacher Core. Remote Start your Tesla? Install the Tesla Core. These Cores could update periodically or at regularly scheduled intervals when you are on wifi in order to keep the data up-to-date and fresh. For instance, every night when I go to sleep at 11:00 pm Siri could be scheduled to update her/his Core Database so that when I wake up Siri is fresh and ready to go. Or, a user could tell Siri to update a specific Core on-the-fly by saying something like “Hey Siri, update your Wikipedia Cores”. You get the idea. Anything you can think of can be packaged as a Core app and installed into Siri. Not only would this create a great user experience because Siri will work blazing fast, but Siri’s functionality can be easier to manage through the Core management system located inside a Siri app. Also, having Siri access and keep data on-device would create a huge demand for increased internal memory storage, which would drive up the average price people spend on their iOS devices in particular. Currently, Apple offers iOS devices with up to 512gb of internal storage and the sales of these more expensive models would only increase as consumers experienced how useful the new on-device Siri would be. Apple’s Unique Market Position Amazon, Google, and Microsoft all have one thing in common that separates them from Apple. They all want your data and they all want your data to live on their servers. They make their money from your data and the services that use your data. This is why these companies have no good incentive to create a voice assistant that lives on-device and processes all or most of the data locally. Apple doesn’t want or need your data to make their money. Apple wants to sell you premium devices and they have built a reputation from their strong stance on security and protecting their customers’ privacy. This fact combined with Apple’s ability to make their own chips creates a unique opportunity for Apple in creating an on-device voice assistant that respects a user’s privacy and securely processes all the data without the need for constantly pinging an external server. Imagine a new Siri Imagine for a moment that Apple has a huge public event and during the keynote, they announce a brand new Siri… “The new Siri Chip functions as the brain which handles all of the functions that Siri uses to process and complete your requests, all locally and on-device. Private. Secure. You decide what Siri can access and can’t access. You can feel safe knowing that Siri is with you at all times, even when you aren’t connected to the internet. We feel that the new Siri completely revolutionizes how people can use voice assistants in their lives. Now, Siri acts more like a real assistant. One that CAN access the internet, but that can fully function without it. This is the new Siri and we think you are gonna love it.” Future Potential The idea of having a personal voice assistant that lives on-device has so much potential that it is easy to imagine the many ways in which this concept could evolve. For instance, Apple could offer different Siri “Personas”, each one giving Siri a different voice and personality when you interact with the assistant. Apple could create a brand new product that acts a home server, router, Apple TV, and Siri database all-in-one device. This theoretical device could be used to store a huge amount of databases that Siri could tap into, but which would still be controlled by the user. This product could store Apple Music downloads, photo libraries, videos, Siri Cores, iCloud files, etc. and serve them up to the user through Siri or directly in-device. Final Thoughts One thing is certain, Apple is barely tapping into the potential of Siri and its unique position within the voice assistant market. Siri should currently be able to do things that are quite simple to pull off for a company with Apple’s resources. I should be able to say… “Hey Siri, play Stranger Things on the living room tv”. Or… “Hey Siri, close all of my apps”. In my opinion, Apple should be creating an entire voice based operating system for utilizing Siri functionality (siriOS?) and there really isn’t a good excuse for not being further along than they currently are. I have high hopes for Apple and what they can do as a technology company. It is my favorite brand and I am excited to see where they take Siri in the future, I just hope that they realize the full potential of voice-operated computing before their competition gets too far ahead of them.

Apple has a unique opportunity to revolutionize voice assistants

0

apple-has-a-unique-opportunity-to-revolutionize-voice-assistants-153d0d60124

2018-05-15

2018-05-15 18:31:07

https://medium.com/s/story/apple-has-a-unique-opportunity-to-revolutionize-voice-assistants-153d0d60124

false

1,800

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Justin Thiel

Life Enthusiast / Husband / Father

4cbd147bfcda

justinthiel

6

38

20,181,104

null

0

null

0

32881626c9c9

2018-04-20

2018-04-20 02:38:08

2018-04-20

2018-04-20 03:08:27

1

false

en

2018-04-20

2018-04-20 05:31:52

21

153db981e6dc

3.392453

1

0

Whether we like it or not, AI (Artificial Intelligence) and robotics are going to shape our future. There are 10 issues that professionals…

5

Musings on Some Digital Issues Whether we like it or not, AI (Artificial Intelligence) and robotics are going to shape our future. There are 10 issues that professionals and researchers need to address in order to design intelligent systems that help humanity. Misinformation and Fake News The flow of misinformation together with our natural inability of perceiving reality based on evidence (a phenomenon called confirmation bias) is a threat to having an informed democracy. Russian hackers influencing the US elections is only one example of how social media can massively spread misinformation and fake news. Recent advances in computer vision make possible to completely fake a video of famous individuals It is an open question how institutions are going to address this threat. Job Displacement The scientific revolution in the 18th century and the industrial revolution in the 19th marked a complete change in society. For thousands of years before it, economic growth was practically negligible. During the 19th and 20th century, the level of society development was remarkable. In the 19th century there was a group in the UK called the Luddites, that protested against the automatization of the textile industry by destroying machinery. Since then, a recurrent fear has been that automation and technological advance will produce mass unemployment. Even though that prediction has proven to be incorrect, it is a fact that there has been a painful job displacement. PwC estimates that by 2030 around 30% of the jobs will be automatized. Under these circumstances, governments and companies should provide workers with tools to adapt to these changes, by supporting education and relocating jobs. Privacy The importance of privacy is all over the news lately due to the Cambridge Analytica scandal, where 87 million Facebook profiles were stolen and used to influence the US election and Brexit campaign. Privacy is a human right and should be protected against misuse. Cybersecurity Cybersecurity is one of the biggest concerns of governments and companies, specially banks. AI can help protect against these vulnerabilities, but it can be also used by hackers to find new sophisticated ways of attacking institutions. Mistakes of AI Last month, a woman was hit and killed overnight by an Uber self-driving car when walking across the street in the US. As any other technological system, AI systems can make mistakes. It is a common misconception that robots are infallible and infinitely precise. Military Robots There is an ongoing debate about controlling the development of military robots and banning autonomous weapons. An open letter, from 25.000 researchers and professionals of AI, asks to ban autonomous weapons without human supervision to avoid an international military AI arms race. Algorithmic Bias We have to work hard to avoid bias and discrimination when developing AI algorithms. An specific example was face detection using Haar Cascades, that has a lower detection rate in dark-skinned people than in light-skinned people. This happens because the algorithm is designed to find a double T pattern in a grayscale image of the person’s face, corresponding to the eyebrows, nose and mouth. This pattern is more difficult to find in a person with dark skin. Haar Cascades are not racists, how can an algorithm be? When programing these algorithms, we need to be mindful of their limitations, transparent with users by explaining how the algorithm works or use a more effective technique with dark-skinned people. Regulation Existing laws have not been developed with AI in mind, however, that does not mean that AI-based product and services are unregulated. Governments must balance support for innovation with the need to ensure consumer safety by holding the makers of AI systems responsible for harm caused by unreasonable practices. Policymakers, researchers and professionals should work together to make sure that AI and robotics provide a benefit to humanity. Superintelligence Some tech leaders have shown concerns about the possible threats of AI, one example was Elon Musk, who claimed that AI is more risky than North Korea. These words generated a strong criticism from the scientific community. Superintelligence is generally regarded to a state where a robot starts to recursively improve itself, reaching a point that easily surpass the most intelligent human by orders of magnitude. Although this may sound as impossible, we still need to take the necessary steps. Robot Rights Should robots have rights? If we think of a robot as an advanced washing machine, then no. However, if robots were able to have emotions or feelings, then the answer is not that clear. A suggestion in the debate around robot rights is that robots should be granted the right to exist and perform their mission, but this should be linked to the duty of serving humans. There is a lot of controversy around this area. Meanwhile, in 2017, Sophia the robot was granted the citizenship of Saudi Arabia. Given these facts, the question we should ask is not whether we should take any action or not, yet which actions shall we take upon the increasing number of issues with regard to the proliferation of AI into our daily lives.

Musings on Some Digital Issues

46

musings-on-some-digital-issues-153db981e6dc

2018-04-20

2018-04-20 12:20:14

https://medium.com/s/story/musings-on-some-digital-issues-153db981e6dc

false

846

Data Driven Investor (DDI) brings you various news and op-ed pieces in the areas of technologies, finance, and society. We are dedicated to relentlessly covering tech topics, their anomalies and controversies, and reviewing all things fascinating and worth knowing.

null

datadriveninvestor

null

Data Driven Investor

info@datadriveninvestor.com

datadriveninvestor

CRYPTOCURRENCY,ARTIFICIAL INTELLIGENCE,BLOCKCHAIN,FINANCE AND BANKING,TECHNOLOGY

dd_invest

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Daily Wisdom

null

ddd120ae7c2

dailywisdom

60

0

20,181,104

null

0

null

0

c8005b22fde

2018-05-08

2018-05-08 15:48:56

2018-05-08

2018-05-08 15:59:22

1

false

en

2018-05-08

2018-05-08 15:59:22

3

153ea727d454

1.50566

0

H&M is moving away from their usual approach to retail by monitoring customer data and implementing artificial intelligence in an attempt…

3

H&M Uses Big Data and AI to Tailor Store Offerings Photo by Ibrahim Boran from Pexels H&M is moving away from their usual approach to retail by monitoring customer data and implementing artificial intelligence in an attempt to comeback from their poor sales slump.Wall Street Journal reporter Saabira Chaudhuri reports that the retail giant is moving towards customizing which products are offered in each of their 4,293 stores. An important shift from their previous mindset geared towards stocking similar products worldwide. Chaudhuri goes on to note that this evolution was likely motivated by the 10 straight quarter slump in sales H&M has experienced. Factors that influence this include a movement away from in-store shopping in favor of online, competition from digital startups and the large discounts they’ve repeatedly had to do in order to clear out $4 billion worth of unsold merchandise. Previously, H&M relied heavily on designers to innovate and produce items consumers would buy. But by using algorithms to analyze returns, receipts and loyalty card data they can better predict and meet customers needs and hopefully reduce the amount of discounts needed to sell goods. H&M’s bid on data is unique to the retail industry. Former retail executive Ludovica Dodero is quoted as saying: “Most companies are still used to taking decisions not driven by analytics at all, only based on the experience of the manager” Despite no longer depending on the intuition of their managers and merchandising heads, H&M realized that this change doesn’t mean they have to replace those positions. Instead they are focusing on providing those merchandising heads with the data and tools to better inform decisions moving forward. Arti Zeighami, head of H&M’s advanced analytics and artificial intelligence, notes that this shift is going to help the company “…be sharper, more accurate and hyper-relevant, not have one solution fits all.” I am interested to see how this move will impact the overall approach of other retailers. As competition continues to grow, putting customer centricity at the forefront will be what helps certain companies stand out and offer value both to investors and shoppers alike.

H&M Uses Big Data and AI to Tailor Store Offerings

0

h-m-uses-big-data-and-ai-to-tailor-store-offerings-153ea727d454

2018-05-08

2018-05-08 15:59:24

https://medium.com/s/story/h-m-uses-big-data-and-ai-to-tailor-store-offerings-153ea727d454

false

346

Tools and strategies to become a leader in Customer Experience.

null

worthix

null

Worthix

blog@worthix.com

worthix

CUSTOMER EXPERIENCE,CX,CUSTOMER SUCCESS,MARKETING STRATEGIES,MARKET RESEARCH ANALYSIS

worthix

Retail

retail

Retail

16,358

Dayana A.

Customer Success Specialist and Junior Content Producer at Worthix. Focus on future and present trends in customer experience.

9c622bfe41e4

dayana.aparicio

8

9

20,181,104

null

0

null

0

null

2018-04-19

2018-04-19 14:28:50

2018-04-19

2018-04-19 14:38:38

1

false

en

2018-04-19

2018-04-19 14:38:38

0

153f582d7dbb

4.94717

0

People tend to fear things that they themselves can not control or do not understand. Artificial Intelligence(AI) is a prime example of…

5

Should Artificial Intelligence Be Regulated? People tend to fear things that they themselves can not control or do not understand. Artificial Intelligence(AI) is a prime example of this. In recent years there has been an increase in the usage of AI which has made many people skeptical of its potential. While artificial intelligence does not currently pose an imminent threat to society it should be regulated. It is the responsibility of both developers and legislatures to do so. Each group within their own spheres of jurisdiction. According to “The Artificial Intelligence Act of 2017–2018”, artificial intelligence is “Any artificial systems that perform tasks under varying and unpredictable circumstances, without significant human oversight, or that can learn from their experience and improve their performance.” This includes, “Systems that think like humans, such as cognitive architectures and neural networks” (H. 4625, 2017). The bill also mentions that any system that passes the Turing test is to be considered artificial intelligence. The Turing test requires the machine to be indistinguishable from a human when asked a series of questions. Legislatures disagree as to whether artificial intelligence can be safely used, whether we can trust the neural network technology itself, and whether it can be safely controlled. Some lawmakers, who want to protect the freedoms of the developers say that “the futuristic fear of existential risks does not justify the overall regulation of development” (Wallach 1). Both parties bring some key points to the table. While artificial intelligence is a threat and could eventually lead to catastrophe, it is also an amazing technology with countless applications. According to Finale Doshi-Velez in “Artificial intelligence is more powerful than ever. How do we hold it accountable?”, artificial intelligence is “used for tasks such as analyzing MRIs, giving financial advice and even composing music.” With all of these amazing applications, it would be a shame to put an end to the development of artificial intelligence entirely. Artificial intelligence poses a number of risks to our society. These risks can be put into three main groups: national security and safety, job markets, and the inevitable singularity instance. Over time as technology gets more advanced, one or a combination of all three events could lead to the eventual downfall of society. For this reason, legislatures must regulate artificial intelligence. Many people fear the use of “human out of the loop” systems where a robot could make life or death decisions. Systems like these could create a national security concern and lead many to question the ethics of the military. Should a robot be given the power to take a life without human consent? Wendell Wallach in “Eyes on AI” makes the point that, “Current learning systems are black boxes, whose output can be biased, whose reasoning cannot be explained, and whose impact cannot always be controlled.” Furthermore, many developers believe that “black box” technologies such as artificial intelligence are becoming increasingly abstract and harder to control. Doshi-Velez, Finale, and Mason Kortz in “Artificial intelligence is more powerful than ever. How do we hold it accountable?” help to further explain this point when they say that, “algorithms, especially those developed using advanced machine learning techniques, like deep neural networks, can be so complex that not even the designers fully understand how they make decisions.” For these reasons, some lawmakers believe the answer is no, robots should not be able to take a life, and are lobbying for “human in the loop” regulations. Others say that regulation of AI in the military would prevent the US from competing with other countries. Vladimir Putin, president of Russia, says that, “Whoever becomes the leader in this sphere will become the ruler of the world”. The next space race is artificial intelligence. Legislatures must create laws that enable companies to compete on an international scale while protecting American interests. Experts predict that one third of today’s jobs will be replaced with robots by the year 2025. This includes machines that do not use AI technology. In, “Tech companies should stop pretending AI won’t destroy jobs”, Kai-FU Lee says that “AI will displace a large number of jobs, which will cause social discontent.” Lee predicts that in the future it may become normal to be unemployed. Some legislatures have proposed a universal basic income for those who have had there jobs taken by robots (Lee 3). In America especially, public skepticism of mass job loss has led many groups to lobby for the ban of AI technologies. Kai-Fu Lee says that, “Truckers are now appealing to President Donald Trump and Congress to stop testing of autonomous trucks.” It is complications such as this that have allowed countries such as China to surpass the US in the development of artificial intelligence. According to Amitai Etzioni in “Should artificial intelligence be regulated?” the singularity instance will occur “after reaching a point of ‘technological singularity,’ computers will continue to advance and give birth to rapid technological progress that will result in dramatic and unpredictable changes for humanity”. Amitai Etzioni also makes the point that “AI is on the path to turning robots into a master class that will subjugate humanity, if not destroy it.” Essentially, the fear is that artificial intelligence in pair with a rapid growth in computing power will allow robots to overtake mankind. Some people may immediately dismiss this claim as science fiction. Others, especially those who work in the tech industry, see the singularity instance as a real threat and are pushing for regulation. Accidents are inevitable, especially those relating to a technology as volatile as artificial intelligence. An autonomous Uber manufactured by Forb in pair with Argo AI recently struck and killed a woman in Tempe, Arizona (Doshi-Velez 3). This incident caused many to lose their trust in AI. In mere seconds trust that was gained after years of development, and eventually led lawmakers to allow fully autonomous vehicles to travel on roads was lost. In order to protect the future of artificial intelligence lawmakers must dismiss freak accidents such as this and focus on all of the good that AI has in store. It is the technology industry’s duty to use artificial intelligence responsibly. It is the programmer as an individual’s duty to never do unethical tasks even if their employer assigns it to them. Nevertheless, it is the duty of the tech industry to defend its freedom to create new technologies despite the public’s skepticism. Accordingly, if the groups abide by these roles there will be balance throughout the industry. Currently, legislators rely on laws, regulations, and industry oversight in order to regulate AI. This approach is inadequate and will soon become nullified. It is not the responsibility of the lawmakers to understand AI. This duty should be left to the developers. It is the lawmakers’ duty to create “An enforcement regime to ensure that industry acts responsibly and that critical standards are followed will also be required” (Wallach 1). The lawmakers job is not to directly regulate artificial intelligence. The lawmakers must put laws into place that hold the technology industry accountable and must avoid hindering development. It is the responsibility of both the programmer and the legislators to regulate artificial intelligence. While technological advances will require the government to further regulate artificial intelligence, it is the job of the programmers to protect the future of technology from tyranny. The current generation will be the remembered by future generations for their advances in artificial intelligence. Lawmakers and developers must reach a consensus as to where to draw the line between freedom and security.

Should Artificial Intelligence Be Regulated?

0

should-artificial-intelligence-be-regulated-153f582d7dbb

2018-04-19

2018-04-19 14:38:39

https://medium.com/s/story/should-artificial-intelligence-be-regulated-153f582d7dbb

false

1,258

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Logan O'Neal

loganoneal.com

42abef6cb454

loganofneal

2

1

20,181,104

null

0

null

0

null

2018-01-08

2018-01-08 01:01:35

2018-01-08

2018-01-08 02:01:30

1

false

en

2018-01-08

2018-01-08 02:01:30

4

153f89fd8fa0

1.339623

1

0

Why this?

3

A Word on AI #1 — January 7th 2018 Why this? My hypothesis is that simply writing down what caught my eyes, week after week, will help me improve and educate myself. My two goals in sharing this are: 1/ inspire you to do the same exercise, 2/ share resources that are tremendously impactful in the way I see the world change. — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — Best of what I consumed this week Venture Beat — “Human involvement is still a necessary component of valuable AI” — Let’s pretend we all start the year and pretend as if people still mater. AI exists but it is still early stage and people play a central role into designing and implementing well thought AI algorithms. Techcrunch — “Rubikloud scores $37 million investment to bring intelligence to retail industry” — A retail AI startup from Toronto raised a $37mm series B to help retailers to manage promotions and loyalty programs. Interesting move for Intel, leading the round through its CVC, to refinforce its presence in the Internet Of Things market. Forbes — “ Predictions For Artificial Intelligence In 2018” — in 2 words, AI will change the way we work and AI will be subject to more scrutiny. As we broadly agree, we do feel it is still too early for AI to transform in society in 2018. Xinhuanet — “Beijing to build technology park for developing artificial intelligence” — China is bullish on AI. With a $2b investment in a tech park, this is only one of the many initiatives deployed by China to attract AI talents and create innovative companies. Definitely more of that coming in 2018…

A Word on AI #1 — January 7th 2018

1

a-word-on-ai-1-january-7th-2018-153f89fd8fa0

2018-01-08

2018-01-08 02:17:51

https://medium.com/s/story/a-word-on-ai-1-january-7th-2018-153f89fd8fa0

false

302

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

A Word on AI

A Word on AI is a collection of articles and discussions on Artificial Intelligence with a business approach — from investing to entrepreneurs stories —

c7a85f2928c0

awordonai

3

8

20,181,104

null

0

null

0

null

2018-02-24

2018-02-24 18:53:35

2018-02-24

2018-02-24 19:02:15

4

false

en

2018-02-24

2018-02-24 19:02:15

0

153f9066ddc5

2.892453

0

Just started to read the book “The Signal and the Noise” by Nathan Silver and getting inspired, though I am still at the first pages and…

4

Some thoughts about “Predictions” Just started to read the book “The Signal and the Noise” by Nathan Silver and getting inspired, though I am still at the first pages and the following does not appeared so far. So let me put down some thought about prediction and put them into the three categories interpolation, regression and classification. How could we put prediction in a formula: (q1) f(x) is known on an interval [0,t] and we are interested in the value f(t+d) with d>0 Most of the time we look at a discrete time interval, so (q2) f(x) is known for {0,…,n} and we are interested in the value f(n+d) with d a positive natural number Depending on the question it is a one-time efford like predicting the outcome of an election or a continuous process of predicting the stock prices for the next day. This is for time prediction, another question might be how a function behaves for some unknown observation, so (q3) f(x_i) is known for i in {1, …, n} and we want to know f(y) with y not in {x_i, i=1, …, n} There is this little difficulty in both cases that we do not know on what information the outcome depends. In the third question we should ask in what room the x_i live, so what is the definition space of f. The same rule applies to q1 and q2, since we should split x in the touple (y,t) with t the time and y from some unknown space. We should also think about the value space of the function. Is it finite, an interval, real, multi-dimensional. Now the art of prediction is actually very old, a lot of really smart people have thought about it and came up with several ideals, depending on the definition and value space. Though a lot of attention is put on prediction algorithms in the last years with the massive increase in computational power. Basically we organize the predictions methods by definition space value space set of functions Interpolation First simplification is to pretend the definition space of f is known (and has finite dimension). Then we speak of an interpolation problem, thus matching a continuous function to the given observations. Usually we have a class of function we try to fit, e.g. polynomials up to degree n. Another approach is to just locally fit the observations, the most popular method is spline-interpolation. Depending on the class of functions the prediction can be very different. Regression A more realistic approach is to accept we do not know everything and introducting an error function epsilon, that has some random distribution to account for the unknown. For ease of use — but not very realistic — the error is assumed to be normally distributed, though there are methods to allow other distributions as well. y = f(x)+eps The most popular method or better class of methods is “linear regression”, roughly speaking putting a line (therefore linear) in the plot with dimensions x and f(x) with minimizing the error function. Classification If the value space is finite, we consider it a classification problem. There is a lot of attention to algorithms tackling these kind of problems as we now are in the field of machine learning. Think of image or speech recognition. As said there are a ton of different algorithms: binomial regression, decision trees, support vector maschines, clustering, neural networks, information filtering systems just to name the most important ones.

Some thoughts about “Predictions”

0

some-thoughts-about-predictions-153f9066ddc5

2018-02-24

2018-02-24 19:02:16

https://medium.com/s/story/some-thoughts-about-predictions-153f9066ddc5

false

581

null

Machine Learning

machine-learning

Machine Learning

51,320

Holger Aust

null

30b2dd57b071

databraineo

1

20,181,104

null

0

null

0

592e0f336366

2018-05-22

2018-05-22 13:48:48

2018-05-22

2018-05-22 13:51:46

1

false

en

2018-05-22

2018-05-22 15:38:47

12

153f9c7912eb

3.090566

4

0

From the time that Blockchain technology started to become increasingly popular and the development of cryptocurrency started to open the…

5

Protecting Yourself From Cryptocurrency Fraud From the time that Blockchain technology started to become increasingly popular and the development of cryptocurrency started to open the floodgates for a completely new tier of assets, there have been several skeptics that have spent their time seeking out and discovering an entirely new category of scams — crypto scams. Although the supporters of Blockchain technology and cryptocurrencies focus on the massive disruption that is taking place in financial institutions and the technology sector as a whole, there are still many unknown factors that have to be assessed before moving forward with certainty, and overcoming the new generation of scammers is one of those factors. Due to the fact that cryptocurrency has a highly transferable nature, combined with the fact that Blockchain technology is intensely complex, creates an environment that is wrought with scammers that base their attempts at success on directly targeting the new and unsuspecting cryptocurrency holders. Because of the newness of the market, it is easier for these individuals to come create intricate and intellectualized scams that will separate you from your money without anyone being the wiser, until it is simply too late. These scams are easy to miss and hard to avoid unless you are aware of what you are looking for, so we’ve put together the three most common types of crypto-scams and will break down a few telltale signs to help you notice and avoid them all together. Twitter/Telegram It is common for scammers to hone in on and target high-profile cryptocurrency and/or Blockchain personalities and profiles in and attempt to gain access to their followers ‘Wallets’ by convincing them to submit their wallet address. This is oftentimes done by using a number of bots to develop statements that appear to be coming from the actual Twitter personality. Following the ‘successful’ sharing of information, it then looks as though other users are experiencing success with receiving their ETH, however, those ‘successful responses’ are simply additional bots aiding in the scam. In order to be on the safe side of things, it is important that, as a general rule of thumb, you do not submit any personal information and/or send any cryptocurrency to wallet addresses that you’ve found on social media. Cryptocurrency Phishing Scams Another highly common scam that is going around the cryptocurrency space is phishing. When you’re intertwined in a phishing scam, the scammers typically ‘clone’ an exchange website, hosting it on a domain that looks like a near replica of the original. Moreover, because of the additional reach that the internet provides, scammers have also begun to pay for Google advertising in order to make their site look more reputable than the others, drawing in more customers to their platform and, in turn, stealing most of, or even all of their cryptocurrency. It is important that you NEVER provide your private keys to anyone and always triple check that the URL you are using is correct and is accompanied by the green “secure” emblem as well. Initial Coin Offering Scams ICO’s have become one of the more disruptive applications that have appeared since the inception of Blockchain technology and although there are definitely some amazing opportunities out there involving the technology, they are also a prime target for scam artists. One of the more common types of ICO scams has been named the ‘exit scam’ and occurs when a fraudulent ICO is set up and proceeds to sell worthless tokens to their investors, disappearing once they have reached their capital investment goals. These types of ‘exit scams’ have been prevalent across the cryptocurrency ecosphere, however, there are a few warning signs that might help you to distinguish between a valid ICO and a fraudulent one. 1: Lack of Whitepaper: Depicts a lack of project validity 2: Minimal Online Presence: No way to verify team members 3: Profit Guarantees: Promises of guaranteed profits or returns 4: Imbalanced Premines: Massive amounts of team capital before an ICO goes live 5: Lack of MVP and Roadmaps: Signifies indifference in the project 6: Unknown Advisory Team: Lack of solid advisory team members Overall, the scams that we see in the market are going to continue to evolve as the technology becomes a greater force behind our everyday lives, so it is imperative that you remain vigilant when ensuring your online safety and security. For more information and to see what else is going on @ IAGON, please follow us at the social media links below, or head over to the IAGON Website! Facebook, Instagram, LinkedIn, Steemit, Reddit Bitcointalk, Twitter, Telegram, Youtube, Medium, Github

Protecting Yourself From Cryptocurrency Fraud

62

protecting-yourself-from-cryptocurrency-fraud-153f9c7912eb

2018-05-29

2018-05-29 10:03:35

https://medium.com/s/story/protecting-yourself-from-cryptocurrency-fraud-153f9c7912eb

false

766

Iagon is a platform for harnessing the storage capacities and processing power of multiple computers over a blockchain grid. Secured and encrypted platform that integrates blockchain, cryptographic technologies & AI, enhancing the overall usability.

null

IagonOfficial

null

Iagon Official

navjit@iagon.com

iagon-official

ARTIFICIAL INTELLIGENCE,CLOUD COMPUTING,BLOCKCHAIN TECHNOLOGY,CLOUD STORAGE,ICO

IagonOfficial

Bitcoin

bitcoin

Bitcoin

141,486

Rose Marie

Project Lead/Content Director @ IAGON

b7347cf5ce45

rosemariewritenow

68

8

20,181,104

null