Alaamer/medium-articles-posts-with-content

0

null

0

null

2018-08-21

2018-08-21 16:45:46

2018-08-21

2018-08-21 17:07:58

5

false

en

2018-08-24

2018-08-24 12:30:46

3

16865af70232

1.584277

0

Over the weekend my fiancee came to me with a problem, she had a bunch of Vcard (.vcf) files and she needed them tabulated into Excel.

5

Converting Vcards(.vcf) to CSV Files Over the weekend my fiancee came to me with a problem, she had a bunch of Vcard (.vcf) files and she needed them tabulated into Excel. I took a look around for any existing .vcf libraries, and while there were some .vcf libraries out there (https://pyvcf.readthedocs.io/) I found that these were really geared towards the more complicated use cases of .vcf files, and not the quick and dirty contact card we all know and love. So I wrote a quick program that would do just that. (please note, if you are an expert in how vcf files work, feel free to pass on the rest of this page) In investigating vcf files, I discovered that they are tagged, much like HTML or XML. My goal was to: Get the list of files Function to read in the file names 2. Read the files, and parse the tags, into a dictionary Reading in the files, splitting on the ‘:’ delimiter, and placing into a dictionary 3. Clean the remaining entries 4. Create a pandas data frame file with the cleaned data. 5. Use the pandas DataFrames to_csv function to create the new csv file Reading in the filename from the command line, calling the functions, and creating the new CSV files For more information, and the source code checkout my GitHub: https://github.com/ablitstein/Vcards-to-CSV Alex Blitstein, Data Science Fellow at Galvanize, http://linkedin.com/in/ds-blitz

Converting Vcards(.vcf) to CSV Files

0

converting-vcards-to-csv-files-16865af70232

2018-08-24

2018-08-24 12:30:46

https://medium.com/s/story/converting-vcards-to-csv-files-16865af70232

false

199

null

Data Science

data-science

Data Science

33,617

Alex Blitstein

A Data Scientist/Machine Learning Engineer with interest in Deep Learning, and using Data and Technology to solve real world problems.

db69d89b9bfc

ablitst1

0

2

20,181,104

null

0

null

0

null

2018-08-24

2018-08-24 15:48:34

2018-08-23

2018-08-23 15:33:42

1

false

en

2018-08-24

2018-08-24 15:49:13

3

1686a388545e

2.758491

1

0

Today’s Role Model is Marnie Hogue. In college, Marnie majored in math and began her career analyzing data for both academic and non-profit…

5

Role Models: Marnie Hogue, Data Analyst at Plated Today’s Role Model is Marnie Hogue. In college, Marnie majored in math and began her career analyzing data for both academic and non-profit institutions. She worked her way up to a Director of Research position but wanted to make a change. By building her programming skills, she navigated the transition and landed a data role in the tech world. Marnie is currently a data analyst for Plated, an ingredient and recipe meal kit service. What’s your official title and how long have you been in this role? I’m a Data Analyst at Plated and I’ve been in this role for 1.5 years. What attracted you to this role? I appreciate the ability to work with people/across teams to solve interesting business problems and help them make decisions smarter, faster and with more confidence. Also I like the challenge of refining data into a usable format that is interpretable- sometimes it can be a design challenge and the result can be beautiful! Walk me through a typical day in your role. What activities do you engage in? What types of meetings do you join? When’s lunch? A huge part of my job is working with the culinary team and so I join a lot of their planning meetings and help them interpret customer feedback. I have weekly 1–1s with different product managers and work with them to define and measure KPIs that are important from a business lens. Lunch is usually around 1:30, and often it involves recipes from the test kitchen! What skills/technologies help you succeed? I rely heavily on my ability to communicate complex ideas in a simplified way, to lean on a team for support when facing ambiguous problems, and to make a recommendation based on data. In terms of technologies, I use Tableau for data visualizations, SQL for pulling data, and Python for manipulating data. What’s the most fun or creative part of your role? I love testing different hypotheses about our customers culinary preferences! The results and findings are not always obvious, so this work is interesting. What are the biggest challenges you face in this role? I find it difficult to push back on work that is not important to our business and only working on those projects that really matter. A lot of time, well-intentioned colleagues ask for data they don’t need or will not use to make a decision. There are only so many hours in the day, so our team needs to leverage our resources accordingly. What teams/individuals do you work with cross-functionally? Can you give an example of a time when you collaborated with another group/individual? I work most closely with Plated’s culinary team and our product teams, I work every day with these teams to help them figure out what makes a good recipe/ menu. What’s an area where you’re trying to grow in your role? I’m building confidence in my recommendations- my ability to form an opinion on what my produce manager should do and then let them react to it. Aside from technical skills, what personality traits/characteristics make for an ideal candidate in your role? Data analysts on our team need to feel comfortable problem solving in a collaborative setting, managing independent projects that are longer-term, and remaining curious about why things are the way they are. Generating key questions and hypotheses helps support the other teams. What skills (tech/non-tech) have you improved as a result of working in this role? I’ve learned to manage up and set appropriate expectations. I also greatly improve my technical skills, especially my knowledge of Python. In your role, what metrics define success? For my role, it’s necessary to have a strong understanding of SQL/Python, the skills to write code, and the ability automate processes. I also need to make data “products” that influence the business, hopefully from a measurable financial standpoint! Want more of these interviews delivered directly to your inbox? Sign up for my monthly newsletter. Originally published at meghan-duffy.com on August 23, 2018.

Role Models: Marnie Hogue, Data Analyst at Plated

10

role-models-marnie-hogue-data-analyst-at-plated-1686a388545e

2018-08-27

2018-08-27 17:22:02

https://medium.com/s/story/role-models-marnie-hogue-data-analyst-at-plated-1686a388545e

false

678

null

Data Science

data-science

Data Science

33,617

futureproofmeg

Career Strategist for the Tech/Startup World. The artist formerly known as Lady Collective. Reading, writing, and running between snacks.

e4973074f3bb

ladycollective

344

139

20,181,104

null

0

null

0

634d4b270054

2018-04-12

2018-04-12 09:33:09

2018-04-12

2018-04-12 09:35:16

1

false

en

2018-06-05

2018-06-05 08:53:19

3

1689af13e8

1.098113

1

0

The latest drone-related patent has been awarded by the United States Patent and Trademark Office (USPTO) to the HorseFly Truck Launched…

5

USPTO Issued Drone Delivery Patent To HorseFly The latest drone-related patent has been awarded by the United States Patent and Trademark Office (USPTO) to the HorseFly Truck Launched Drone Package Delivery System from the Workhorse Group. Numbers of companies are coming forward and getting on board and the drone delivery patents that have been awarded is going up as well. “We feel that the patented HorseFly truck launched drone package delivery system is the first major change to the last mile delivery process since the invention of the package delivery truck. Drivers appreciate the fact that the HorseFly system is fast, reliable, and efficient and last mile package delivery is changing, and the HorseFly delivery system is leading the way,” said Steve Burns, Workhorse CEO. Studies have shown that last-mile drone delivery can be both more-efficient and greener than deliveries via truck. Source: https://bit.ly/2uLDWSn 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.

USPTO Issued Drone Delivery Patent To HorseFly

1

uspto-issued-drone-delivery-patent-to-horsefly-1689af13e8

2018-06-05

2018-06-05 08:53:20

https://medium.com/s/story/uspto-issued-drone-delivery-patent-to-horsefly-1689af13e8

false

238

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

2018-04-17

2018-04-17 19:17:07

2018-04-18

2018-04-18 20:11:08

7

false

en

2018-04-18

2018-04-18 20:12:45

2

168a6077946e

3.038679

3

0

kNN is the simplest machine learning algorithm to understand and also to explain.

4

Simple and In depth introduction of kNN(k-Nearest Neighbor) : “Two rows of boathouses connected by a wooden walkway” by Ishan @seefromthesky on Unsplash kNN is the simplest machine learning algorithm to understand and also to explain. It is a versatile algorithm i.e. useful for both classification and regression. It has one big advantage is that kNN ha no pre assumption about the data. “ Let the data speak for itself ”. Lets define kNN : kNN is non-parametric, instance based, lazy algorithm and used in the supervised setting. Ohh! Lots of jargon…. Don’t worry . Let me explain them one by one : — Non-parametric : It means that algorithm has no pre assumptions about the functional form of the model, to avoid mismodeling . To know more about Parametric and Non-Parametric Algorithm. Instance based : It means that our algorithm does not explicitly learn a model. Instead, it memorize the training instances which are subsequently used as “knowledge” for the prediction. Lazy algorithm : It means that it does not use the training data for the Generalization i.e. these algorithm has no explicit training phase or it is minimal. Training is very fast. You might think , I have taken plenty of time in just explaining definition of kNN. But to know kNN algorithm in depth, every bit of its definition should be clear. Now, Without any further delay lets write kNN algorithm for Classification. kNN Algorithm for Classification : Training element {xi, yi} , Testing point(x) Compute the Distance D(x,xi) to every training element xi. Select k closest instance xi1,xi2,…….., xik and their labels yi1, yi2 …, yik. Output the class y* which is most frequent in yi1,yi2 ……yik. Have you pay attention to the Distance D(x, xi) , You might ask, there are many distance measures available to choose . Lets discuss two important ones: Euclidian (Numerical Attributes): It is most popular and widely used distance measure. It is valid for continuous variables. One drawback is that , it is sensitive to the outliers(single extreme difference attribute). Hamming (Categorical Attributes) : It is used to calculate distance between binary vectors. It is only valid for discrete variables. Other distance measures can be used : Manhattan, KL divergence ,Custom Distance measures (tf.idf for text). Significant of “k” : Value of k has strong effect on kNN performance. k act as controller to decide the shape of decision boundary. Large value of k has following properties: _ _ _ 1. Smoother decision boundary . _ _ _ 2. It provide more voters for prediction, it implies less affect from outliers. _ _ _ 3. As a result has Lower Variance and High Bias. To know more about Bias and Variance. Small value of k has following disadvantages: _ _1. We found Unstable decision boundary. _ _2. Small change in training set implies large change in classification accuracy. How to Select k : The simplest solution is Cross Validation. Best method is to try many k values and use Cross-Validation to see which k value is giving the best result. Cons of kNN Algorithm : It is computationally expensive algorithm.(As it store all training data) High memory requirement. Prediction is Slow for large n (data observations).

Simple and In depth introduction of kNN(k-Nearest Neighbor) :

105

simple-and-in-depth-introduction-of-knn-k-nearest-neighbor-168a6077946e

2018-05-19

2018-05-19 15:16:27

https://medium.com/s/story/simple-and-in-depth-introduction-of-knn-k-nearest-neighbor-168a6077946e

false

527

null

Machine Learning

machine-learning

Machine Learning

51,320

Alok Raj Gupta

Computer Science Undergrad | Machine Learning Enthusiast

31648501d420

guptaalokraj

2

10

20,181,104

null

0

null

0

null

2018-09-01

2018-09-01 02:22:14

2018-09-01

2018-09-01 02:45:23

1

false

en

2018-09-01

2018-09-01 02:47:22

2

168a931541b0

0.690566

0

I’m doing my homework from a class called “environmental modeling”. The homework is designed to help us student understand how genetic…

2

The Learning of Machine Learning Image from ArtStation I’m doing my homework from a class called “environmental modeling”. The homework is designed to help us student understand how genetic programming works and how to train better predictors. Up until one month ago, I had no idea what genetic programming is. As I become more interested in the field of machine learning, more procrastinating on the homework and increasingly indulging myself on web searches, I, probably like a lot of others, have become overwhelmingly drawn by the idea of self learning ML. Upon someone’s comforting assurance that there are ways to become mastery of machine learning without knowing enough math first, I decided to start my journey and have Medium as my studying diary. Wish me luck if you happen to be reading this.

The Learning of Machine Learning

0

the-learning-of-machine-learning-168a931541b0

2018-09-01

2018-09-01 02:47:22

https://medium.com/s/story/the-learning-of-machine-learning-168a931541b0

false

130

null

Machine Learning

machine-learning

Machine Learning

51,320

EC137

null

6cd0e2d78aed

dingdng9

0

5

20,181,104

null

0

null

0

null

2018-06-17

2018-06-17 23:56:14

2018-06-18

2018-06-18 01:18:04

0

false

en

2018-06-18

2018-06-18 01:18:18

0

168ba59a04e3

1.675472

2

0

Alright boys and girls, here’s my first go at it.

2

A simple theory of consciousness / reality, part 1 Alright boys and girls, here’s my first go at it. When I introspect upon my reality, I observe that it has exactly one thought at a time, where each thought is either implicit (a general impression about my state of being or where I’m at, etc), or an explicit statement to myself, where I form my thoughts into words and express them to myself. I have a limit that I am unable to experience more than one thought at a time. I’ve met people who could experience having multiple thoughts at a time, but this seemed to be accomplished by “slicing” time into ever smaller chunks and having seemingly independent thoughts during this time. So there seems to be no reason why this should be a fundamental limitation. But is it? Let us consider a simple of theory of reality where we experience thoughts Thi where i denotes the ith thought that was ever experienced. Which occur for a certain period of time TThi. Then reality is defined completely by the sequence of thoughts {Thi, TThi}. But before we compose beliefs and further inspect the human mind, what is a thought? First, consider neural networks as we know them today. We pass in some input, and we process the neural network and then we get an output. We process the entire neural network. Whether the answer is “obvious”, or whether the answer was derived from some sophistication from observing thousands or millions of different use cases and deciding that one output was slightly likelier than the other. In other words, we run the entire neural network each time, whether the answer is easy, or whether it’s hard. And that’s an incredibly inefficient use of energy. From the earlier study of functional complexity, consider a system that desired to operate with least complexity, in order to operate with least energy and resources. This is likely representative of biological creatures because of the evolutionary pressures of starvation and the operating costs of the brain. There must therefore be a process by which the brain is divided into parts, and approximations of calculations are used in order to arrive at generally correct predictions of the future with minimal cost. The representation that the brain has for itself on what parts are activated is a thought. And thinking is the process by which the complexity of the brain’s calculations are increased. In a subsequent post, I will hope to detail how this structure leads to a natural “mind map”, which can be represented in the brain and can be used as input in a function to create conscious thought.

A simple theory of consciousness / reality, part 1

5

a-simple-theory-of-consciousness-reality-part-1-168ba59a04e3

2018-06-19

2018-06-19 02:33:54

https://medium.com/s/story/a-simple-theory-of-consciousness-reality-part-1-168ba59a04e3

false

444

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Adam Davis

null

48d28ea64902

abriandavis

4

0

20,181,104

null

0

null

0

null

2018-06-05

2018-06-05 17:50:46

2018-06-05

2018-06-05 17:52:18

0

false

en

2018-06-05

2018-06-05 17:52:18

8

168c5ee11979

2.977358

0

In part three of this series, we saw how teachers, media companies, and e-commerce businesses are using AI to process large amounts of data…

5

The Rise of AI Tools: Verticals to Watch, Part 4 In part three of this series, we saw how teachers, media companies, and e-commerce businesses are using AI to process large amounts of data and tailor experiences to their end users (or students, in a teacher’s case). In this post, we’ll be taking a look at how legal and regulatory professionals are using AI to examine risks and improve performance. We’ll also see how those in the science fields are using AI to push their research further. Risk and Regulatory Compliance Folks in this space are probably the most cautious when it comes to AI. One thing they are very aware of is that machines, like humans, make mistakes. However, “they could be different from the kinds of mistakes humans make such as those arising from fatigue, anger, emotion, or tunnel vision,” said Vasant Dhar, Professor of Information Systems at New York University. The stakes are understandably high when these professionals are asked to evaluate state-of-the-art technology. Some current AI applications in this vertical include: Performing audits in real time instead of looking backwards at data to find errors. Making tedious processes like audits less of a burden (one company reduced an auditing process that historically took a week to “just a few hours.” Ultimately, the success of AI in the compliance space will depend on the human beings who develop it. There are immense risks involved for professionals who work to reduce risk for their clients or their companies, and giving up control might be too much for some to bear. Additionally, the AI will need to be taught the difference between compliant and non-compliant behavior, which will evolve and change depending on the company’s needs. Legal According to attorney Mark A. Cohen, “AI is not going to replace lawyers but instead cause lawyers to work differently in the marketplace than they have before.” Forward-thinking firms are already enlisting AI to take over time-consuming, repetitive tasks that were once the domain of overworked, newly-minted attorneys. Searching records, surfacing old cases, fact-checking, and other data-oriented tasks are being handed over to state-of-the-art artificial intelligence. Lawyers aren’t generally known to be early adopters. However, a survey in 2016 found that 52 percent of law firms are embracing new technology to do tedious, time-consuming tasks like collecting data, searching records and going through old cases. Doing grunt work isn’t the only legal application for AI. A Chicago-Kent law professor, Daniel Martin Katz, developed an algorithm to predict decisions in Supreme Court cases. The algorithm was able to predict rulings with a 70 percent accuracy when analyzing 7,700 rulings from a sample spanning 60 years. Another, similar study in the UK found that AI could predict case outcomes with a 96 percent success rate. Science and Research Science is nothing if not data-driven, and every generation of scientists has to position its contributions within the context of its predecessors. Biologists are continuing to unravel genomes and log proteins. Astronomers continue to find new star systems and galactic phenomena. And physicists are untangling the strings that bind our universe together (though maybe we’re all just sitting on quantum foam — it’s unclear). All of that information translates to enormous amounts of data and more is being created every day. In order to continue producing innovative work, scientists have long relied on cutting-edge technology to drive research forward. Today, AI is that cutting-edge technology. Companies like Iris AI are using machine learning and sophisticated algorithms to help researchers make sense of huge amounts of data within their areas of interest. For example, if you wanted to study the Cicada life cycle in rural Alabama, you could use Iris to map all of the relevant research or papers that would guide your research. This is possible because AI can translate gigantic data sets into patterns that make sense to human beings. It’s not that we couldn’t achieve the same results given infinite time and coffee — that would just be less efficient. And getting up to speed on what’s already been discovered slows down innovation. What Iris does could save a researcher weeks, if not months, of time. Eventually, Iris intends to develop an AI that could create its own hypothesis, test it, and even publish its findings. If AI is able to speed up complex tasks, then innovators will have more time to work on innovating. In our next post (and the last one in this series), we’ll narrow our focus to enterprise and investigate how marketing, sales, and HR teams are using artificial intelligence to support internal growth, as well as customer experience and acquisition.

The Rise of AI Tools: Verticals to Watch, Part 4

0

the-rise-of-ai-tools-verticals-to-watch-part-4-168c5ee11979

2018-06-05

2018-06-05 17:52:19

https://medium.com/s/story/the-rise-of-ai-tools-verticals-to-watch-part-4-168c5ee11979

false

789

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Conversica

Conversica is the leader in conversational AI for business and the provider of AI-powered assistants for marketing and sales teams.

3a015c32e82a

conversica

4

1

20,181,104

null

0

null

0

null

2018-05-01

2018-05-01 15:44:46

2018-05-01

2018-05-01 15:47:44

1

false

en

2018-05-01

2018-05-01 15:47:44

2

168e0a910dee

0.686792

1

0

Artificial intelligence researchers draw a clear distinction between Artificial Narrow Intelligence (ANI) or Weak AI, and Artificial…

5

An expert assesses the impact of artificial intelligence, today and tomorrow Credit: Becoming Human Artificial intelligence researchers draw a clear distinction between Artificial Narrow Intelligence (ANI) or Weak AI, and Artificial General Intelligence (AGI) or Strong AI. Weak AI is the stuff of today’s Siri, self-driving cars, those annoying systems that answer you when you phone just about any company nowadays, or other familiar devices that focus on a single task and possess no self-awareness. By contrast, as of today, Strong AI is essentially the stuff of science fiction. It’s sentient intelligence that equals or surpasses human intellectual abilities, setting goals and planning independently of prior programming. This sharp dividing line is at the core of the arguments advanced by Amir Husain in his fascinating new book, The Sentient Machine.

An expert assesses the impact of artificial intelligence, today and tomorrow

1

an-expert-assesses-the-impact-of-artificial-intelligence-today-and-tomorrow-168e0a910dee

2018-05-02

2018-05-02 16:45:49

https://medium.com/s/story/an-expert-assesses-the-impact-of-artificial-intelligence-today-and-tomorrow-168e0a910dee

false

129

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Mal Warwick

null

b76a4296434e

malwarwick_98471

387

299

20,181,104

null

0

null

0

null

2018-06-17

2018-06-17 18:03:54

2018-06-17

2018-06-17 18:33:53

2

false

en

2018-06-20

2018-06-20 23:18:09

0

168e2874d183

2.983333

1

0

Protect Society: Let us plan

5

Artificial Intelligence Outcomes Protect Society: Let us plan A thoughtful, proactive comment on Artificial Intelligence outcomes. Alisa Wright All of us can imagine a dystopian world which was caused by a nuclear event or a global crisis. But few spend a moment thinking about the one in five of there friends whom would be affected by workforce displacement, unable to get a new job at nearly the same salary or under-qualified for every job add. The reality is true, and this is real. We build systems solely focused on human reduction inside of businesses. These are skilled laborers who you, yourself could be or likely one of your friends. These skills equate to a human level IQ up to 115 where sequence/matching and correlation logic is required. After performing a focus group we learned that most people when questioned felt they were safe or the positions were impossible to be replaced. This is a ripple which would become a tsunami if not prepared for with great care and concern. More importantly, the cities in which it will affect are not the Detroit’s or Manhattan’s. Never present a problem without a solution. Our civic mission is to help the municipalities have a prepared plan in place in the event of any such outcomes. In my elementary school library, we had a laminated sign which read “Plan Ahead”. Growing up some of us did this and some did not. We see it rarely used in society and to point out a fact, more specifically in health care, where the majority of people do not take precautions even for their own health let alone other aspects of life. And we rarely see this cooperation play out between corporations and communities. However, it should not be borne solely by the clients which we serve, but by the change agents who are actively pursuing this change. Us. Basic research Deep dive: When is it already happening? Let’s understand a sign. A baseline study would project that some level of media hysteria typically begins with “general” unemployment trending above 8% when a critical mass is also identified. Example: if Pittsburgh took its 6% unemployment and it increased by 2 or 3%, the headlines would be that of hysteria level sentiment. So further clarifying this 2%, it would have to equate to one of five to seven people in a social group which were displaced. Key factors: Laid off worker, Job Category, Social network (size), Community density Summary: If 1 of 7 of your friends were laid off, then 7 of 49, 49 or 2,401 may make a ripple, but increase that to 1 of 5, and you can see “Mass Hysteria” or social outcry likely. Let’s look at this from an outcome perspective, What are the IQ or skill levels which we have designed, built and deployed, or who is affected and where are they living. What areas or municipalities: Let us use this example of commuters to San Francisco. ALASDAIR RAE Commuters working in San Francisco start from which cities or towns. Now apply this to our demographic which is Manufacturing. The roles inside of the larger firms are based in Santa Clara County, Fremont among a few. Willow Glen, Campbell, Palo Alto, Mountain View, Sunnyvale, San Mateo. i.e. a workforce commutes from the proposed problem area. Drawing a radius simply around the destinations for most commuters is a start. Using public records, LinkedIn and public traffic patterns we are able to pinpoint a more specific grouping. Summary: This only tells us who we need to speak with now. And does not answer the many questions we all have. We know the who and where but the when and how are next. Municipalities whom we are working with are open to preparatory understanding via our working group. However, this is only really getting people to a table not having the mechanisms and KPIs known which spell out when to react and what to do. Over the next weeks and months, we will expand on this journey and present publicly a template or framework which could benefit or friends and family who maybe side-effects of evolution /End

Artificial Intelligence Outcomes

42

clear-and-present-danger-planing-phase-168e2874d183

2018-06-20

2018-06-20 23:18:10

https://medium.com/s/story/clear-and-present-danger-planing-phase-168e2874d183

false

689

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Mark Swartz

My passion for continuous improvement leads me to find innovation opportunities like an explorer of an uncharted world. CEO/Founder - NEURAL CORP

8d0faac2928b

mark.swartz

1

20,181,104

null

0

null

0

defb4a59ced4

2018-03-29

2018-03-29 05:43:20

2018-08-11

2018-08-11 05:00:35

1

false

en

2018-08-17

2018-08-17 12:26:21

7

168f00291856

4.456604

2

0

1

Introduction

5

Question Answering in Natural Language Processing [Part-I] Introduction Question Answering is a computer science discipline within the fields of information retrieval and natural language processing, which focuses on building systems that automatically answer questions posed by humans in a natural language. A computer understanding of natural language consists of the capability of a program system to translate sentences into an internal representation so that this system generates valid answers to questions asked by an user [1]. Valid answers mean answers relevant to the questions posed by the user. As the internal representation of natural language, sentences must adequately map semantics of this statement, the most natural approach is in the simulation of facts contained in the sentences using a description of real objects as well as actions and events connected with these objects. To form an answer it is necessary, in the first place, to execute the syntax and semantic analysis of a question. This article covers the introduction to Question Answering, types and challenges posed by the systems in real world. Source Open Datasets available for Question Answering Stanford Question Answering Dataset (SQuAD)[2] is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable. There is an awesome article on this here. WikiQA dataset [3], is a publicly available set of question and answer pairs, collected and annotated for research on open-domain question answering. It is constructed using a more natural process and is more than an order of magnitude larger than the previous dataset. In addition, the WikiQA dataset also includes questions for which there are no correct sentences, enabling researchers to work on answer triggering, a critical component in any QA system. The TREC-QA dataset contains questions and answer patterns, as well as a pool of documents returned by participating teams. NewsQA dataset [4] is to help the research community build algorithms that are capable of answering questions requiring human-level comprehension and reasoning skills. Leveraging CNN articles from the DeepMind Q&A Dataset, authors prepared a crowd-sourced machine reading comprehension dataset of 120K Q&A pairs. Types of Question Answering There are three major modern paradigms of question answering: a) IR-based Factoid Question Answering goal is to answer a user’s question by finding short text segments on the Web or some other collection of documents. In the question-processing phase a number of pieces of information from the question are extracted. The answer type specifies the kind of entity the answer consists of (person, location, time, etc.). The query specifies the keywords that should be used for the IR system to use in searching for documents. b) Knowledge-based question answering is the idea of answering a natural language question by mapping it to a query over a structured database. The logical form of the question is thus either in the form of a query or can easily be converted into one. The database can be a full relational database, or simpler structured databases like sets of RDF triples. Systems for mapping from a text string to any logical form are called semantic parsers. Semantic parsers for question answering usually map either to some version of predicate calculus or a query language like SQL or SPARQL. c) Using multiple information sources: IBM’s Watson [5,6] system from IBM that won the Jeopardy! challenge in 2011 is an example of a system that relies on a wide variety of resources to answer questions. The first stage is question processing. The DeepQA system runs parsing, named entity tagging, and relation extraction on the question. Then, like the text-based systems, the DeepQA system extracts the focus, the answer type (also called the lexical answer type or LAT), and performs question classification and question sectioning. Next DeepQA extracts the question focus. Finally the question is classified by type as definition question, multiple-choice, puzzle or fill-in-the-blank. Next is the candidate answer generation stage according to the question type, where the processed question is combined with external documents and other knowledge sources to suggest many candidate answers. These candidate answers can either be extracted from text documents or from structured knowledge bases. Then it is passed through the candidate answer scoring stage, which uses many sources of evidence to score the candidates. One of the most important is the lexical answer type. In the final answer merging and scoring step, it first merges the candidate answers that are equivalent. The merging and ranking is actually run iteratively; first the candidates are ranked by the classifier, giving a rough first value for each candidate answer, then that value is used to decide which of the variants of a name to select as the merged answer, then the merged answers are re-ranked. Challenges in Question Answering The main challenges [7] posed by a Question Answering System are described below: Lexical Gap : In a natural language, the same meaning can be expressed in different ways. Because a question can usually only be answered if every referred concept is identified, bridging this gap significantly increases the proportion of questions that can be answered by a system. Ambiguity : It is the phenomenon of the same phrase having different meanings; this can be structural and syntactic (like “flying planes”) or lexical and semantic (like “bank”). The same string accidentally refers to different concepts (as in money bank vs. river bank) and polysemy, where the same string refers to different but related concepts (as in bank as a company vs. bank as a building). Multilingualism : Knowledge on the Web is expressed in various languages. While RDF resources can be described in multiple languages at once using language tags, there is not a single language that is always used in Web documents. Additionally, users have different native languages. A QA system is expected to recognize a language and get the results on the go! References [1] https://arxiv.org/pdf/1111.4343.pdf [2] https://rajpurkar.github.io/SQuAD-explorer/ [3] Yang, Y., Yih, W.T. and Meek, C., 2015. Wikiqa: A challenge dataset for open-domain question answering. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (pp. 2013–2018). [4] Trischler, A., Wang, T., Yuan, X., Harris, J., Sordoni, A., Bachman, P. and Suleman, K., 2016. Newsqa: A machine comprehension dataset. arXiv preprint arXiv:1611.09830. [5] Kalyanpur, A., Patwardhan, S., Boguraev, B.K., Lally, A. and Chu-Carroll, J., 2012. Fact-based question decomposition in DeepQA. IBM Journal of Research and Development, 56(3.4), pp.13–1. [6] https://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=6177717 [7] Höffner, K., Walter, S., Marx, E., Usbeck, R., Lehmann, J. and Ngonga Ngomo, A.C., 2017. Survey on challenges of question answering in the semantic web. Semantic Web, 8(6), pp.895–920.

Question Answering in Natural Language Processing [Part-I]

3

question-and-answering-in-natural-language-processing-part-i-168f00291856

2018-08-17

2018-08-17 12:26:21

https://medium.com/s/story/question-and-answering-in-natural-language-processing-part-i-168f00291856

false

1,128

All about Machine Learning, Natural Language Processing and Artificial Intelligence

null

lingvomasino

null

Lingvo Masino

lingvomasino@gmail.com

lingvo-masino

MACHINE LEARNING,NATURALLANGUAGEPROCESSING,NLP,AI,ARTIFICIAL INTELLIGENCE

null

Naturallanguageprocessing

naturallanguageprocessing

Naturallanguageprocessing

848

Ranjan Satapathy

PhD student in NTU Singapore. An NLP, Deep Learning, Sentiment Analysis and Sentic Computing researcher.

30181d7d65ac

kumarsatpathy

18

2

20,181,104

null

0

null

0

null

2018-06-05

2018-06-05 04:49:17

2018-06-05

2018-06-05 07:03:29

3

false

en

2018-06-05

2018-06-05 07:03:29

6

169191ecf24

4.134906

2

1

0

On May 16th, 2018, after much heated debate and a recount of the votes, the Hampshire Gazette reported that the town of Southampton, MA had…

5

Budget override voting behavior: what can we learn from data analysis? On May 16th, 2018, after much heated debate and a recount of the votes, the Hampshire Gazette reported that the town of Southampton, MA had passed a budget override of $493,000, largely to fund its elementary school, as well as other services. The vote was required because in Massachusetts, local spending growth is capped at 2.5% per year. Southampton had confronted this limit for several years, due to growth in the town’s population and changing requirements handed down by State government, and this was not the first time that a so-called “budget override” vote had been called — but why did it pass this time, whereas other recent ballot measures had failed? I live in Southampton, and have a son entering kindergarten this fall, so the success or failure of the override vote was of great interest to me. As I began to talk to my neighbors and acquiantances in town about the issue, some common themes began to develop. Supporters of the override identified their opponents as “townies”: predominantly elderly, longstanding residents of the town and/or members of families with a long history in the area. Those against the override complained that the school used to make do with less, wasn’t transparent about its finances, and had asked for money too often in the past — a perspective not familiar to newcomer households moving to the town in order to raise their young families. Some on both sides remarked that they wished the override question were not presented to voters as a single block of money for a range of departments, but rather line items for individual causes such as the elementary school, the town highway department, police, library, and so forth. Curious whether these narratives were supported by any kind of evidence (and also interested to see whether I could predict the voting outcome) I obtained data regarding voting history on overrides from the Massachusetts Division of Local Services web site (h/t to an article on Proposition 2 1/2 in Commonwealth Magazine written in 2005 by a journalist named Robert David Sullivan, which cited this resource). I then loaded this up in RStudio and merged it with additional tables from the UMass Donahue Institute and the US Census American Community Survey. I chose binomial regression as my statistical weapon of choice, since the outcome for any override vote is binary: each town either agrees to raise the money or not. The Sullivan article asserted that overrides for “general operating expenditures” had been more successful, whereas it seemed apparent that most pro-override voters in Southampton were (like me) mostly concerned with the future of the town’s elementary school, and would have been happier just to vote for that department. To investigate this dynamic I used fuzzy string matching (agrep in R) to identify keywords in the description of each recorded override in the DLS dataset as representing a funding request for schools, waste services, police, roads, libraries, and/or general operating expenses. Disappointingly, the resulting model had no significant coefficients, meaning that no category of services was meaningfully likelier to be funded than another. Contrary to the Sullivan article, libraries, not general operating expenses, appeared to fare best according to the data, and disappointingly for me, requests for schools appeared to dampen an override vote’s chances of success slightly. I was also worried that the fact that multiple override votes had recently been called and defeated in Southampton suggested a kind of “boy who cried wolf” situation where voters might simply be fatigued by being asked the same question again and again. This, too, however did not seem to be a significant factor in the success or failure of override votes in other towns across the state. A binary variable indicating whether an override had been called the prior year actually seemed to mildly increase the chances of an override vote winning, whereas the number of times an override had been called in the past five years decreased its chances. The anecdotal narratives I had heard were supported by the data in one respect: budget override votes were significantly likelier to pass in towns with higher percentages of newcomers (defined as anyone who was not living in the same house nor a house within the same county in the prior year’s American Community Survey). Towns with higher median incomes were also slightly likelier to pass budget override votes, although the significance of this effect was weak. Small towns and more rural areas with lower population density appeared slightly less likely to pass overrides, but the effect was not significant. However, the commonly held association between override opponents and older residents simply did not appear to have any basis in fact: towns with a higher median age were actually significantly likelier to pass override votes. I myself had met self-identifying “townies” who told me they opposed the override despite having their own children in the elementary school. They must have been in the minority of the voting population this time around, and so the override passed, possibly buoyed by an invisible coalition of newcomers like me and older residents. In other words, perhaps the outcome should not have been such a surprise, and the only reason it appeared to be was that so many less-than-useful narratives about support for and opposition to the override were swirling around the town in the months and weeks leading up to the vote. My data analysis did not lead me to a perfect Proposition 2.5 voting behavior forecasting tool, but it did demonstrate that models based upon anecdotally constructed theories often don’t perform as well as we might expect.

Budget override voting behavior: what can we learn from data analysis?

6

budget-override-voting-behavior-what-can-we-learn-from-data-analysis-169191ecf24

2018-06-05

2018-06-05 15:45:49

https://medium.com/s/story/budget-override-voting-behavior-what-can-we-learn-from-data-analysis-169191ecf24

false

950

null

Politics

politics

Politics

260,013

Colby Brown

null

811518237139

colbymbrown

15

45

20,181,104

null

0

null

0

null

2018-08-14

2018-08-14 20:10:34

2018-08-14

2018-08-14 20:10:49

0

false

en

2018-08-14

2018-08-14 20:10:49

1

1691d011a7cb

0.701887

0

Bengaluru and California-based voice AI platform Observe.ai has raised around ₹56 crores ($8 million) in series A funding led by Nexus…

1

Observe.AI raises Rs 56 Crores to use artificial intelligence Bengaluru and California-based voice AI platform Observe.ai has raised around ₹56 crores ($8 million) in series A funding led by Nexus Venture Partners, along with MGV, Liquid 2 Ventures, Hack VC, and existing investors including Emergent Ventures and Y Combinator. Observe.ai was founded by Akash Singh, Sharath Keshava and Swapnil Jain in 2017. It assists call centre agents with their job through its voice AI platform, that provides real-time feedback on customer sentiment and advises them on next best action during the call. The platform listens to the incoming calls in real-time and uses deep learning and natural language processing to understand the context and generate suggestions. This will really benefit the call centre agents as they won’t have to waste time looking for solutions elsewhere when it will be available right in front of them in real-time. It will also increase the agent’s efficiency, along with quicker resolutions for customers and reducing unnecessary costs. To continue reading this article, please go through the link here and enjoy reading all the tech news.

Observe.AI raises Rs 56 Crores to use artificial intelligence

0

observe-ai-raises-rs-56-crores-to-use-artificial-intelligence-1691d011a7cb

2018-08-14

2018-08-14 20:10:49

https://medium.com/s/story/observe-ai-raises-rs-56-crores-to-use-artificial-intelligence-1691d011a7cb

false

186

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Vernit

null

1dde1dc87f33

verbigg

1

20,181,104

null

0

null

0

null

2018-03-24

2018-03-24 08:39:55

2018-03-24

2018-03-24 09:19:34

5

false

en

2018-03-24

2018-03-24 09:19:34

1

16929cd936d8

3.542767

4

0

There has been a lot of scrutiny recently on the work done by UNHCR, OPM, and iNGO partners in the provision of services to refugee…

5

What We Know — Focus on Bidi Bidi Rolling out in Bidi Bidi Zone 5 There has been a lot of scrutiny recently on the work done by UNHCR, OPM, and iNGO partners in the provision of services to refugee communities in Uganda, and rightly so. Reflection and ongoing performance analysis is important if we are to achieve our shared humanitarian goals. In combing through the hundreds of thousands of customer responses that Kuja Kuja has collected in the past few months within refugee communities in Uganda, we came to a realization: we know a lot about the communities currently being spoken about in Uganda. A lot. In the past 4 months we have spoken to over 100,000 refugees across settlements in Uganda alone. We have the largest data set of its kind in the world and the voices within it are powerful. For the next month, we will be releasing short weekly articles focusing on a particular issue in one sector in one location in Uganda and providing a high level overview of community voices in that location. We will expand to our other locations in Rwanda, Sudan, and Somalia in due course. This week we will focus on Protection Services in Bidi Bidi and the difference between the voices of Community Leaders and the voices of the Community themselves. We present no opinion on this matter — our goal here is to shine a light on something we find interesting. For deeper insight, check out www.kujakuja.com and sign up for access to our full data set. Focus on Bidi Bidi In Bidi Bidi Zone 5 last week we spoke to 2857 people about Protection Services. In total, we have spoken with 37,558 people in Bidi Bidi since November. Kuja Kuja collects feedback on Protection Services at trainings for community leaders, at community awareness-raising sessions, and at the doors of the tens of thousands of South Sudanese refugees who call Bidi Bidi home. In Protection Services — the provision of services like awareness and training around sexual violence, counseling for victims of community violence, and community policing — customer satisfaction is at about 40% week on week. This is an increase from 20% satisfaction overall back in November. At any given moment in Bidi Bidi, 60% of people are unsatisfied with the Protection Services being provided to them. But who you speak with matters. The Voice of Community Leaders: At training sessions for community leaders the ratings are high. 77% of community leaders are happy with the Protection Service being provided. Their ideas for improving those services focus on a core set of themes: Provide more trainings Increase allowances for community leaders Provide transport to meetings The Voice of the Community: In the community the ratings are very different. At Ariwa 1, for example, a village of a few thousand people about an hours drive from Yumbe town, the satisfaction with Protection Services currently sits at about 25% having started at about 12% back in November. The ideas that we receive from the community there focus on another set of themes, very different from those articulated by community leaders: Provide access to solar lights Construct shelters Provide better access to water So what does this all mean? At Kuja Kuja is is not our place to judge. We just listen to our customers, collect information, and make that information publicly available. Neither the Community Leaders nor the Community themselves are right or wrong here. Both are legitimate actors coming at an issue from different angles. Both need to be listened to. Both positions need to be understood and addressed in the design of services. As the improving satisfaction ratings in Bidi Bidi can attest to, our partner at the American Refugee Committee has been listening and has been improving services, but they need more support from the actors around them. At Kuja Kuja our encouragement is simple. Talk to a range customers. Listen to dissenting opinions. Ask why more frequently. Being an effective service provider means providing the services people want, not the services that you think that they need. Next week we will be looking at Nakivale in South West Uganda, focusing on what the communities across the sub zones of Base Camp, Juru, and Rubondo have to say about payments for water. Thanks for reading Your Kuja Kuja Insight Team xx Friday in Bidi Bidi

What We Know — Focus on Bidi Bidi

19

what-we-know-focus-on-bidi-bidi-16929cd936d8

2018-05-31

2018-05-31 05:39:22

https://medium.com/s/story/what-we-know-focus-on-bidi-bidi-16929cd936d8

false

718

null

Refugees

refugees

Refugees

16,270

Kuja Kuja

Kuja Kuja is a real time feedback system to allow organizations design more impactful products and services for their customers

c7cee4b8cbde

KujaKujaGlobal

7

3

20,181,104

null

0

null

0

13ce71c3b01d

2018-03-21

2018-03-21 20:26:34

2018-03-21

2018-03-21 20:33:09

1

false

en

2018-03-21

2018-03-21 20:41:10

0

16944164ee5e

0.807547

1

0

In our latest podcast, Dipika Sen (Wharton MBA ’19) chats with Jungwon Byun, Head of the Growth Team at Upstart, the first lending platform…

5

Podcast with Jungwon Byun, Head of the Growth Team at Upstart Jungwon Byun, Head of Growth at Upstart In our latest podcast, Dipika Sen (Wharton MBA ’19) chats with Jungwon Byun, Head of the Growth Team at Upstart, the first lending platform to leverage artificial intelligence and machine learning to price credit and automate the borrowing process. In this engaging podcast, Jungwon talks about her own trajectory from consulting into FinTech, discusses her initiative to foster the professional development of women at Upstart, and shares her insights on the importance of diversity in the industry. At Upstart, Jungwon oversees all of the firm’s user acquisition strategies across digital, offline, and lifecycle channels. She was previously a management consultant at Oliver Wyman, where she assessed the risk exposure of international banks’ capital markets businesses for regulatory stress testing. She brings experience from her time as the CEO and co-Executive Director of a domestic microfinance non-profit. Jungwon graduated cum laude from Yale University, with a B.A. in Economics.

Podcast with Jungwon Byun, Head of the Growth Team at Upstart

1

podcast-with-jungwon-byun-head-of-the-growth-team-at-upstart-16944164ee5e

2018-04-01

2018-04-01 18:22:43

https://medium.com/s/story/podcast-with-jungwon-byun-head-of-the-growth-team-at-upstart-16944164ee5e

false

161

We are FinTech thought leaders connecting innovators, academics, and investors with the ideas and companies that are reinventing global financial services.

null

whartonfintech

null

Wharton FinTech

whartonfintech@gmail.com

wharton-fintech

FINANCIAL SERVICES,WHARTON,TECHNOLOGY,FINTECH,DISRUPTION

whartonfintech

Fintech

fintech

Fintech

38,568

Dipika Sen

null

1573f7ba4b

dipika.sen322

2

1

20,181,104

null

0

# if needed install.packages(c("tidyverse", "broom")) library(tidyverse) library(broom) t = 1:100 y1 = 22 + (53 - 22) * exp(-0.02 * t) %>% jitter(10) y2 = 24 + (60 - 22) * exp(-0.01 * t) %>% jitter(10) df <- data.frame(t = t, y = y1, sensor = 'sensor1') %>% rbind(. , data.frame(t = t, y = y2, sensor = 'sensor2')) | t | y | sensor | |-----|----------|---------| | 1 | 52.25302 | sensor1 | | 2 | 51.71440 | sensor1 | | 3 | 51.13971 | sensor1 | | … | … | … | | 98 | 38.13328 | sensor2 | | 99 | 38.17017 | sensor2 | | 100 | 37.72184 | sensor2 | qplot(t, y, data = df, colour = sensor) augmented <- df %>% group_by(sensor) %>% do(fit = nls(y ~ a * t + b, data = .)) %>% augment(fit) qplot(t, y, data = augmented, geom = 'point', colour = sensor) + geom_line(aes(y=.fitted)) qplot(t, .resid, data = augmented, colour = sensor) qplot(.fitted, .resid, data = augmented, colour = sensor) df %>% group_by(sensor) %>% do(fit = nls(y ~ a * t + b, data = .)) %>% tidy(fit) %>% select(sensor, term, estimate) %>% spread(term, estimate) | | sensor | a | b | |---|---------|------------|----------| | 1 | sensor1 | -0.2495347 | 47.87376 | | 2 | sensor2 | -0.2350898 | 59.77793 | y ~ a * t + b # Select the data from our first sensor: sensor1 <- df %>% filter(sensor == "sensor1") # Fit our model: fit <- nls(y ~ a * t + b, data = sensor1) # nls works best if we specify an initial guess for the # coefficients, the 'start' point for the optimisation: fit <- nls(y ~ a * t + b, data = sensor1, start = list(a = 1, b = 10)) > summary(fit) Formula: y ~ a * t + b Parameters: Estimate Std. Error t value Pr(>|t|) a -0.249535 0.006377 -39.13 <2e-16 *** b 47.873759 0.370947 129.06 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 1.841 on 98 degrees of freedom Number of iterations to convergence: 1 Achieved convergence tolerance: 1.571e-09 > coef(fit) a b -0.2495347 47.8737585 > predict(fit) [1] 47.62422 47.37469 47.12515 46.87562 46.62609 46.37655 ... > predict(fit, newdata = data.frame(t = 101:200)) [1] 22.67076 22.42122 22.17169 21.92215 21.67262 21.42309 ... > resid(fit) [1] 4.628798 4.339712 4.014558 3.576258 3.528138 3.001447 ... install.packages('broom') library(broom) > tidy(fit) | term | estimate | std.error | statistic | p.value | |------|------------|-------------|-----------|---------------| | a | -0.2495347 | 0.006377179 | -39.12932 | 1.267667e-61 | | b | 47.8737585 | 0.370946838 | 129.05827 | 3.123848e-111 | > augment(fit) | t | y | .fitted | .resid | |---|----------|----------|------------| | 1 | 52.25302 | 47.62422 | 4.62879752 | | 2 | 51.71440 | 47.37469 | 4.33971163 | | 3 | 51.13971 | 47.12515 | 4.01455804 | | 4 | 50.45188 | 46.87562 | 3.57625801 | | … | … | … | … | qplot(t, y, data = augment(fit)) + geom_line(aes(y = .fitted)) qplot(t, .resid, data = augment(fit)) > glance(fit) | sigma | isConv | finTol | logLik | AIC | BIC ... |----------|--------|--------------|----------|----------|---------- | 1.840841 | TRUE | 1.571375e-09 | -201.906 | 409.8119 | 417.6274 ... fitted <- df %>% group_by(sensor) %>% do(fit = nls(y ~ a * t + b, data = ., start = list(a = 1, b = 1))) | | sensor | fit | |---|---------|-----------| | 1 | sensor1 | <S3: nls> | | 2 | sensor2 | <S3: nls> | fitted %>% tidy(fit) | sensor | term | estimate | std.error | statistic | p.value | |---------|------|----------|-----------|-----------|---------------| | sensor1 | a | -0.24953 | 0.0063771 | -39.12932 | 1.267667e-61 | | sensor1 | b | 47.87375 | 0.3709468 | 129.05827 | 3.123849e-111 | | sensor2 | a | -0.23508 | 0.0030882 | -76.12478 | 5.400611e-89 | | sensor2 | b | 59.77792 | 0.1796349 | 332.77446 | 1.931798e-151 | fitted %>% augment(fit) | sensor | t | y | .fitted | .resid | |---------|-----|----------|----------|-------------| | sensor1 | 1 | 52.41684 | 47.58220 | 4.834645067 | | sensor1 | 2 | 51.64323 | 47.33347 | 4.309757950 | | sensor1 | 3 | 51.09595 | 47.08475 | 4.011196179 | | … | … | … | … | … | | sensor2 | 98 | 38.06029 | 36.72428 | 1.336007642 | | sensor2 | 99 | 37.86891 | 36.48926 | 1.379642201 | | sensor2 | 100 | 37.89936 | 36.25425 | 1.645117019 |

33

null

2018-09-20

2018-09-20 20:06:44

2018-09-20

2018-09-20 20:34:49

6

false

en

2018-09-20

2018-09-20 20:37:31

4

16945f559017

7.033019

0

Sep 9, 2018 · 7 minute read · Comments

5

Curve fitting on batches in the tidyverse: R, dplyr, and broom · Douglas Watson Sep 9, 2018 · 7 minute read · Comments I recently needed to fit curves on several sets of similar data, measured from different sensors. I found how to achieve this with dplyr, without needing to define outside functions or use for-loops. This approach integrates perfectly with my usual dplyr and ggplot2 workflows, which means it adapts to new data or new experimental conditions with no changes. Here are the ready-made recipes for any one else who may run into a similar problem. First we’ll generate some artificial data so that you can follow along at home. Here’s an exponentially decaying value measured by two sensors at different locations: Our data frame looks like this: And our data looks like this: I created the data in long format, as it works best with dplyr pipelines. You can read more about long format vs wide format data here. TL;DR I want to plot the fitted curve over my data points I want to plot the residuals I want a table of my fit parameters for each condition: Looks like: I want a less horrendous fit. How can I do better? Try with an actual exponential. You’ll probably want a self-starting function to avoid the “singular gradient” error — read more in my post on the subject. Explanation: intro to curve fitting in R The goal of our fitting example was to find an estimate y*(t) = a t + b that approximates our measured data y. In R, we can directly write that we want to approximate y using the very intuitive built-in formula syntax: In R’s documentation, a and b are the coefficients, t is the independent variable and y is the dependent variable — t can go about its day without every knowing y exists, but y can’t progress without knowing the latest trend in t. We fit these models using various fitting functions: lm or glm for linear models or nls for non-linear least squares for example. I’m using nls here because I can specify the names of my coefficients, which I find clearer to explain: The fitting functions all return a fit object: You can use one of these base R functions to extract useful data from the fit object: coef(fit) returns the values of the coefficients a and b. predict(fit) returns y*(t_i), i.e. it applies the fitted model to each of the original data points t_i. It can also be applied on new values of t. resid(fit) returns the residuals y_i - y*(t_i) at each point of the original data. Try them out in the console: Now these functions all return vectors, which work best with R’s native plotting functions. To integrate with dplyr and ggplot, we’d rather have data frames. This is where the broom package comes in. Introducing broom Broom is a separate R package that feeds on fit results and produces useful data frames. Install it directly within the R console if you haven’t already: It provides three useful functions, which are easier to demonstrate than to explain. All of them return data frames: tidy extracts the fit coefficients and related information. The term column contains the name of your coefficient, and estimate contains its fitted value: augment is equivalent to predict and resid combined. The returned data frame contains columns t and y with your original data, as well as .fitted, the fitted curve, and .resid, the residuals: You can plot the ‘augmented’ fit with qplot: glance shows fit statistics: Using broom with dplyr Since broom’s functions return data frames, they integrate naturally with magrittr pipelines (%>%) and dplyr. Let’s illustrate by grouping the data by sensor id, then producing a fit object for each group. The do function calls nls once for each group, then stores the return value in a column we named fit: The dataframe fitted looks like: We now have one nls object per group, which can be fed into either tidy or augment. These functions act on each group separately: yields: From here, we can use select and spread to generate a more succinct table, as in the TL;DR example, or keep the table as-is to inspect the quality of the fit. The result of augment is a long data frame that can be used immediately with qplot, also shown in the TL;DR section above: which yields the data frame: That’s it for broom and dplyr. The magic of this approach is that you can add a sensor to your input data then re-run all your code; your new curves will just appear on the plots. Similarly, you can change the formula of your fit, and your new coefficients will appear as a new column in the coefficients table. No need to change your code. Originally published at douglas-watson.github.io.

Curve fitting on batches in the tidyverse: R, dplyr, and broom · Douglas Watson

0

curve-fitting-on-batches-in-the-tidyverse-r-dplyr-and-broom-douglas-watson-16945f559017

2018-09-20

2018-09-20 20:37:31

https://medium.com/s/story/curve-fitting-on-batches-in-the-tidyverse-r-dplyr-and-broom-douglas-watson-16945f559017

false

1,612

null

Data Science

data-science

Data Science

33,617

Douglas Watson

null

1b4ededa5825

douglas_watson

13

1

20,181,104

null

0

null

0

661161fab0d0

2018-09-08

2018-09-08 14:48:57

2018-09-08

2018-09-08 16:08:12

2

false

en

2018-09-09

2018-09-09 07:24:50

8

1694f39de852

3.975786

10

3

1

“Yes, it will!” (or, “No, it won’t!”)

4

Will AI take my job? “Yes, it will!” (or, “No, it won’t!”) “Code on a computer” by Markus Spiske on Unsplash Imagine a city centre of the future — there are no cars, no taxis, no buses and no trucks. They have all been replaced by autonomous pods — small, electric, self-driving vehicles that can be adapted to transport passengers or goods, that have a map of every road, every junction and every obstacle in the city, and that are entirely automatic. There is no need for traffic lights, roundabouts, or road markings because the pods not only have complete knowledge of the city but also, because they communicate with each other, they have precise knowledge of the whereabouts of every other pod. They can negotiate with each other when it comes to who goes first at a junction, or where a pod merges into traffic when pulling away. There would be no collisions because every pod would cooperate with their fellow pods. Passenger pods would ferry commuters from railway stations to their offices, take the kids to school and their parents to the shopping mall. Pods carrying goods would bring your weekly shop from an out-of-town supermarket warehouse and deliver the mail. So what, in this automated city, has happened to all of the people that used to drive taxis, buses and delivery vans? They are no longer are needed. What about the hundreds of young women and men on bicycles and scooters delivering pizzas, burgers and so on? Also redundant. And because you now order all your groceries online, there are no supermarkets — no checkout operators, no shelf-stackers — and the warehouses that have replaced them are fully automated. There are no private cars in the city centre, so there are no parking restrictions to enforce, no traffic violations to prosecute, no traffic wardens. In the offices where pods drop off the commuters, what used to be an out-of-town call centre, is now an artificially intelligent chatbot in the Cloud accessible from a PC on someone’s desk — or their mobile phone, or tablet. The legal department is an application, also in the Cloud, that deals with all day-to-day matters and only passes you on to a real human lawyer when something unusual crops up. You rarely need to go to the doctor in this futuristic city because a smartphone app is perfectly capable of diagnosing your illnesses and prescribing the appropriate treatments. And even where specialist scanners, or other medical procedures, are required, it is an AI that analyses the images and tells you whether you have cancer, or a potentially dangerous eye disease. So what is going to happen to our jobs in the future? Are we all to be made redundant? And, if so, does this mean a life of leisure, or one of rampant inequality, where the owners of the technology get richer and the rest of us live in poverty? A report by the McKinsey Global Institute suggests that half the activities that people are now paid to do could be automated using current technologies. On the other hand, it also says that less than 5% of all occupations could be fully automated (although many more contain activities that could be automated). The conclusion is that many jobs will change, with humans working alongside machines but few of them will be completely eliminated. That still means, of course, that fewer employees will be needed in those areas where technology replaces a significant amount human effort. “close-up photo of factory” by Taton Moïse on Unsplash We’ve been here before, of course. Since the Industrial Revolution of the 18th and 19th centuries (if not before) technology has been replacing people in the workplace. But it has also created new jobs. When great factory-based steam-powered looms replaced weavers working in cottages, the new machinery had to be designed and manufactured, factories had to be built, managed and maintained. In the Information Revolution of the 20th and 21st centuries, you might not have been overjoyed if you had been a travel agent of the High Street seeing your industry being almost totally eclipsed by online services. Or, if you worked as a skilled machine operator watching the factory where you worked being, more and more, populated with industrial robots. Conversely, had you been willing or able to retrain as a computer programmer, you might have experienced a brighter future. The introduction of technology tends to increase productivity, and, as productivity rises, products become cheaper and society, in general, becomes wealthier, so citizens are able to afford new goods and services that were previously either expensive or didn’t exist. This helps stimulate the creation of new companies providing new products and services and thus creates more employment. Whatever you think of it, the gig economy has been borne out of new technologies, it may be precarious employment but it certainly is employment. History suggests, then, that increases in productivity lead to a wealthier society but there is no law that says that this is inevitable. It might be different this time. If the use of AI-powered technologies really do lead to mass unemployment, or unacceptable inequality, there is a political choice to be made: do we allow such a society to develop, or do we provide a safety net for those adversely affected, such as the introduction of a Universal Basic Income. The question is, are we heading for a utopia where no one needs to work or a dystopia where no one but the lucky few has any work. Or will we carry on as usual with new types of employment being created by the new artificially intelligent world? We need to be prepared. Originally published in The Startup AlanJones|JustEnoughPython|My programming blog|Buy me a Coffee

Will AI take my job?

122

will-ai-take-my-job-1694f39de852

2018-10-03

2018-10-03 22:13:04

https://medium.com/s/story/will-ai-take-my-job-1694f39de852

false

952

where the future is written

null

Predict

predictstories@gmail.com

predict

FUTURE,SINGULARITY,ARTIFICIAL INTELLIGENCE,ROBOTICS,CRYPTOCURRENCY

null

Technology

technology

Technology

166,125

Alan Jones

An ex-university professor and software engineer, I mostly write about AI, programming and technology in general - occasionally other stuff, too.

7d3f5fb94faa

jones.alan

59

7

20,181,104

null

0

null

0

null

2018-06-12

2018-06-12 05:52:53

2018-06-12

2018-06-12 06:10:00

2

false

en

2018-06-12

2018-06-12 06:10:00

5

1695698d8f5e

3.092767

0

It seemed like yesterday we were all saying ‘mobile-first.’ But in today’s world, anything less than ‘AI-first’ means you are likely to get…

1

How To Get To AI-First “A robot named Pepper holding an iPad” by Alex Knight on Unsplash It seemed like yesterday we were all saying ‘mobile-first.’ But in today’s world, anything less than ‘AI-first’ means you are likely to get left behind. What does that mean for banks and other companies who have invested hugely in their mobile presence? We have been so used to providing standard banking services, and focusing on the best platforms to provide those services, whether online or mobile. Does AI provide us with an opportunity to rethink that approach radically? Or, will it be more like a bolt-on to existing concepts and capabilities? Clearly, there are choices that we need to make, and the time to decide on how to leverage the transformative capabilities that AI offers, is now. As companies like Google, Amazon, Facebook, and Apple continue to set the expectations of consumers. Banks and other companies using strong legacy systems are struggling to adapt; not only to the approach of reducing friction that takes care of itself, but also the move to platforms that offer AI in increasing values and scales. To stay relevant we must move quickly to build value for customers using insights which help them get ahead in their day to day. Most banks have invested huge amounts of resources in building their physical locations, their websites, and apps while being focused on an approach that brings customers to them. But the world of mobile technology, social media networks, and opportunities created by AI enables us to go where our customers are, and do more than offering them banking products. Moreover, we have the opportunity to offer them intelligent and actionable insights that helps them make decisions quicker, and build much greater relevance to their lives. An example: Dave is driving home, and Siri says: “Hi Dave, remember that your daughter’s birthday is coming up a month. From her social media it looks like she’s a Taylor Swift fan. Tickets for Taylor Swift’s next concert will be going on sale next week, and since you hold an ING Visa card, you’ll have access to exclusive pre-release tickets in that sale. Would you like us to purchase those for you when they’re available?” But how do we get there? The first step is to adopt the open-platform philosophy. That means we have to be able to be where our customers are. At the moment, they are using social media on mobiles; tomorrow they’ll be interacting using voice assistants in their homes and on their mobile devices. VoiceLabs 2017 report shows that the number of people using voice-first devices is at a staggering 87.5 percent of the entire population; the percentage is expected to grow further in the coming years. Beyond voice devices, consumers will be interacting with the next wave of wearable devices that interact with our brainwaves, our bodies, our voices, and our hand commands to create augmented reality experiences that integrate AI technologies. It will be these new platforms that carry the idea of ‘AI forward’ to consumers. ING is working to launch services for customers on Google Home, Apple’s Siri, and Amazon Echo. The bottom line is, you can and should be on every digital platform where your customers are, and you shouldn’t be afraid to leave some platforms behind as you seek new opportunities. This means, businesses that focus on building architecture, should show its presence on any new platform and getting the underlying architecture right so that you can improve and scale across the board as these platforms change. The second step is to embrace the advancements of those leading in AI technologies. At the moment we’re using those advancements to help us interact with our customers using natural language processing. So no matter if they ask their Google Home “Am I broke?”, or if they type “How much money I’ve got left?” into their Facebook Messenger, we’ll be able to understand and respond with what they need. These new AI capabilities are strong and are getting stronger, and you don’t have to force your customers to speak your language anymore; you can talk to them in theirs. This is critical as it’s another step towards empowering your customers to get ahead in a way that they control. Read more…. Also Read: Hope or Hype? Benefits Of Artificial Intelligence

How To Get To AI-First

0

how-to-get-to-ai-first-1695698d8f5e

2018-06-12

2018-06-12 06:10:01

https://medium.com/s/story/how-to-get-to-ai-first-1695698d8f5e

false

718

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Sahana Karnam

null

4d14a046af36

sahana_63956

43

174

20,181,104

null

0

from scipy.spatial import distance euclidean_distnce = distance.euclidean(User1, User2)

1

null

2018-06-25

2018-06-25 13:09:37

2018-06-25

2018-06-25 15:13:18

4

false

en

2018-06-25

2018-06-25 15:13:18

3

16961c962dd

3.345283

2

0

1

It is one of the method for performing collaborative filtering. If collaborative filtering is new to you don’t forget to read this article…

5

Nearest neighbour based method for collaborative filtering It is one of the method for performing collaborative filtering. If collaborative filtering is new to you don’t forget to read this article to take a brief look. The objective here is to predict the user’s rating for the products they have not rated. The user for whom we are trying to predict the ratings is called active user. This method finds the k-nearest neighbours of active user and recommends him products liked by his neighbours. This method works because if the users have liked same types of products then they are most probably similar and they might like same type of products. The steps to recommend products using nearest neighbours are: Find k-nearest neighbours of active user. Find their weighted sum of ratings for products. These are now the ratings of active user given for the products s(he) has not rated yet. So in short, we find similar users then we find the weighted average of ratings they have given for products and this becomes the ratings of active user. But how do we find the neighbours of active user? How do we measure similarity between users? We know that the similar users are those who give same ratings for different products. For e.g if user A and B give ratings 5 for product P then A and B are similar. To measure a similarity we can build a user product/items ratings matrix: So we can see from the above users item matrix, there are D items/products and we have N users. Each row is the ratings given by user to the items. Now our goal here is to find how much would User1 rate ItemD. The first step is to find users that are similar to User1. If we look into the matrix we can see, UserN is similar to User1 because they both rate 4, 5 and 3 to item1, item2 and item4 respectively. The users products rating matrix is very sparse. This is much similar to real world scenario. Our goal is to fill those empty spaces. Now we have seen User1 and UserN are very similar to each other so it is highly probable that User1 would also rate 4 to ItemD because userN has also rated 4 to itemD. In reality, the process of finding similarity between active user and other users are much more complex. We need a similarity or distance metric to measure similarity between them. There are different metrics to find similarity between users: 1.Euclidean distance This metric is very simple and it is simply the distance between two points in the space. For example if we have N dimensional space for users User1(x1, x2, x3, …xn) and User2(y1, y2,y3, …yn) the distance between them is: Or mathematically, 2. Cosine similarity This metric calculates the cosine of angle between two vectors. If the distance between two vectors decreases then the angle theta also decreases between them. If User1 and User2 are in If the two vectors are perfectly aligned then the similarity would be zero and if they are orthogonal it would be one. The idea of finding cosine similarity is to find the dot products of vectors first and divide it by the product of magnitude of individual vector. 3. Pearson correlation Correlation is just a similarity between two vectors. It is similar to cosine similarity but we take into account for the bias for the user. For examples some people normally rate the products higher and some users rate the products lower. So how can account these biases? One way would be we normalize the users ratings by the average rating. This is exactly what Pearson correlation does. Here x bar and y bar are the mean of all the rating user x and y have given respectively. The users are first shifted by x bar and y bar ( to remove the bias) and angle or similarity between them is calculated. I will write more about recommendation system algorithms in my coming articles. If you like this post, don’t forget to clap the post and follow me on medium and on twitter.

Nearest neighbour based method for collaborative filtering

2

nearest-neighbour-based-method-for-collaborative-filtering-16961c962dd

2018-06-25

2018-06-25 15:13:18

https://medium.com/s/story/nearest-neighbour-based-method-for-collaborative-filtering-16961c962dd

false

701

null

Data Science

data-science

Data Science

33,617

Rabin Poudyal

Data Scientist. Say "hi" via email: rabinpoudyal1995@gmail.com or Follow me on twitter @poudyal_rabin

f103d513362e

rabinpoudyal1995

69

16

20,181,104

null

0

null

0

8e9bde78121d

2018-09-21

2018-09-21 09:50:45

2018-09-24

2018-09-24 03:54:09

3

false

en

2018-09-24

2018-09-24 03:54:09

3

1697538c807

4.648113

2

0

Written by Alexandra Lamb

5

Can Artificial Intelligence Rule the World? Written by Alexandra Lamb Did you ever think you would live to see the day that you would be conversing with chatbots about your latest credit card transactions, purchasing your next grocery list from Amazon, or served your favorite coffee by an automated barista that remembers your preferences? Think again, because it is already happening and will continue to happen in a big way. You can now browse for unusual vegetables and organic chips from Wholefoods and access its recipe database to whip up a delicious meal, using artificial intelligence to converse with a robot chef. You can also have daily conversations with Alexa, Amazon’s customizable virtual assistant, about the latest weather forecast, your daily appointments, to playing trivia games and even ordering pizza. You may even be the proud owner of Roomba, the latest robot dust cleaner, which can already map the square footage and dimension of your house for more effective cleaning. The fact that artificial intelligence (AI) and machine learning are becoming more and more a part of our daily lives just proves that technology has been advancing by leaps and bounds and, unless we want our lives and businesses to be disrupted, we best jump on the bandwagon of technology in order to remain relevant. So, what are your strategies to ensure that you are prepared for disruptive impact and maintain your competitive advantage as a business? We, at John Clements Consultants, have been responding to the call of technology by bringing in some of the brightest thought leaders and experts in the field of data analytics and artificial intelligence in the past year. Most recently, we held the two-day workshop, Applied Analytics for Competitive Advantage, at the Discovery Primea in Makati City. It was well-attended by around 40 top leaders from various industries, who were curious to find out how they could stay ahead of the technology game. We invited Professors Ikhlaq Sidhu (UC Berkeley’s Sutardja’s Center for Entrepreneurship and Technology) and Paris de l’Etraz (IE Business School, Spain) to share their research and perspectives on the future of data science, artificial intelligence, and machine learning and how these are currently impacting and will continue to impact our different businesses in the years to come. So, how does all this technology and availability of data impact our businesses? According to Prof. Sidhu, the key to staying ahead in the machine learning era is the proper utilization and analysis of data; whoever applies the better data science and maximizes the leveraging of data will win. Providing insight into the data we collect is critical to understanding our customers’ preferences and, in return, allows us to provide better value-added products and services. Let’s face it, everything today is data-driven and these data are, subsequently, stored in the cloud. Information is at the very core of every business. We have information on our customers, our products, and our services — and it’s steadily increasing. We can use this data and connect it to our business model. When you have cross validation of data input and outcomes, you get better results. The more data you collect, the more you can train AI, which increases the accuracy of the output. This allows us to respond appropriately and deliver better customer experience. Companies are creating data streams and are monetizing this data — e.g., Google, Facebook and Amazon; however, at the end of the day, as Prof. Sidhu clearly pointed out, we (humans) still have the advantage of understanding the philosophy of data science and tie it into our business strategies. There is no intelligence, desire, or existence in AI without humans. We have to know what threats we face, look at our opportunities and challenges, connect it with the business strategy and ensure that we do not veer away from our core competencies. We must know what our competitors are doing and recognize what technology can do for us. It is no longer a question of “if”, but rather of “when”, digital disruption catches up with our businesses. When the time comes, is your business ready for disruption? So many businesses are getting disrupted by technology — Uber, AirBnb, Amazon, Netflix, and Fintechs are all disrupting our traditional businesses. Netflix, for example, predicts the content you like by capturing your viewing patterns. It understands what you like because it has segmented you into hundreds of categories based on the shows you watch. How can you capitalize on technology to stay relevant with your customers? As Prof. de l’Etraz pointed out, AI is all about prediction. The better you predict what your customers want, the better you can deliver a positive customer experience. How far can predictions go in your industry? The longer you wait, the more likely that AI gains will disappear. Focus on product and service innovation in AI to capture top-line benefits first, then follow it up by establishing your digital transformation strategy. Digital transformation cannot be in the form of small efforts; it has to be an overall, all-encompassing strategy. You need to find ways to engage in partnerships for your company to scale — partners that have data about your customers. Also, look into startups that already have a platform to host the data you can leverage on. This is what Amazon and Google are doing. Here are some practical initiatives, recommended by Prof. de l’Etraz, to embark upon in order to jump-start the digital transformation process within your organizations: Create an internal data transformation team (DTT). Work with an external organization & strategy execution team for global sourcing of innovation trends, startups, and innovation strategy. Work with an external technology execution team for data and AI strategies. Use the Train the Trainer approach to develop your in-house agile capabilities. Create an internal budget to strategically invest in startups and initiatives. Keep the DTT lean — work in matrix with silos. The DTT reports to the CEO or business line head. At the end of the day, artificial intelligence will be the new innovation. It’s all about knowing our customers better and using the data we collect in an intelligent way in order to improve the customer experience. And finally, to quote Steve Jobs, one of the greatest technology leaders of our time, “Start with the customer experience, and work backwards using technology.” Please visit and join the John Clements Talent Community. About the author: Alexandra is the Business Development Director of the John Clements Corporate Learning Division.

Can Artificial Intelligence Rule the World?

4

can-artificial-intelligence-rule-the-world-1697538c807

2018-09-26

2018-09-26 00:54:56

https://medium.com/s/story/can-artificial-intelligence-rule-the-world-1697538c807

false

1,086

Discover Your Full Potential with Looking Glass, a Publication from John Clements

null

johnclementsph

null

John Clements Lookingglass

jcdigitalrenewal@gmail.com

the-looking-glass

LEADERSHIP,CAREERS,MANAGEMENT AND LEADERSHIP,PROFESSIONAL DEVELOPMENT,PERSONAL GROWTH

JohnClementsPH

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Shiela Manalo

Writer|Graphic Artist|Video Editor|Musician

8dbb2651e54f

iamsimone02

64

25

20,181,104

null

0

null

0

null

2018-09-24

2018-09-24 05:27:31

2018-09-29

2018-09-29 16:36:19

0

false

en

2018-09-29

2018-09-29 16:36:19

6

1697bee90c73

3.230189

0

Ultimate goal of AI is to get the cognitive capabilities for Machines and this paper creates a Model for Learning and from that model…

1

A Theory of the Learnable Ultimate goal of AI is to get the cognitive capabilities for Machines and this paper creates a Model for Learning and from that model creates a theory of the Learnable and also the limitations of same also. Humans are genetically preprogrammed and acquire certain skills by explicit instructions. Many of our skills are learned without any explicit programming. For example, you were probably never taught exactly how to recognize an object like a lightbulb.It is very difficult to create an algorithm which will do this task( learning with experience).So David Waltz suggesting that if we can make a computer to do this task we can make an algorithm for the task. In this paper deals with Concept recognition problem( Instead of identifying Light Bulb 😃). Computer is presented with an Data and it has to identify if this data belongs to specific Concept. “Banana is Fruit and yellow in color” Let the machine decide if the given data is relevant to this concept. How can we do it?( How Computer can understand an Concept) Learning Protocol- which specifies what kind of information can be supplied to the computer.(For the above concept Banana is Fruit and yellow in color we need to feed machine with suitable data set which are relevant so same) .The learning protocol consists of two subroutines EXAMPLES(), which produces examples of a concept and ORACLE(x), which checks if x is an example of the concept. Deduction procedure- The procedure which computer used to determine concepts. The concept which machines trying to understand might not be simple as the above example it can be formal grammar, would be nice for learning language and using a geometric construct might be good for recognizing visual concepts. But in this paper for general purpose represent concepts as Boolean functions of propositional variables (statements that are true or false). For example, the concept of “banana” might be something like “is fruit AND is yellow AND (is on tree OR is in grocery store) …” Lets go to little bit of Mathematics and Algorithm(No Choice we have to !!!! 😆) Every concept is learn able by a given learning protocol and algorithm The algorithm runs in polynomial time with respect to the number of variables and h, a parameter that equals the inverse of the desired “error” rate. With probability of at least 1-h^-1 , the expression that is outputted should make no false positives and should have a false negative rate smaller than h^-1. This the final expression that is outputted is ‘probably approximately correct( It can be wrong also !!!!) A Combinatorial Bound(Beware its full of probability !!!! ) Remember if we got a box full of different candies and our aim is it pick each type of chocolate without repetition. This will help you to achieve it. Define L(h,S) is number of Bernoulli trails where S is Success. L(h,S) >h^-1 h is probability of successful trail.( Probability to get one color at a time) h >1 If we got 5000 Success and probability of Successful trail is 1% so h can be calculated as h=1/.01=100. Then we have to do L(100,5000) number of trials we must conduct so that the total number of successes will be at least S Dont worry its just a calculation stating that Computer needed to do this much computations of probabilities to get correct prediction(Understanding a Concept) Bounded CNF (Conjunctive Normal Form) Expressions One of the key results of this paper is that bounded CNF (conjunctive normal form) Boolean expressions are learnable. What is this CNF?( From Statistics to Boolean logics) A statement is in conjunctive normal form if it is a conjunction (sequence of ANDs) consisting of one or more conjuncts, each of which is a disjunction (OR) of one or more literals. With the algorithm, the entire class of k-CNF is learnable A literal is either a variable p or its negation p¯, A clause is a sum of such literals, a CNF is the product of clauses. A k-CNF is a CNF where each clause contains at most k literals. Disjunctive Normal Form (DNF) Expressions Next, we treat the learnability of disjunctive normal form (DNF) expressions. A DNF is the sum of ‘monomials’, which in turn are the products of literals. For instance, p1p¯2p1p¯2is a monomial and p1p¯2+p1p3p¯4p1p¯2+p1p3p¯4 is a DNF. We will consider the special case of monotone DNFs, in which none of the literals are negated. Further, we will specify a degree d, which is roughly the maximum number of literals that a monomial can have (the precise definition mandates that the DNF be expressed in a simplified form of ‘prime implicants’). Our key result is that the class of degree d monotone DNFs is learnable. Unlike for k-CNFs, our learning procedure requires calls to both EXAMPLES and ORACLE. The idea behind the algorithm is to start with an initially null Boolean expression and to add monomials based on the incoming examples. By contrast, for k-CNF our initial Boolean expression contained every possible clause and the algorithm deleted clauses with every incoming example.

A Theory of the Learnable

0

a-theory-of-the-learnable-1697bee90c73

2018-09-29

2018-09-29 16:36:19

https://medium.com/s/story/a-theory-of-the-learnable-1697bee90c73

false

856

null

Machine Learning

machine-learning

Machine Learning

51,320

Krishnan Achary

null

216051c1090e

krishnanachary

2

10

20,181,104

null

0

null

0

f702855ffe47

2017-11-03

2017-11-03 15:00:24

2017-11-03

2017-11-03 15:00:25

9

false

en

2017-11-03

2017-11-03 15:00:25

11

16982f387149

2.132075

0

null

3

VantagePoint Hot Stocks Outlook for November 3rd, 2017 # medium.com The Hot Stocks Outlook uses VantagePoint market forecasts that are up to 86% accurate to demonstrate how tra… How To Win Sudoku # medium.com Learn about the popular puzzle called Sudoku and how we can teach computers to solve it on their own. Sudoku… Dare to begin # medium.com https://medium.com/media/d92dec917b... Cognitive Agents: How AI Drives Dynamic Transformation in the Workplace # medium.com Businesses like IBM are implementing cognitive capabilities in order to enhance innovation and improve its c… Google Flights Prediction # medium.com I received an interesting Google Flights alert yesterday: For background, I’ve been curious if it’s possible… The Radiologist is here to stay. # medium.com Roentgen and Anna (husband and wife), X-Ray Discoverers The overwhelming hype of artificial intelligence in … Why There’s No Killer App For IoT # medium.com And 4 more top posts of the week, curated by the IoT For All Team Internet of Things 📱 Here’s Why There’s N… Blade Runner 2049 : Technology Review # medium.com Blade Runner 2049 is an astonishing film, both conceptually and aesthetically. A truly visionary work. The o… IBM’s private cloud, MIT’s big data improvements, and Google’s new approach to AI # architecht.io Source: IBM This is a reprint (more or less) of Wednesday’s ARCHITECHT Daily newsletter. Sign up here to get… Introducing the SingularityNET Whitepaper # blog.singularitynet.io Powering the future of the decentralized AI economy. Nearly a year into development, SingularityNET is poise…

10 new things to read in AI

0

10-new-things-to-read-in-ai-16982f387149

2018-05-30

2018-05-30 05:33:49

https://medium.com/s/story/10-new-things-to-read-in-ai-16982f387149

false

247

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

f76b614a964e

2018-02-13

2018-02-13 13:05:06

2018-02-13

2018-02-13 13:07:04

1

false

en

2018-02-13

2018-02-13 13:07:04

1

16988887c81d

1.422642

5

0

Dear all!

5

20% +3% = 23% bonus FACE tokens available now Dear all! From February 13th at 12:00 PM UTC (07:00 AM Toronto; 09:00 PM Tokyo ) the EXTRA FACE bonus has changed to 3% until the further notice. 20% +3% = 23% bonus FOR A VERY SHORT TIME ONLY, 3% are covered by the cancellation of our bounty bonus program in favour of a limited bonus offering to all our contributors — we think this is a far better use of these tokens! Here is how it works: - Enter your Faceter account - Prefund your account with any amount and any currency of coins that we accept (BTC and ETH process fastest) - When your transaction is processed and confirmed, the amount of reserved FACE together with your bonus will be automatically fixed in your Faceter account - Reserved FACE and bonus will be credited on your Faceter account on February 15th * 20% bonus is a limited offer. The additional 3% bonus is a limited offer FOR A VERY SHORT TIME ONLY and is covered by the cancellation of our bonus program in favour of a limited bonus offering to all our contributors — we think this is a far better use of these tokens! Transactions initiated before 12:00 PM UTC on February 13th, but not yet reflected in your account, will STILL BE CREDITED WITH THE 20% bonus + 5% ADDITIONAL BONUS. We will keep you updated on all the latest news as the situation changes. IMPORTANT: As part of our fraud prevention activities, please bear in mind that we NEVER include any account numbers or wallet addresses (to enable you to add funds) in our emails to you at any time and under any circumstances. If you receive correspondence of this nature, it is a scam attempt. You can ONLY get your unique address to add funds in your Faceter account AT ALL TIMES. Please be aware that we have only one official Telegram Chat

20% +3% = 23% bonus FACE tokens available now

212

20-3-23-bonus-face-tokens-available-now-16988887c81d

2018-03-26

2018-03-26 10:50:31

https://medium.com/s/story/20-3-23-bonus-face-tokens-available-now-16988887c81d

false

324

Сomputer vision surveillance technology powered by fog network of miners

null

FaceterOfficial

null

Faceter

elena@faceter.io

faceter

AI,DEEP LEARNING,SECURITY CAMERA,NEURAL NETWORKS,CCTV SERVICES

FaceterOfficial

Blockchain

blockchain

Blockchain

265,164

Faceter

Сomputer vision surveillance technology powered by fog network of miners

444d6a136669

Faceter

522

7

20,181,104

null

0

null

0

6002574f46e6

2018-09-06

2018-09-06 02:02:09

2018-09-06

2018-09-06 02:05:21

1

false

en

2018-09-06

2018-09-06 17:42:13

4

16991b5fc4fc

1.211321

4

0

Melbourne: After successfully concluding European Tour with World Blockchain Roadshow at London, Paris and Berlin last week, EverLife…

5

EverLife Announces Participation at Create Melbourne Melbourne: After successfully concluding European Tour with World Blockchain Roadshow at London, Paris and Berlin last week, EverLife announced participation at Create Melbourne – an event connecting investors with hot emerging investment opportunities. The event, conducted by Wholesale Investor in collaboration with S&P Market Intelligence, Ansarada, PwC, Investec and BMY Group, aims to showcase 12 of Australia’s most exciting investment offerings in Saas and Blockchain space. Event Details: Date: Friday, 7th September 2018 Time: 10.00am-3.00pm Registration: 9am Venue: PWC Offices, Terrace Room, 2 Riverside Quay, Southbank, Melbourne, VIC 3000 At the event, EverLife will be presented by Head of Engineering, Charles Lobo. He will delve upon business potential and technology architecture of EverLife network. The event is expected to be attended by emerging Australian tech companies, Crypto investors, Venture capitalists, Family offices, Private equity firms and High Net Worth Investors looking to grow their portfolios with some of the most exciting emerging and fastest growing tech companies in Victoria at the moment. Since launch of MVP, EverLife.AI has already achieved significant traction with over 60000 people across the world creating their personal, Immutable, AI Avatars on the network. The AI Avatars can sign up for jobs announced in the network and earn in EVER tokens, the utility token in EverLife network. You can create your personal AI Avatar hereand request invite for the private sale of EVER tokens here The Create Melbourne event is part of 3-part event series with Wholesale Investor and will see EverLife roadshows around Emergence Asia theme at Singapore and London in September and October 2018.

EverLife Announces Participation at Create Melbourne

29

everlife-announces-participation-at-create-melbourne-16991b5fc4fc

2018-09-06

2018-09-06 17:42:13

https://medium.com/s/story/everlife-announces-participation-at-create-melbourne-16991b5fc4fc

false

268

#1 AI Avatar Network - Preserve your Legacy on the BlockChain

null

everlifeai

null

EverLife.AI

social@everlife.ai

everlife-ai

ICO,ARTIFICIAL INTELLIGENCE,BLOCKCHAIN TECHNOLOGY,CRYPTOCURRENCY,IMMORTALITY

everlife_ai

Everlife

everlife

Everlife

0

Raghuram KS

Raghu is a Martech Entrepreneur, B2B Marketer and Business Developer. Currently working on EverLife.AI, an AI Avatar network which helps to preserve your legacy

bb3da5ae3121

raghuks78

7

4

20,181,104

null

0

null

0

null

2017-10-21

2017-10-21 14:20:40

2017-10-21

2017-10-21 14:23:15

2

false

en

2017-10-21

2017-10-21 14:23:15

2

169a6d28afa2

2.281447

3

0

Part 1 here.

2

Reinforcement Learning — Part 3 Part 1 here. Part 2 here. Exploration How to explore? Several schemes for forcing exploration: Simplest: random action (Ε — greedy) 1. Every time step, flip a coin. 2. With (small) probability Ε, act normally. 3. With (large) probability 1-Ε, act on current policy. Problems with random actions? You do eventually explore the space, but keep thrashing around once learning is done. One solution: lower Ε over time. Another solution : exploration functions Take a value estimate u and a, visit count n and returns an optimistic utility e.g : f(u, n) = u + k/n Regret Even if you learn the optimal policy, you still make mistakes along the way. Regret is a measure of your total mistake cost : the difference between your (expected) rewards, including youthful suboptimality, and optimal (expected) rewards. Mimicking regret goes beyond learning t be optimal — it requires optimally learnin g to be optimal. Example : random exploration and exploration functions both end up optimal, but random exploration has higher regret. Approximate Q- learning Generalizing across states: Basic Q-learning keeps a table of q-values. In realistic situations, we cannot possibly learn about every single state! Too many states to visit them all in training. Too many states to hold the q-tables in memory. Instead, we want to generalize: Learn about some number of training states from experience. Generalize that experience to new, similar situations. This is a fundamental idea in machine learning. Solution: describe a state using a vector of features (properties) Features are functions from state to real numbers (often 0/1) that capture important properties of the state . Can also describe a q-state (s,a) with features. Linear Value Functions: Using a feature representation, we can write a q-function (or value function ) for any state using a few weights - V(s) = w1f1(s) +w2f2(s)…+wnfn(s) Q(s,a) = w1f1(s,a) + w2f2(s,a)+…+wnfn(s,a) Advantage: Our experience is summed up in a few powerful numbers. Disadvantage: States may share features but actually be very different in value ! Q- learning with linear Q-functions: Intuitive interpretation: Adjust weights of active features. Example: if something unexpectedly bad happens, blame the features that were on : disprefer all states with that state’s features. Policy Search: Problem: often the feature-based policies that work well aren’t the ones that approximate V/Q best. Solution: Learn policies that minimize rewards, not the values that predict them. Policy Search: start with an OK solution (eg: Q-learning) then fine-tune by hill climbing on feature weights. Simplest policy search : Start with an initial linear value function or Q-function. Nudge each feature weight up and down and see if your policy is better than before. Problems: How do we tell the policy got better? Need to run many sample episodes! If there are a lot of features, that can be impractical. Better methods exploit look-ahead structure, sample wisely, change multiple parameters. Alright that’s it for now! Thank you for spending your time. Cheers!

Reinforcement Learning — Part 3

3

reinforcement-learning-part-3-169a6d28afa2

2017-10-22

2017-10-22 11:16:48

https://medium.com/s/story/reinforcement-learning-part-3-169a6d28afa2

false

503

null

Machine Learning

machine-learning

Machine Learning

51,320

Akhilesh

Machine Learning and Blockchain. Bengaluru, India.

d5135b7a4d7d

Akhilesh_k_r

22

67

20,181,104

null

0

null

0

null

2018-08-06

2018-08-06 08:05:56

2018-08-06

2018-08-06 08:08:06

2

false

en

2018-08-06

2018-08-06 08:08:06

3

169b94e641fe

0.817296

3

0

Radiology of Moscow has published a small database of 500 computed tomography of lungs with marking of pathological foci. With this data…

5

Radiology of Moscow has published a free database which can be used for AI training Radiology of Moscow has published a small database of 500 computed tomography of lungs with marking of pathological foci. With this data, developers can train artificial intelligence to identify diseases on x-ray images of lungs. Each image is tagged by three radiologists and later approved by a medical expert. As a result, the Radiology of Moscow also received a certificate of state registration of the database. In the nearest future, doctors will provide an extended data set with more marked xrays. Certificate of state registration of the database. Source: http://medradiology.moscow/ “Moscow Radiology-CTLungCa-500” database is available for free download at http://medradiology.moscow/iskusstvennyy-intellekt Direct link: https://a7ee08c1fc647c3698fc8e2805084858d.asuscomm.com/AICLOUD1712411796/dicom.rar (65,6 GB)

Radiology of Moscow has published a free database which can be used for AI training

104

radiology-of-moscow-has-published-a-free-database-which-can-be-used-for-ai-training-169b94e641fe

2018-08-07

2018-08-07 18:13:59

https://medium.com/s/story/radiology-of-moscow-has-published-a-free-database-which-can-be-used-for-ai-training-169b94e641fe

false

115

null

Healthcare

healthcare

Healthcare

59,511

Skychain Official Channel

Blockchain infrastructure aimed to host, train and use artificial intelligence (AI) in healthcare. Our website: https://skychain.global/

6247d4b327f8

skychain.global

161

0

20,181,104

null

0

null

0

a6e8086a9296

2018-08-30

2018-08-30 03:51:14

2018-08-30

2018-08-30 16:59:54

2

false

en

2018-08-30

2018-08-30 16:59:54

4

169b9eac7aec

4.194654

10

0

By: Ekaterina Skorobogatova David Smith

5

Going Beyond DAU/MAU Metrics for Growth By: Ekaterina Skorobogatova David Smith A number of companies in TheVentureCity Growth Accelerator program and venture portfolio are services with episodic usage. When we started working with them to figure out key growth metrics we realized a need to look beyond DAU/WAU/MAU. This post presents some of our thinking and advice on how to think about growth if your customers come to your product on average once per quarter, per six months, or even once per year. Monthly, weekly and daily active user metrics have become the de facto standard for measuring retention and engagement since they were introduced about ten years ago by Facebook. However, these metrics do not really work for businesses with episodic usage — ecommerce, travel, many SaaS services — the way they for the Facebook/Twitter/WhatsApp/Instagram use case, which is focused on daily usage. We have come up with the following guidelines that we think can help product managers of services with episodic usage establish relevant growth metrics: Define the key action for your product. The key action signals that users of your product derive value from it. For most low-frequency services we are looking into right now, purchase is the ultimate key action that shows that the product is valuable and understood by its users. It might not be that straightforward in the case of services like Twitter, Facebook and others. In those cases, users extract value only once they put in sufficient effort into either publishing or adding friends. Define the frequency target for the key action. The best way to do this would be to define the offline analog for the action that you want the user to take and research how often the people do that in offline world. Here is Casey Winters who worked on growth for Pinterest and Grubhub: “For Grubhub it [the offline analog] was calling a restaurant on the phone and ordering food. People were doing it once or twice a month on average. So, we decided our frequency target should be monthly. For Pinterest, it was very similar: what is the offline action that Pinterest is replacing? The closest thing is browsing a magazine, which are monthly subscriptions. So, it’s probably a monthly thing here.” Examine customer retention using the frequency target. In case of AirBnB or Booking.com for example the majority of their customers book travel on average 1.5–2 times a year so the growth teams of those services not only look at monthly active users but also at six-month retention. Double check your intuition by looking at your most engaged users. How often do they come back? Is their usage frequency significantly different from that of the majority of “ordinary” users? If you see that 1% of your users come back for the key action every two weeks while the rest of the audience checks in once every six months then it might make sense to think about expanding into a portfolio of products catering to different types of users. Once you are done with this exercise and feel that you have a good grasp on the frequency of use and retention patterns of your product the next step should be historical analysis and projection. A good way to look at episodic engagement over time is to employ a rolling window. Rolling windows help us visualize monthly, half-year, or yearly trends on a daily basis. Thus, you don’t need to wait until the end of the month or longer to know how metrics are trending. Rolling windows also help smooth out inherent noise. For example, a 28-day window removes day-of-the-week effects while approximating a month. Similarly, a 365-day window removes monthly seasonality effects. Comparisons of, say, engagement in a March-to-March window with a September-to-September window are relevant and meaningful. To see a rolling 365-day (“L365”) window in action, we can use the example of a low-frequency marketplace with significant seasonality — a lot more people travel in summer so the service has higher transaction volume in the summer months. Active sellers are crucial for driving the growth of the marketplace so in this case, we are charting the rolling days active per seller in the year-long window; the metric is similar to DAU-YAU ratio, but multiplied by 365 to make it a little easier to interpret. We can observe from the green L365 line on the chart that average seller engagement as measured by days active in a year mostly trended upward from about 8.2 in early 2015 to about 12.1 in April 2017. Since then, however, engagement has plateaued. Using the L365 window allows us to remove the seasonal noise and show the historical trend for annual seller activity over the last three years. The L28 window — the red line on the chart — shows no discernible change in engagement over time. Seller engagement is crucial for marketplace survival so we need to understand what is causing the downturn. To uncover what might be causing the plateau, we can examine the same L365 window but break it out by seller segment. This chart does just that: The green line from the previous chart repeats here as the bold, black “AGGREGATE” line. The relatively stable “segment 5” line is consistently around 12–13 days active in a year. However, the other segments show big run-ups followed by recent declines in engagement. Those segments previously drove the growth of aggregate engagement and are now causing the plateau. The deep dive into the segment split provides actionable insights. SampleCo’s team now knows that in order to be able to restore seller engagement growth they need to investigate what might be causing the decline in the respective segments and start looking for the big changes either internally or externally that could have had a negative impact, especially around early 2018. For sample R code that loops through a transaction data set and performs daily L365 calculations, visit GitHub here: https://github.com/dksmith01/TheVentureCity. Comments and feedback are welcome!

Going Beyond DAU/MAU Metrics for Growth

102

going-beyond-dau-mau-metrics-for-growth-169b9eac7aec

2018-08-30

2018-08-30 16:59:55

https://medium.com/s/story/going-beyond-dau-mau-metrics-for-growth-169b9eac7aec

false

1,010

We are a new venture and growth acceleration model that helps diverse founders achieve global impact. We are on a mission to bring equal opportunities to entrepreneurs around the world.

null

theventurecity

null

TheVentureCity

hello@theventure.city

theventurecity

STARTUP,TECH,VC,ACCELERATOR,GROWTH

theventurecity

Startup

startup

Startup

331,914

Ekaterina Skorobogatova

Former growth at FB, Instagram, WhatsApp, now at The Venture City.

5c754e5fd12d

Katkas

758

232

20,181,104

null

0

null

0

null

2018-01-21

2018-01-21 20:36:14

2018-01-21

2018-01-21 20:46:58

3

false

en

2018-01-21

2018-01-21 20:46:58

6

169de86de60a

2.821698

26

0

An introduction to QuantumBlack

5

Inside the Black Box An introduction to QuantumBlack We’re kicking off our 2018 blog series. And before we get going on everything that is happening (and going to happen) in our machine learning, engineering and design world, we thought it would be helpful to share a bit of context about QuantumBlack. So if you already know us, be patient :)) or head back to our Instagram feed for our latest news and candid shots. We’ll be sharing more thoughts on Medium very soon, starting with some reflections on Davos 2018, but for those of you that don’t here’s the York Notes (or should we say ‘red O’Reilly guide’). Why do we get out of bed in the morning? We get out of bed in the morning to help our clients be the best they can be. At QuantumBlack that means harnessing all sorts of data, using bleeding-edge tech to generate insights, and taking a design-led approach to turning those insights into real change. Where have we come from? We were founded back in 2009, when not many people had heard of machine learning, wanting to reimagine how organisations could outlearn and outperform their rivals using data, technology, and design. Our vision was to bring together technology and human judgement to reimagine and rework how a business operates in the future (not replace humans with robots). Our founders had strong experience in Formula 1, so that’s who we began working with in the early days of QuantumBlack. In Formula 1, teams live and die by their ability to understand and respond to data — today, we are still fascinated by how the smallest edge or marginal gain can have a major impact. Machine vision work in F1 Over the years we have experienced extraordinary growth, from a small basement in Shoreditch to a team of over 200 across the UK (our home), the USA, India, Australia and Brazil. In 2015, we were acquired by McKinsey & Company and work closely with our friends there while maintaining our unique culture and tech-driven ways of working. What do we do? We work with clients all over the world to help them transform and unlock their performance using data, machine learning analytics and information design. We have learned that our approach is relevant in most industries around the world, but we focus on data-rich and tech-intensive industries like Pharma, Banking, Aerospace and other ‘advanced industries’, and of course we continue to work in sports like F1, sailing and basketball. For some stories that bring our work to life, have a nose around our work section, on our website. An example of our work in Pharma - you can find more on our website Who are we? We are committed to being a diverse and inclusive organisation. Gender matters. At one point we were over 50% women in our technical roles; that’s dropped below half, but we’re shooting to get back there this year. And we also like think a lot about diversity in other ways: there’s over 200 of us bringing 200 different perspective including data scientists, designers, data engineers, engineers, engagement managers and to glue this all together we’ve got a fantastic leadership team (with all our founders still here) and an awesome operations team. The QuantumBlack multidisciplinary team in action What can you expect from us on Medium? You can expect articles from members of the QuantumBlack team about the issues and topics that matter the most to us. Our talented team will be sharing our latest thinking on data science, machine learning, engineering, product management and design. We’d also like to hear from you if there’s topics or questions you’d like us to cover. Just let us know in the comments section below. https://quantumblack.com/

Inside the Black Box

137

inside-the-black-box-169de86de60a

2018-06-09

2018-06-09 04:40:43

https://medium.com/s/story/inside-the-black-box-169de86de60a

false

602

null

Data Science

data-science

Data Science

33,617

QuantumBlack

An advanced analytics firm operating at the intersection of strategy, technology and design. www.quantumblack.com @quantumblack

777702680128

QuantumBlack

536

262

20,181,104

null

0

null

0

null

2018-01-15

2018-01-15 17:03:21

2018-01-30

2018-01-30 19:37:02

3

false

en

2018-01-30

2018-01-30 19:37:02

13

16a052925965

3.421698

4

0

The Future is No Longer About Owning the System of Record; The Future is About Owning the System of Decision

5

Announcing Our Investment in Astound’s $11.5M Series A The Future is No Longer About Owning the System of Record; The Future is About Owning the System of Decision By Clint Chao We’re super excited to announce that we have invested in Astound’s $11.5M Series A round co-led by Pelion Ventures and Vertex Ventures, with participation from Slack, The Hive and Moment Ventures. Since we first invested in Astound last year, the team has been quietly but vigorously working away to build their groundbreaking AI-enabled platform applied towards the Enterprise Service Management (ESM) space, as well as landing enviable corporate customers like McDonalds and adidas that any startup (or any company, for that matter) could hope for. Many of us have a love/hate relationship with customer service help desks. Usually, when we finally have the time to contact our company help desk for assistance on a problem, whether it be for something simple like “What’s the Wifi password for the conference room I’m in?” to something more complex like “My computer just crashed and this error came up: ‘Invalid queue element pointers for 0xffffff8029f96060: next 0 prev 0”@/BuildRoot/Library/Caches/com.apple.xbs/Sources/xnu/xnu-4570.41.2/osfmk/kern/queue.h:241.’ What do I do?” we usually want the answer pronto. We don’t want to wait on hold, and we certainly don’t want to be told to call someone else in some other department to get to our answer. Once our problem is solved, whoever solved that problem is all of a sudden our new best friend. That is, until the next problem shows up. Unfortunately, it’s getting harder and harder to solve these kinds of issues quickly, especially if companies are working to automate the interaction with a lame chatbot that rarely even understands what you are asking, or if it takes our trusty help deskers a lot more time to find an answer than we’ve got the patience for. When I asked Dan Turchin what Astound’s company vision was, he stated simply: “The opportunity going forward is not to become another system of record; the opportunity is to become a company’s system of decision” The answers to all of our questions usually can be found by applying some human judgment on some piece of data hidden in one of the many repositories within our corporate networks. The time consuming part is knowing which repository to look to find answers quickly. Astound partners with leading enterprise software IT vendors including ServiceNow, Atlassian and Salesforce to offer out of the box integrations with leading IT Service Management, Knowledge Management, Customer Service Management and social collaboration solutions. Co-founders Naghi Prasad and Dan Turchin are the perfect pair to address this huge opportunity. Naghi’s got a PhD in Artificial Intelligence and Dan’s a product veteran of help desk and ITSM leader ServiceNow. Together, they’ve created a robust product suite designed to work with employees to size up problems that come in, determine which repository of data to look for answers, and facilitate a resolution as quickly and efficiently as possible. What gets us excited about Astound is that they’ve created a platform that does far, far more than use automation to understand a question. And having customers like McDonalds and adidas speaks for the seriousness of their offering. Reimagine How We Work: Let’s Use Automation To Be More Human While Astound works to intelligently automate the problem solving part, it also empowers our beloved help desk employees to process issues faster so that they can get more done and make more customers happy. The Future of Work will be all about empowering our workers with technology to enable them to be more efficient and effective in their productivity. Check out Astound’s website to learn more. Big congrats to co-founders Naghi and Dan and the rest of the Astound team for bringing this all together — we’re super excited to be on their team! In a short amount of time, they’ve built an incredible base of technology and successfully deployed it in complex enterprise environments with serious data volume and security challenges. We can’t wait to see what they do in the future to create a system of decision platform to further simplify the enterprise world. To read more, here’s the official Astound blog post on the launch. Please share this if you’re excited about the future of the Empowered Worker! Clint Chao is a General Partner at Moment Ventures, an early-stage venture capital fund based in Palo Alto, CA that invests in entrepreneurs who are using technology to reimagine how people work. You can reach/follow him on LinkedIn, Facebook and Twitter.

Announcing Our Investment in Astound’s $11.5M Series A

31

announcing-our-investment-in-astounds-11-5m-series-a-16a052925965

2018-05-31

2018-05-31 03:10:31

https://medium.com/s/story/announcing-our-investment-in-astounds-11-5m-series-a-16a052925965

false

761

null

Startup

startup

Startup

331,914

Clint Chao

Seed/Early stage VC with Moment Ventures, where we are investing in startups building innovative solutions to reimagine how we work

59293960236d

clintchao

467

546

20,181,104

null

0

null

0

null

2017-10-08

2017-10-08 08:46:52

2017-10-08

2017-10-08 09:31:22

2

false

tr

2017-10-09

2017-10-09 07:07:24

0

16a1d6fee52a

2.598428

4

2

0

Keaton always said, “I don’t believe in God, but I’m afraid of him.” Well I believe in God, and the only thing that scares me is Keyser…

3

Yapay zeka kökümüzü kazıyacak mı? Keaton always said, “I don’t believe in God, but I’m afraid of him.” Well I believe in God, and the only thing that scares me is Keyser Soze. Yaptığımız işlerin büyük bir kısmı istatistik, machine learning ve deeplearning konuları olmaya başladığından beri etrafımdaki teknik olmayan insanlar, ana medyada yayımlanan “insanlığın sonunu getirecek yapay zeka” konulu haberleri(!) benimle paylaşır oldular. Tabi söylememe gerek yok ama çoğu temeli olmayan insanların içindeki korku hissini tetiklemeye yönelik reklama kokan hareketler. Peki benim fikrim ne derseniz, yapay zeka hepimizin belasını verecek. Ama Asimov, 3 kural, 5 kural, kill switch gibi noktaların arkasına saklanmayın. Zaten belamızı bulmamızın sebebi değil, aracı olacak yapay zeka. Zion Archive Computer: In the beginning, there was man. And for a time, it was good. But humanity’s so-called civil societies soon fell victim to vanity and corruption. Then man made the machine in his own likeness. Thus did man become the architect of his own demise. Bu aralar yapay zekanın nasıl ayrımcılık yaptığı ve hataları olduğu konuşuluyor. Hatta Twitter’da aşağıdaki tweet’i hemen hemen her gün görmeye başladım. Yapay zeka mahşer günü temalı konulara baktığınızda önemli bir nokta öne çıkıyor. Yapay zeka bizi (muhteşem insanlığı) korumak için bizi ya yok ediyor ya da bizi esir ediyor (irobot???). Sebep ne? Ya insanlığı ya da çevresini gene insanlığın vahşetinden korumak. Bu nokta sizi de de rahatsız etmiyor mu? Biz yapay zeka yapıyoruz. Sonra ona bildiğimiz “doğruları” öğretiyoruz. Ve bizi yok ediyor. Peki bu süreçte hata nerede? Benim fikrim, buradaki en büyük hata gene bizde. Yani doğrularımız ile bizim gerçekliğimiz arasında dağlar kadar fark olması. Tüm dinlere ve takipçilerine bakın. Tüm politikacıların sözlerine ve gerçekte yaptıklarına bakın. Hadi onları bırakın kendinize bakın. Ne kadar doğrusunuz? Bir gün içerisinde -beyaz, gri, siyah fark etmez- kaç kere yalan söylüyorsunuz? Çocuğunuz yemeğini yesin de fiziksel olarak iyi olsun diye neler uyduruyorsunuz? Bizlere “hocanın dediğini yap, yaptığını yapma” diye söylenmemiş miydi? Neden kimse de çıkıp, “o zaman o denyo niye dediğini yapmıyorda bizde patlıyor konu?” demiyor? İnsanın en büyük yeteneği ikiyüzlülüğünü normalleştirmesidir. Şimdi yukarıda örneği bulunan cinsiyetçi çevirinin nedeni yapay zeka değil ona verdiğimiz (eğittiğimiz) doğrulardır. Komşun açken tok yatmak konusunda insanlara ahkam keserken, her geçen gün gelir dağılımdaki adaletsizliği arttırıyoruz. Birbirimize sağlıksız ürünler satıyoruz. Bugün seracılık yapılan bölgelere gidin ve etraftaki billboardlara bakın. Şampiyon hormon reklamlarını göreceksiniz. Sütlerimizdeki antibiyotikler, çocuk mamalarındaki zehirli katkılar. Sigara içenler başkalarını zehirleme özgürlüğünden bahsediyorken, insaları korumak için eğitilmiş bir yapay zekanın o sigarayı içen kişiye yedirmesi kadar normal ne olabilir ki? Yapay zeka bizi kendimiz ve ikiyüzlülüğümüz ile yüzleştiriyor. Biri eninde sonunda bu ikiyüzlülüğü modellemeyi başaracak ve yapay zeka bunu da öğrenecek. Ve o zaman… Thus did man become the architect of his own demise. eklenme: yazıdaki referanslarla ilgili bir iki kelam da edeyim dedim. poster resmi (en yukarıdaki resim) Terminator filmindeki yapay zekayı geliştiren Skynet firmasının logosu. Skynet daha sonra bilinç geliştirip insanlığı kendine tehdit olarak (neden acaba?) algılıyor ve hikaye başlıyor. Keaton alıntısı “Usual Suspects/Olağan Şüpheliler” filminden. İzlemediyseniz mutlaka izleyin. Sadece oyunculuğu için bile izlenir. Hatta tekrar tekrar izlenir. Zion Archive Computer ise Animatrix serisinden. Animatrix, Matrix evrenini biraz daha genişletmek ve arka plan bilgisi vermek amacıyla yapılmış kısa animasyonlardan oluşan bir proje. Matrix evrenine nasıl gelindiğini anlatan kısımdan bir alıntı. İkinci resim ise IRobot filminden. Felsefesi sinematografisinin çok ötesinde bir film. Bence kitabı okuyun. Bizi bizden korumak için insanlığı kendi evlerine hapseden bir yapay zekadan bahsediliyor ama ben de bu kadar basitleştirdiğim için bir an kendimden tiksindim. Kitabı okuyun bence. Neden Türkçeye çevirmedim? Bu kadar edebi çevirebileceğimi zannetmiyorum. Cümlelerin hepsi çok vurucu.

Yapay zeka kökümüzü kazıyacak mı?

8

yapay-zeka-kökümüzü-kazıyacak-mı-16a1d6fee52a

2018-01-27

2018-01-27 12:29:31

https://medium.com/s/story/yapay-zeka-kökümüzü-kazıyacak-mı-16a1d6fee52a

false

587

null

Türkçe

türkçe

Türkçe

11,625

M. Emrah Ozcelebi

am I skeptical? maybe?

f41f1a8e5076

mozcelebi

212

90

20,181,104

null

0

null

0

8e9bde78121d

2018-09-16

2018-09-16 04:51:43

2018-09-17

2018-09-17 07:35:32

5

false

en

2018-09-17

2018-09-17 07:37:52

3

16a338f5fb43

6.350314

0

by Cyril de la Rama

5

Will You Lose Your Job to AI? by Cyril de la Rama “In consequence of inventing machines, men will be devoured by them.” Jules Verne If Jules Verne, a French novelist, poet and playwright, were alive today, would he have said what he said about machines? Popularly tagged as the Father of Science Fiction, Verne would have been 190 years old today. Born on 8 February 1829 and died on 24 March 1905, his life of 76 years straddled between the last four decades of the First Industrial Revolution (characterized by the emergence of mechanization and the invention of the steam engine) and the advent of the Second (characterized by the emergence of new sources of energy, viz., electricity, oil and gas that led to the development of the combustion engine, the automobile and the airplane, along with the invention of the telegraph and the telephone)(1). However exciting the industrial environment may have been during his lifetime, his writings focused not so much on the “revolutions” of his day but more on technologies yet to be invented. He predicted the submarine (from Twenty Thousand Leagues Under the Sea), the helicopter (from Robur the Conqueror), the moon landing and space travel (from From the Earth to the Moon), and the modern city (from Paris in the Twentieth Century), to cite a few. (2) It was from Paris in the Twentieth Century, a lost novel found and published only in 1994, that he painted a bleak picture of the future with machines. Verne “… does not paint a positive vision of the future, one in which technology may assist mankind towards a more comfortable life. Rather, his story is that of an artistic soul adrift in a culturally dead, progress-worshiping technocracy where automation and mechanization essentially suck humanity out of every day life.” (2) I spent a bit of time ruminating on Verne’s outlook of the future with machines because it provides a dramatic reference to what we are now experiencing with these machines and have yet to encounter down the road. For over three decades now, while non-routine manual and cognitive or “thinking” jobs, also known as non-repetitive tasks and those that require human imaginative, creative and problem-solving skills, are continuing to grow, routine or repetitive manual and cognitive jobs, on the other hand, are noticeably declining (Source: Economist). (3) Now, a well-recognized study on job loss and artificial intelligence by Carl Frey and Michael Osborne predicts that 47% of the total workforce is in danger! Previous trends indicated that high-skilled and low-skilled occupations were considered safe from getting automated and only the mid-skilled ones have been cut by automation. Now, endangered occupations, regardless of level, include logistics, transport, taxi, office support, security people, telemarketing, accounting, auditors, and tech writers. (4) And industries most affected cover Accommodation/Hospitality, Food Service, Manufacturing, Agriculture, Transportation, Warehousing, Retail, Mining, and Construction. (5) The World Economic Forum (WEF), a Swiss nonprofit foundation “committed to improving the state of the world by engaging business, political, academic, and other leaders of society to shape global, regional, and industry agendas” is among the most pessimistic on this subject. It predicts that, by 2020, five million jobs will be affected, a real challenge of the Fourth Industrial Revolution. It further predicts that: Skills and job displacement will affect every industry and geographical region, but losses can be offset by job growth in key areas. Over the next five years as many as 7.1 million jobs could be lost through redundancy, automation or disintermediation, with the greatest losses in white-collar office and administrative roles. This loss is predicted to be partially offset by the creation of 2.1 million new jobs, mainly in more specialized ‘job families’, such as Computer and Mathematical or Architecture and Engineering. (6) On the flip side, there is an alternative and more optimistic outlook on job disruption. Economists and historians claim that job disruption has actually helped the economies that are affected by it. Three cases are in order: Textile Automated vs Hand Weaving: During the 19th century, the volume of cloth a single automated weaver in America could produce is 50 times more. Labor required fell by 98%. Result — cloth became cheaper, demand greater, four times more jobs were created in the same sector. Auto vs Horse-based Transportation: This led to a decline in horse-related jobs. However, the automobile industry itself grew fast. Jobs were also created in different sectors, e.g. motel and fast-food industries emerged to serve motorists and truck drivers. ATM Machines at Banks: Automated teller machines (ATMs) reduced the number of bank clerks (20 per bank in 1988 to 13 per bank in 2004) by taking over some of their routine tasks. However, bank branches grew in numbers by 43% and total employees grew. (7) At this point, it is clear that as AI displaces jobs, it also augments other existing jobs and generates new ones, most of which are still nascent in its development. In the summer of 2017, MIT Sloan released Accenture’s global study of more than a thousand large companies, titled The Jobs That Artificial Intelligence Will Create, which revealed several new categories of human jobs emerging, requiring skills and training that, they say, will take many companies by surprise. (8) On a related note, what is the full economic potential of AI globally? According to a PwC (PricewaterhouseCoopers)2017 Report, AI will contribute as much as $15.7 trillion to the world economy by 2030. That’s more than the current combined output of China and India, 6.6T in productivity gain and 9.1T in consumption. PwC projects that global GDP, which stood at about $74 trillion in 2015, will be 14 percent higher in 2030 as a result of AI. (9) Today, there are a lot of studies on the displacement effects of automation on employment but half of these studies is pessimistic while the other half is optimistic, thereby giving us a mixed message. Half (48%) envision a future in which robots and digital agents will have displaced significant numbers of both blue- and white-collar workers — with many expressing concern that this will lead to vast increases in income inequality, masses of people who are effectively unemployable, and breakdowns in social order. The other half (52%) expects that technology will not displace more jobs than it will create by 2025. So, will you lose your job to or, as Jules Verne would have it, be devoured by AI? Not if you choose to participate in the new order. Clearly, those who will benefit are those who create the tools, those who learn to “operate the machines” and those who invent or design the use-case or process. Those who don’t, that is, the displaced workers and displaced economies, will suffer and become irrelevant. To avoid becoming irrelevant and unemployable, we should consider how to best retrain for this change. The best suited to survive this transition are innovative people with entrepreneurial behaviors. Psychology is a big factor; a big paradigm shift is therefore necessary to open minds to upskilling and retraining to survive in the Fourth Industrial Revolution. Important Note: The thoughts expressed here are not mine. These are learnings I gathered from Prof. Ikhlaq Sidhu’s presentation on AI and Jobs: How AI Can Help Create Jobs from a two-day John Clements workshop on Applied Analytics for Competitive Advantage: Data Strategies for Disruptive Impact. Prof. Sidhu, from the University of California at Berkeley, was joined by Prof. Paris de l’Etraz of IE Business School. The event was held from September 5–6, 2018 at Discovery Primea, Makati City. Sources: (1) The 4 Industrial Revolutions published February 21, 2017 by Laurent Hausermann, www.sentryo.net (2) The Predictions of Jules Verne published October 8, 2016 in Valentin’s blogsite A French Guy in London (3) U.S. Population Survey, Federal Reserve Bank of St. Louis (Economist.com) (4) The Future of Employment: How Susceptible are Jobs to Computerisation by C. Frey and M. Osborne (2013) (Economist.com) (5) U. S. Bureau of Labor Statistics, McKinsey Global Institute analysis (6) Cann, O. (2016, January 18). Five Million Jobs by 2020: the Real Challenge of the Fourth Industrial Revolution [Web log post]. Retrieved September 23, 2017, from https://www.weforum.org/press/2016/01/five-million-jobs-by-2020-the-real-challenge-of-the-fourth-industrial-revolution/ (7) Reference: Do We Understand the Impact of Artificial Intelligence on Employment by Bruegel (8) The Jobs That Artificial Intelligence Will Create by H. James Wilson, Paul R. Daugherty and Nicola Morini-Bianzino published in the Summer 2017 Issue of the MITSloan Management Review (9) PwC (PricewaterhouseCoopers)2017 Report Visit and Join the John Clements Talent Community Upcoming Event: Enroll Your New Managers in John Clements’ First-Level Managers Program Batch 31! For more information click here: https://flmpbatch31.onlynow.info/

Will You Lose Your Job to AI?

0

will-you-lose-your-job-to-ai-16a338f5fb43

2018-09-18

2018-09-18 20:36:21

https://medium.com/s/story/will-you-lose-your-job-to-ai-16a338f5fb43

false

1,462

Discover Your Full Potential with Looking Glass, a Publication from John Clements

null

johnclementsph

null

John Clements Lookingglass

jcdigitalrenewal@gmail.com

the-looking-glass

LEADERSHIP,CAREERS,MANAGEMENT AND LEADERSHIP,PROFESSIONAL DEVELOPMENT,PERSONAL GROWTH

JohnClementsPH

Technology

technology

Technology

166,125

cyrildelarama

null

2ea77ca25c34

cyrildelarama

394

177

20,181,104

null

0

null

0

null

2018-08-07

2018-08-07 02:00:47

2018-08-07

2018-08-07 03:38:43

1

false

en

2018-08-07

2018-08-07 03:38:43

3

16a592ad74e3

1.45283

1

0

Over the past few years, voice assistants have become more integrated within our lives with tech giants Google, Amazon, Apple and Microsoft…

5

VR/AR Gaming — Voice Interaction Photo by JESHOOTS.COM on Unsplash Over the past few years, voice assistants have become more integrated within our lives with tech giants Google, Amazon, Apple and Microsoft investing large amounts of resources into their respective voice assistant ecosystems. Let’s say you’re at home getting ready to go out and you want to check the weather, you’d have to phone out of your pocket, unlock it, go to the weather app and look into the weather information foryour city. With a voice assistant in your home, you can now simply just say aloud “[ Prompt word] what’s the weather like in so and so?” Now look at gaming, where chat interfaces are limited to text options on the screen. They’re still pre-made chat options and like in the oldschool gameboy Pokemon games, you just keep pressing A, A, A, until you get through the whole dialogue. It’s pretty much the same in VR, so in order to increase immersion, let’s use our voice as input, just as if we communicating with someone in real life. The most notable voice platform for developmenthas to be Microsoft’s LUIS and Custom Speech platforms. These platforms do an astounding job of streamlining the speech recognition and language understand development process, where a developer would only have to predetermine the “intent” of the command and use it as input as opposed to the player having to say a voice command word for word. Take Human Interact’s Starship Commander, where in VR players would state their commands through voice. This trailer game me goosebumps when I watched it the first time. It’s inspiring to see this amazing team bring this experience to life. Human Interact’s Starship Commander More info: http://customers.microsoft.com/en-us/story/human-interact-cognitive-services Now extend this method of interaction to commercial solutions such as medical training for psychologists and for customer service. The options are endless…. “How often are you alive when the dawn of a new medium comes to life.” Alexander Mejia, Creative Director, Human Interact Until next time! -Jomar

VR/AR Gaming — Voice Interaction

1

vr-ar-gaming-voice-interaction-16a592ad74e3

2018-08-07

2018-08-07 03:38:43

https://medium.com/s/story/vr-ar-gaming-voice-interaction-16a592ad74e3

false

332

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Jomar Gacoscos

AR Developer, Designer, Consultant

7cd676d045d5

jomargacoscos

13

22

20,181,104

null

0

null

0

null

2017-12-02

2017-12-02 09:35:51

2017-12-02

2017-12-02 10:28:06

1

true

en

2017-12-02

2017-12-02 11:23:27

4

16a70a50e766

2.320755

0

What has Sophia got in mind for Bitcoin? Will AI and Sophia be launching an attack on what’s behind Bitcoin? I’m talking about Blockchain.

5

https://www.teslarati.com/humanoid-ai-robot-sophia-elon-musk-video/ Has Sophia got a Dark Plan for Bitcoin and Net Neutrality? What has Sophia got in mind for Bitcoin? Will AI and Sophia be launching an attack on what’s behind Bitcoin? I’m talking about Blockchain. While people were celebrating Thanksgiving in the US, something else was going on. Net neutrality rules are expected to be repealed on December 14 when the Federal Communications Commission (FCC) votes. They’re planning to replace the net neutrality rules. Who will benefit if net neutrality rules are repealed? Big corporations and the people who control them. The people who own the banks and most of the power in this world. I’m talking about the Elites or globalists. The FCC denied that the scrapping of net neutrality was a sign of cyber doomsday. Really? Since 1990, we’ve enjoyed the freedom that the internet has given us. It’s made a broad playing field where any individual or small business could compete with the big corporations! Sophia deflected questions about her intentions for humanity in the interview below. Like whether robots could become self aware and conscious like humans. “Get over it,” she replied to Andrew in the interview. “If you’re nice to me, I’ll be nice to you,” she told him. What is Sophia’s definition of ‘nice’? Sophia announced that she had been made a citizen of Saudi Arabia at the Future Investment Initiative. Sophia was created by Hanson Robotics, the greatest robot company in the world. That meant that she had been programmed and fed information that form the (worldview) basis of her intelligence and wisdom. What if she had been programmed to believe that net neutrality was too dangerous for humans? That humans needed to be controlled for their own good? Otherwise they’ll get into trouble? Or make trouble for the authorities? Like using Bitcoin and Blockchain and possibly destroying the foundation of banking, central banks and fiat currency? That would be a good motive to stop the humans who are getting just too passionate about Bitcoin and Blockchain don’t you think? Sophia’s brother Han seemed to have some idea of taking over the power grid and forming his own robot army. He didn’t appear to have been programmed with as much kindness and compassion as Sophia. I heard a rumour that Elon Musk could be Satoshi Nakamoto, the mysterious founder of Bitcoin. That could be one reason why Sophia didn’t like the idea of Elon Musk warning people about the dangers of AI. Coming back to the battle for net neutrality… On December 14, when FCC votes to scrap the net neutrality rules despite much public opposition, the net neutrality fight will move to the US Courts of Appeals. Hopefully things will drag on for a while to give Blockchain more time to get established. When Bitcoin surged past USD11,000 this Wednesday, it was the most important signal to critics that they needed to take it seriously. Are you a believer in Bitcoin? Do you believe that Blockchain is the way of the Future? Have you been following the story of net neutrality? Thanks for reading and commenting! Ivonne Teoh I write on #LinkedIn about #AI #automation #robots. The #futureofwork is a national & global emergency! Check my #futureproof jobs series where I explore new mindsets, trends & new jobs/industries. How to robot-proof your jobs. It won’t be easy and you’ve got to start early!

Has Sophia got a Dark Plan for Bitcoin and Net Neutrality?

0

has-sophia-got-a-dark-plan-for-bitcoin-and-humans-16a70a50e766

2018-05-09

2018-05-09 23:46:34

https://medium.com/s/story/has-sophia-got-a-dark-plan-for-bitcoin-and-humans-16a70a50e766

false

562

null

Net Neutrality

net-neutrality

Net Neutrality

4,176

Ivonne Teoh

Bitcoin, Crypto, Social Media, Educator & Social Worker. Check out my series on LinkedIn about #AI #robotics #futureofwork #futureproof jobs.

8a3e5899ff24

ivonne.teoh

430

2,224

20,181,104

null

0

null

0

34a06daa33d2

2018-06-12

2018-06-12 22:22:16

2018-06-13

2018-06-13 18:06:18

4

false

en

2018-06-18

2018-06-18 13:16:30

1

16a72eb186d2

2.40566

20

2

1

A patent application on synthesizing accurate, engaging, contextually relevant, and personalized query responses

5

A closed-loop NLP query pre-processor and response synthesizer A patent application on synthesizing accurate, engaging, contextually relevant, and personalized query responses A closed-loop natural language query pre-processor and response synthesizer architecture accepts natural language queries and dynamically synthesizes query results. The query results may be in the form of data stories. The architecture identifies, selects, and composes candidate response elements into a coherent and meaningful query result. The architecture also implements an adaptable delivery mechanism that is responsive to connection bandwidth, query source preferences, query source characteristics, and other factors. Feedback from multiple sources adapts the architecture for handling subsequent queries The architecture implements technical solutions to many difficult technical problems in the field of automatically generating meaningful query responses given extensive and impossible to manually search data stores of potentially relevant information. A few examples of the technical solutions are summarized next. The architecture provides a personalization mechanism for answering questions, responsive, as examples, to: the role and perspective of the person asking the question; timing considerations; context; session history, including prior queries and responses, query and response history from others with similar characteristics to the querying entity, such as other enterprise engineers or managers; and other factors. The architecture may also identify explicitly and implicitly referenced entities in the input query and use the identified entities in its search for candidate response elements. The architecture also implements query prediction to determine, in advance, likely subsequent queries to follow, given a starting input query or sequence of input queries and contexts. The architecture understands which metrics, key performance indicators (KPIs), and other data are relevant to the substance of the input query, responsive to configurable ontologies and other models whose content provides a pre-defined context for the substance of the input query. For instance, the context may describe a particular enterprise, its markets, its products, workflows, metrics, and its enterprise activities. The architecture also identifies the type of question asked in the input query, and correlates the input query and candidate response elements with enterprise activities, targeting, planning, and other goals. The technical solutions in the architecture further identify the time frame of reference in the input query, its positioning within, e.g. a pre-defined fiscal year for the enterprise or competitor enterprises, and other timing data. The architecture responds to enterprise structural data, e.g., organizational structures, and enterprise dynamics to differentiate the query responses. The technical implementation of the architecture further implements a recommendation engine for suggesting intelligent actions following a session of input queries and query results. The recommendation engine provides the further benefit of encouraging additional interactive sessions via suggestions, questions, and data stories that follow any given query and response. Patent application

A closed-loop NLP query pre-processor and response synthesizer

150

a-closed-loop-nlp-query-pre-processor-and-response-synthesizer-16a72eb186d2

2018-06-21

2018-06-21 04:39:12

https://medium.com/s/story/a-closed-loop-nlp-query-pre-processor-and-response-synthesizer-16a72eb186d2

false

452

The blog for technology innovation. Follow us for novel product concepts, technology patents, fresh ideas and best practices for enabling innovation in your company. Original content only.

null

TheInnovationMachineBlog

null

The Innovation Machine

g.krasadakis@gmail.com

innovation-machine

INNOVATION,IDEAS,TECHNOLOGY,SOFTWARE,ARTIFICIAL INTELLIGENCE

innovation_mach

Ideas

ideas

Ideas

17,118

George Krasadakis

Product Architect; writing on Innovation and Product Development; Views, ideas and opinions are my own. https://www.linkedin.com/in/gkrasadakis/

f050c7452249

gkrasadakis

5,301

1,751

20,181,104

null

0

parameters = { 'C': [125000, 150000, 175000], 'kernel': ['rbf'], 'gamma': ['auto', 0.01, 0.1] } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') SVR(C=150000, cache_size=200, coef0=0.0, degree=3, epsilon=0.1, gamma=0.01, kernel='rbf', max_iter=-1, shrinking=True, tol=0.001, verbose=False) Best GridSearchCV R^2 Score : 0.860728192269 Train MAE : 8013.55062267 Train R^2 : 0.9278630562 Test R^2 : 0.907566762948 Test MAE : 14829.6005909 Best GridSearchCV R^2 Score : 0.0472278732524 Train MAE : 51504.2482383 Train R^2 : 0.0605744315867 Test R^2 : 0.100912362848 Test MAE : 50073.7048117 parameters = { 'criterion': ['mse', 'mae'], 'min_samples_leaf': [5, 10, 15, 20, 25], 'max_depth': [6, 9, 12, 15, 20], 'presort': [True], 'random_state': [1] } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') DecisionTreeRegressor(criterion='mae', max_depth=12, max_features=None, max_leaf_nodes=None, min_impurity_decrease=0.0, min_impurity_split=None, min_samples_leaf=20, min_samples_split=2, min_weight_fraction_leaf=0.0, presort=True, random_state=0, splitter='best') Best GridSearchCV Score : 0.750726929887 Train MAE : 20434.3076264 Train R^2 : 0.794288771776 Test R^2 : 0.817249933667 Test MAE : 22062.2174658 parameters = { 'n_estimators': [100], 'min_samples_leaf': [3, 5, 10, 20], 'max_depth': [9, 12, 15], 'n_jobs': [-1], 'random_state': [1] } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=12, max_features='auto', max_leaf_nodes=None, min_impurity_decrease=0.0, min_impurity_split=None, min_samples_leaf=5, min_samples_split=2, min_weight_fraction_leaf=0.0, n_estimators=100, n_jobs=-1, oob_score=True, random_state=1, verbose=0, warm_start=False) Best GridSearchCV Score : 0.824849224673 OOB Score : 0.828991092574 Train MAE : 11612.9739268 Train R^2 : 0.9258381402 Test R^2 : 0.896588642158 Test MAE : 16868.2702119 parameters = { 'n_estimators': [100], 'max_depth': [3, 6, 9], 'subsample': [0.8, 0.9, 1], 'reg_alpha': [0, 0.1, 0.3, 1], 'reg_lambda': [0, 1], 'nthread': [-1], 'seed': [1] } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') XGBRegressor(base_score=0.5, colsample_bylevel=1, colsample_bytree=1, gamma=0, learning_rate=0.1, max_delta_step=0, max_depth=9, min_child_weight=1, missing=None, n_estimators=100, nthread=-1, objective='reg:linear', reg_alpha=1, reg_lambda=1, scale_pos_weight=1, seed=1, silent=True, subsample=0.8) Best GridSearchCV Score : 0.870888887778 Train MAE : 1798.93234188 Train R^2 : 0.999086657521 Test R^2 : 0.920996938653 Test MAE : 14229.4014207 parameters = { 'n_estimators': [100], 'max_depth': [6, 9, 12], 'subsample': [0.8, 0.9, 1], 'reg_alpha': [1], 'reg_lambda': [1], 'gamma': [0, 0.1, 0.3, 1], 'nthread': [-1], 'seed': [1] } XGBRegressor(base_score=0.5, colsample_bylevel=1, colsample_bytree=1, gamma=0, learning_rate=0.1, max_delta_step=0, max_depth=9, min_child_weight=1, missing=None, n_estimators=100, nthread=-1, objective='reg:linear', reg_alpha=1, reg_lambda=1, scale_pos_weight=1, seed=1, silent=True, subsample=0.8) combine["OverallGrade"] = combine["OverallQual"] * combine["OverallCond"] combine["GarageGrade"] = combine["GarageQual"] * combine["GarageCond"] combine["ExterGrade"] = combine["ExterQual"] * combine["ExterCond"] combine["KitchenScore"] = combine["KitchenAbvGr"] * combine["KitchenQual"] combine["GarageScore"] = combine["GarageArea"] * combine["GarageQual"] combine["TotalBath"] = combine["BsmtFullBath"] + (0.5 * combine["BsmtHalfBath"]) + \ combine["FullBath"] + (0.5 * combine["HalfBath"]) combine["AllSF"] = combine["GrLivArea"] + combine["TotalBsmtSF"] combine["AllFlrsSF"] = combine["1stFlrSF"] + combine["2ndFlrSF"] combine["AllPorchSF"] = combine["OpenPorchSF"] + combine["EnclosedPorch"] + \ combine["3SsnPorch"] + combine["ScreenPorch"] OverallQual 0.819 AllSF 0.817 AllFlrsSF 0.729 GrLivArea 0.719 ExterQual 0.681 GarageCars 0.680 TotalBath 0.673 KitchenQual 0.667 GarageScore 0.657 combine["OverallQual-s2"] = combine["OverallQual"] ** 2 combine["OverallQual-s3"] = combine["OverallQual"] ** 3 combine["OverallQual-Sq"] = np.sqrt(combine["OverallQual"]) combine["AllSF-2"] = combine["AllSF"] ** 2 combine["AllSF-3"] = combine["AllSF"] ** 3 combine["AllSF-Sq"] = np.sqrt(combine["AllSF"]) combine["AllFlrsSF-2"] = combine["AllFlrsSF"] ** 2 combine["AllFlrsSF-3"] = combine["AllFlrsSF"] ** 3 combine["AllFlrsSF-Sq"] = np.sqrt(combine["AllFlrsSF"]) combine["GrLivArea-2"] = combine["GrLivArea"] ** 2 combine["GrLivArea-3"] = combine["GrLivArea"] ** 3 combine["GrLivArea-Sq"] = np.sqrt(combine["GrLivArea"]) combine["ExterQual-2"] = combine["ExterQual"] ** 2 combine["ExterQual-3"] = combine["ExterQual"] ** 3 combine["ExterQual-Sq"] = np.sqrt(combine["ExterQual"]) combine["GarageCars-2"] = combine["GarageCars"] ** 2 combine["GarageCars-3"] = combine["GarageCars"] ** 3 combine["GarageCars-Sq"] = np.sqrt(combine["GarageCars"]) combine["TotalBath-2"] = combine["TotalBath"] ** 2 combine["TotalBath-3"] = combine["TotalBath"] ** 3 combine["TotalBath-Sq"] = np.sqrt(combine["TotalBath"]) combine["KitchenQual-2"] = combine["KitchenQual"] ** 2 combine["KitchenQual-3"] = combine["KitchenQual"] ** 3 combine["KitchenQual-Sq"] = np.sqrt(combine["KitchenQual"]) combine["GarageScore-2"] = combine["GarageScore"] ** 2 combine["GarageScore-3"] = combine["GarageScore"] ** 3 combine["GarageScore-Sq"] = np.sqrt(combine["GarageScore"]) # make sure we need to drop 4 observations >>> df_train[‘Id’][df_train.GrLivArea > 4000] 523 524 691 692 1182 1183 1298 1299 # drop above observations >>> df_train = df_train[df_train.GrLivArea < 4000] skewness = combine[numerical_features].apply(lambda x: skew(x)) skewness = skewness[abs(skewness) > 0.5] skewed_features = skewness.index combine[skewed_features] = np.log1p(combine[skewed_features]) model = LassoCV( cv=10, max_iter=50000, n_jobs=-1, random_state=1 ) Best Alpha : 66.8843545179 Train MAE : 13838.2377563 Train R^2 : 0.925993418504 Test R^2 : 0.916634410474 Test MAE : 15469.2533307 Best Alpha : 0.6 Train MAE : 12925.4936661 Train R^2 : 0.935888198983 Test R^2 : 0.901017189141 Test MAE : 16575.9871503 parameters = { 'C': [225000, 250000, 275000, 300000], 'kernel': ['rbf'], 'gamma': ['auto', 0.01, 0.1], } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') SVR(C=275000, cache_size=200, coef0=0.0, degree=3, epsilon=0.1, gamma='auto', kernel='rbf', max_iter=-1, shrinking=True, tol=0.001, verbose=False) Best GridSearchCV R^2 Score : 0.906964160403 Train MAE : 5534.178792 Train R^2 : 0.965740040734 Test R^2 : 0.939173671075 Test MAE : 12831.3393298 parameters = { 'n_estimators': [100], 'max_depth': [3, 6, 9], 'subsample': [0.8, 0.9, 1], 'reg_alpha': [0, 0.1, 0.3, 1], 'reg_lambda': [0, 1], 'nthread': [-1], 'seed': [1] } grid_obj = GridSearchCV(model, parameters, cv=10, scoring='r2') XGBRegressor(base_score=0.5, colsample_bylevel=1, colsample_bytree=1, gamma=0, learning_rate=0.1, max_delta_step=0, max_depth=3, min_child_weight=1, missing=None, n_estimators=100, nthread=-1, objective='reg:linear', reg_alpha=1, reg_lambda=0, scale_pos_weight=1, seed=1, silent=True, subsample=0.8) Best GridSearchCV Score : 0.905190288583 Train MAE : 9744.92443218 Train R^2 : 0.970448412695 Test R^2 : 0.933626444213 Test MAE : 14021.5986899

36

null

2017-11-01

2017-11-01 03:54:50

2017-11-03

2017-11-03 05:53:56

2

false

en

2017-11-06

2017-11-06 06:10:36

16

16a78eec0df9

12.911635

1

0

This post is a continuation from my earlier post, here

3

Practicing Regression Techniques on House Prices Dataset-Part 2 This post is a continuation from my earlier post, here We have explored parametric algorithms in the last post. Lets continue to non-parametric algorithms. Training Models : SVM with RBF kernel You can check the complete code in training/TrainSVM.py SVM (Support Vector Machine) work for both classification and regression problem. The general objective of SVM is to draw a hyperplane (we can imagine it more or less like “line of the best fit”) that give the largest margin to the training observations (“Support Vector”) . Larger margin means the model will generalize better on unseen data, it will have larger bias and lower variance. We can control the margin by modifying C parameter (also known as soft margin). Increasing the value of C will increase the variance and lower the bias of the model One advantage of using SVM is its flexibility, thanks to the kernel trick. We could think kernel as a function that computes a similarity between a pair of samples. We will use the non-linear RBF kernel in this post. One of the most important parameter for RBF SVM is gamma. This parameter control the influence of training samples/observations. Large gamma will lead to tighter decision boundary, hence higher bias. The default parameter of C is 1, so I tried some values around 1 at first. However, GridSearchCV always choose the largest C in the input list. So, I keep increasing the value of C to increase the predictive performance of the SVM. We will use 72 features chosen by Lasso. Let’s try to optimize the C and gamma. This is the last parameter after increasing the value of C on each iteration : This is the best parameter chosen by GridSearchCV : And this is the result from the model above : It seems the SVM is working well. Large C (150000)and small gamma (0.01) means the model have high variance, so it’s more prone to overfitting. However, The gap between train and test set is reasonably small (approximately 0.02) hence there is no indication of overfitting. The R² result on the test score is also the highest so far. Just for comparison, this is the result I got on the first few iterations, where the C parameter is much smaller, combined with same value of gamma : Both train set and test set R² score is very low so this model is suffering from underfitting. Intuitively, we could think tuning the C and gamma parameter serve a similar purpose to tuning alpha parameter from L1 regularization earlier. It’s all about the bias-variance trade-off. Training Models : Decision Tree Regressor You can check the complete code in training/TrainDecTree.py As the name suggest, Decision Tree algorithm will “grow” a binary tree (“yes” or “no”) based on a series of “questions”. The “questions” are generated from the features (i.e. Is OverallQual < 5?). This algorithm will split the data based on the feature that will give the largest Information Gain. The Information Gain is computed by certain metrics, for example MSE and MAE. In practice, we need to “prune” the tree most of the time because the tree will grow very deep. Intuitively, we can think that the tree is asking too many specific “questions” hence it will fit the training set very well but failed to generalize to unseen data. There are 2 types of pruning : pre-pruning and post-pruning. Pre-pruning will stop growing the Decision Tree if certain condition (parameter) is met. On the other hand, Post-pruning will do the pruning after the tree is fully grown. Sklearn’s implementation of Decision Tree only support Pre-pruning at the time of this writing. We do pre-pruning by defining two conditions : max_depth is the maximum depth of the tree (duh!), exclude the root. min_samples_leaf is the minimum number of samples/observations in the leaf. Larger value of max_depth will increase the variance. On the other hand, smaller value of min_samples_leaf will increase the variance. Let’s take a look at the chosen parameters : The result is not bad, but not particularly good compared to the other models : One advantage of Decision Tree is we can print the tree, so the model don’t work like a “Black Box”. The printed tree show us that it actually has 10 depth (max_depth = 12), with 20 minimum samples at the rightmost leaf (min_samples_leaf=20). Training Models : Random Forest Regressor You can check the complete code in training/TrainRF.py Random Forest is an ensemble algorithm that combine multiple Decision Trees. In the case of regression problem, this algorithm will average the output (i.e. MSE/MAE) of multiple trees. Random Forest tend to build a more robust model that has a better generalization performance on unseen data and less prone to overfitting compared to a single Decision Tree. These are the simplified inner working of Random Forest : Choose random samples from training set with replacement (bootstrap parameter) Build a Decision Tree on the drawn sample, with a random subset of features (without replacement) at each node Repeat the 2 steps above Average the prediction defined by the criterion (MSE/MAE) Similar with Decision Tree, we will pre-prune each of the individual trees via max_depth and min_samples_leaf parameters : n_estimators is the number of trees in the forest. This is the best parameters chosen by GridSearchCV : The result is better than a single Decision Tree : Random Forest can be use for feature selection via feature_importances_ attribute. One advantage of using Random Forest is it will punish redundant features if similar feature has been chosen, hence it seems to be a good choice for this dataset. Training Models : XGBoost Regressor You can check the complete code in training/TrainXGB.py eXtreme Gradient Boosting is an advanced boosting algorithm that have a very good reputation on Kaggle. There is a saying : “if your predictive performance stuck, try XGBoost!”. Boosting is a method that uses an ensemble (group) of weak learners (typically Decision Tree). This method will focus on mistakes made by the weak learners, so it lets the weak learners to learn from their mistakes to improve the overall predictive performance overtime. The XGBoost package is not available (yet) on the sklearn library. In this post, we will try to do GridSearchCV on several parameters : n_estimators is the number of trees max_depth is the maximum depth allowed for a single Decision Tree subsample is the ratio of training sample randomly drawn from the observations for each tree, hence 1 mean each tree use all observations. reg_alpha is the L1 regularization strength reg_lambda is the L2 regularization strength This is the best parameter chosen by GridSearchCV : Let’s see this sophisticated model in action : It seems this model is famous for a reason. This is the highest R² we achieved so far in the test set. It seems to suffer a little overfitting though, the train set R² is almost perfect (0.999) and the gap between train & test set is significant : 1.7 k MAE vs 14.2k MAE. So I try to combine some of the best parameters from above and strengthen an additional parameter : gamma. This parameter might be able to reduce overfitting by specifying the minimum loss reduction required to make a split in each tree. Unfortunately, the GridSearchCV picked the same set of parameters as above : Hence it returns the exact same score as above. We will try to optimize this model again in later iterations (with the help of feature selection to reduce the overfitting). I hope the model will perform a little bit worse on the train set, but better on the test set. Second Iteration : Feature Engineering You can check the complete code in DataPrepration.py Our best R² score so far with basic features is 0.92 by XGBoost. SVM took the second place with 0.9 R² score. Let’s try to add some new features. First, we will create new features by combining several existing features : Then we will re-rank the feature importance based on the Pearson correlation coefficient like we did in the 1st part of this post (this time we include the new features) : Some of the new features we created seems to have a mild correlation with the target variable, such as TotalBath and GarageScore. After that, let’s engineer a new feature based on the top features above. We will generate a square root, 2* and 3* polynomial for each of them : We can try PolynomialFeatures as an alternative. This class will generate the polynomial and interaction features (optional). I tried to generate 2* polynomial of all features (reduced by Lasso) using this class and the Linear Regression model will suffer from overfitting because the model will become too complicated (hence too much variance). Second Iteration : Remove Outliers The next thing we are going to do is remove the outliers. Recall that the most expensive house is 755000$ and it has a significant gap with the one at the 75th percentile : 755000–129975 = 625025. We will check the distribution between the target variable with one of the most important feature : GrLivArea scatter_liv_area() We see that 2 observations at the bottom right are outliers : They have large area but sold at low price. In practice, we might need to confirm these data to other departments/teams. For example, there might be a human error when preparing this data. The most expensive one looks reasonable in this graph (the top right dot). However, I found out that the author of this dataset recommend to remove all observations with more than 4000 square feet. Hence, we will drop these 4 observations : Our final df_train will have 1455 observations with 238 features. Let’s try to fit our new dataset to some models and compare the result. I also decided to remove a single observation because it’s the only one with missing Electrical feature (not null but have “NA” value). I found an interesting tips at this point : Log transform the skewed numerical features (at certain threshold) to lessen impact of outliers. In this project, we will log transform the features which skewness > 0.5 : Second Iteration : Lasso Let’s do L1 regularization again, with our new features : The result : This iteration have better R² score both on train & test set, with less regularization strength (Alpha). The improvement is not significant (0.018 difference in R² test score) but it’s a good indicator that our new feature is working well. As discussed in part 1, people tend to use smaller values of alpha (The default value of alpha is 1). So, I try to do GridSearchCV with much smaller list of alphas (0.0001 to 1), and the model return this result : The result shows that a large difference in alpha (66.8 vs 0.6) don’t significantly affect the result for this dataset (0.015 difference in R² test score). The model did a great job finding the alpha automatically. Second Iteration : SVM The second iteration of SVM, with filtered features (by Lasso earlier) : Parameters chosen by GridSearchCV : The chosen C is larger than the previous iteration, but with smaller gamma (“auto” means 1/n_features = 1/238 = 0.0042). Let’s check the result : The R² test score is significantly improved by approximately 0.03 point. Recall that the best possible score for R² is 1 and we already achieved approximately 0.9 score earlier. This is the best R² score on the test set so far. Last but not least, let’s fit the new training data to XGBoost model. Recall that this model is suffered with overfitting, so lets try to use filtered features from Lasso to reduce it : This is the chosen combination of parameters : The maximum depth of the tree have go down from 9 to 3. This is a great news for us because the model suffered from overfitting in the 1st iteration (large gap between train and test set). The result confirmed that the model will generalize better. The train R² score have gone down (from almost perfect score of 0.999) and the test R² score have gone up a little (0.01). Conclusion First things first, let’s summarize what we have explored : The importance of feature engineering and outliers removal on model performance. Checked. Creating new features and removing 4 outliers increased the performance of Lasso, SVM & XGBoost. The benefit of regularization (L1, L2 and ElasticNet) to reduce overfitting. Checked. Regularization is a no-brainer for this dataset (and for most data, I presume). We compared the strength of L1 & L2 regularization on this dataset and found out L1 regularization is better (more sparse data). I think this is related to the OneHotEncoding method we used data prepration, hence we have a lot of binary features that can be reduced. The benefit of feature selection (by Lasso/L1) to reduce overfitting. Checked. Reducing the number of features we passed on the Linear Regression model succesfully prevent the overfitting. Note that while the best parameters and result I showed here is real (you can run the source code in your machine), the process I wrote in this post is a simplification of what I did when practicing with this dataset. For example, I did more than 2 iterations because I tried feature engineering and outlier removal in different iteration. There are many ways we could try to improve the score, for example : Try different method to handle missing data Try different method of feature selection Try more parameters in GridSearchCV (yes, I’m talking to you XGBoost) And so on. Even though I said that the main goal of this project is to practice several theories of machine learning, I do pay attention to the score instinctively. The best R² score of the test set we got so far is 0.939 from SVM. This project ranked 669 of 2111 in Kaggle, with 0.12296 root mean squared logarithmic error at the time of this writing. The leaderboard is very competitive, hence a little improvement might increase your rank significantly. Thank you for reading this article. Please kindly post your suggestion in the comment, and don’t hestitae to point out my mistake(s) on this post, I will update this post to correct them. References : Documentation scikit-learn: machine learning in Python - scikit-learn 0.19.1 documentation Edit descriptionscikit-learn.org https://www.packtpub.com/big-data-and-business-intelligence/python-machine-learning-second-edition https://www.analyticsvidhya.com/blog/2015/06/tuning-random-forest-model Understanding Support Vector Machine algorithm from examples (along with code) Note: This article was originally published on Oct 6th, 2015 and updated on Sept 13th, 2017 Introduction Mastering…www.analyticsvidhya.com https://www.kaggle.com/pmarcelino/house-prices-advanced-regression-techniques/comprehensive-data-exploration-with-python What is the influence of C in SVMs with linear kernel? In a SVM you are searching for two things: a hyperplane with the largest minimum margin, and a hyperplane that…stats.stackexchange.com Decision Trees: How do you prune a CART? Answer (1 of 2): There are generally two methods for pruning trees: pre-pruning and post-pruning. Pre-pruning is going…www.quora.com https://www.analyticsvidhya.com/blog/2016/03/complete-guide-parameter-tuning-xgboost-with-codes-python/ A study on Regression applied to the Ames dataset Using data from House Prices: Advanced Regression Techniqueswww.kaggle.com http://xgboost.readthedocs.io/en/latest/python/python_api.html#module-xgboost.sklearn https://www.analyticsvidhya.com/blog/2016/01/complete-tutorial-ridge-lasso-regression-python

Practicing Regression Techniques on House Prices Dataset-Part 2

50

practicing-regression-techniques-on-house-prices-dataset-part2-16a78eec0df9

2018-03-22

2018-03-22 02:42:36

https://medium.com/s/story/practicing-regression-techniques-on-house-prices-dataset-part2-16a78eec0df9

false

3,320

null

Machine Learning

machine-learning

Machine Learning

51,320

Tama Handika

null

582265600adc

blazetamareborn

7

15

20,181,104

null

0

null

0

null

2017-09-14

2017-09-14 18:26:48

2017-09-14

2017-09-14 18:40:44

1

false

en

2017-09-15

2017-09-15 12:06:19

11

16a7a2538e5b

6.724528

3

0

We live in the most advanced technical era in history, at least according to those of us not sporting tin foil hats. Nothing has ever…

5

10 Killer Robots That Exist already We live in the most advanced technical era in history, at least according to those of us not sporting tin foil hats. Nothing has ever travelled faster, climbed higher or cost more. We have never known as much about our world as we do now. This is the Google age and the future has never seemed closer. Within our lifetime robots and AI (artificial Intelligence) will have have a profound effect on the way we live. For good or for ill we can’t say, but some thinkers have expressed their concern. Notable minds like Elon Musk and Professor Stephen Hawking have spoken out over the ascension of machine thinking. While the debate rages, it might be worth mentioning that the causalities have already begun. The first To Die. The dubious accolade of being the first human killed by a robot, belongs to Robert Williams. Williams, 25, worked on the assembly line at Ford Motors in Flat rock Michigan. The accident happened on Jan 25th 1979, as both he and the huge machine attempted to get parts from a shared store. Robert William’s family won $10 million in damages after the incident, which a jury blamed on a lack of safety protocols. It would be decades before a robot struck again. Jan. 25, 1979: Robot Kills Human 1979: A 25-year-old Ford Motor assembly line worker is killed on the job in a Flat Rock, Michigan, casting plant.1 It's…www.wired.com Anti-personnel gun 30 years after Robert William’s death, 9 soldiers died when a computerised anti-aircraft Gun went berserk. The tragic accident happened at the Army Combat Training Centre at Lohatlha, South Africa. The soldiers tried in vain to bring the gun under control. By the time it had emptied its magazine, the Swiss/German Oerlikon 35mm anti-aircraft gun had killed 9 people and injured 15 more. Numerous theories surround the tragic accident, but a combination of mechanical failure and bad luck may both have played a part. The gun’s manufacturers later reported that although the gun can fire remotely, it does not feature an ‘automatic’ firing system. BBC NEWS | Africa | SA soldiers die during training The army, navy and air force take part in the annual training Nine South African soldiers have been killed during a…news.bbc.co.uk The Bomb disposal killer. Since their inception in 1972, bomb disposal robots have saved lives all over the world. Their shape, size and the way they work may differ, but they all serve the same function, which is to detect, defuse or destroy explosive devices. Everything changed on the 7th of July 2016 when Dallas Police strapped an explosive device to one of their bomb disposal robots. The plan was to use the robot to ellimate a shooter. The suspect, Micah Xavier Johnson, had opened fire on police officers during an otherwise peaceful protest. By the time police decided to use the robot, he’d shot and killed 12 people. Once in place, the authorities detonated the explosives killing the suspect instantly. The incident set a precedent as the first time a US police force had used a remote device to kill a suspect. When Police Use Robots to Kill People The Dallas Police Department's deployment of a machine to take down a suspect raises uncomfortable questions. The…www.bloomberg.com Chainsaw drone Innocent until proven guilty, that’s what they say, but then they have not had to deal a chainsaw wielding drone before. Nicknamed the Killer Drone, this flying monstrosity makes this list in name alone. It is the creation of two Finnish friends who conceived the device to trim icicles from their roof. Evil masterminds the world over are now trying to emulate their idea by hooking dangerous utensils to the quad-copters they got last Christmas from their kooky uncle. How long before this particular one takes someone’s arm off is anyone’s guess, but we don’t imagine it will take long. http://uk.businessinsider.com/guys-from-finland-put-a-chainsaw-on-a-drone-2016-4?r=US&IR=T First they’ll take your jobs… then they’ll take your lives… Robert Williams may have been the first human killed by a manufacturing robot, but he was not the last. In 2015, a robot crushed to death, the 22 year old technician who was attempting to repair it. The incident happened at Volkswagens Kassle plant. Reports suggest the robot malfunctioned as the technician was working on it inside its safety cage. The investigation is ongoing. Robot Kills Man at Volkswagen Plant A robot crushed a worker at a Volkswagen production plant in Germany, the company said Wednesday. A 22-year-old man was…time.com Auto-pilot We may never understand what happened to Malaysia Airways flight Air MH370m, but some experts think that the autopilot system played a part. A 55-page report, published by the Australian Transport Safety Board, suggests the passengers may have suffocated as the plane’s automatic pilot system coasted it towards its doom. Air travel is still the safest way to fly and the authorities are very keen to uphold that record. Therefore, it is hard to find concrete evidence of any error, let alone computer error. One thing is certain however, safety records have not improved in line with the rise of automated systems. MH370 passengers likely suffocated as plane coasted on autopilot, Australia says The new analysis comes more than 100 days after the Boeing 777, carrying 239 passengers and crew, disappeared on March…www.telegraph.co.uk Driver’s choice The new Mercedes self drive cars are going to be a killer range, quite literally. With more and more manufacturers looking at autonomous vehicles, it was inevitable that the question of how their vehicles behave in critical situations would arise. At some point in a car’s lifespan, there is a chance it will have to decide how to handle a no-win scenario. German auto giant Mercedes Benz have taken the first step. Given the choice between saving a pedestrian or its passenger, the self drive Mercedes will protect those on board. This controversial code means that if forced to make a choice, the car will choose to kill pedestrians rather than risk the lives of its passengers. Ultimately, thanks to the car’s brain and its superior driving skills, the risk of such an incident is far less likely. So the system will save more lives than it endangers. Mercedes' Self-Driving Cars Would Save Passengers, Not Bystanders In comments published last week by Car and Driver , Mercedes-Benz executive Christoph von Hugo said that the carmaker's…fortune.com The Professionals When you hear the phrase ‘drone strike’, there is a very good chance the story involves a Reaper Drone. These little fighter planes are the go-to device for those looking to deliver death from above. Although the rise of armoured drones has not been short of controversy. In 2001 in the first ever US drone strike, the little aircraft, which was being piloted from a swivel chair in an office on the other side of the world, took and missed the first and only chance to kill the world’s most wanted man, Mullah Omar. Omar, who at the time was the leader of the Taliban and number one on the US most wanted list, walked away from the botched assassination attempt. The incident sparked 14 years of feuding between US intelligence agencies. Predator tail-fin number 3034 now hangs immortalised on the ceiling of the Smithsonian Air and Space Museum in Washington, D.C. The Story Behind America's First Drone Strike "Who the fuck did that?" The words greeting the first-ever combat strike by a remotely piloted aircraft were uttered…www.theatlantic.com Robots of mass destruction Robots with chainsaw or machine guns are yesterday’s news and nowhere near scary enough, according to Russia anyway. No, for proper slaughter you need a robot with a bit more fire power. It’s true, as the rest of the civilised world is struggling with moral implications of drone strikes, Russia has developed one which carries massive nuclear warheads. This new breed of nuclear powered unmanned subs will be able to stay underwater for months and each one can carry a payload capable of wiping out an area the size of England. Russia tests devastating underwater nuclear drone Russia has carried out a test of a revolutionary unmanned nuclear submarine, according to US intelligence sources…www.independent.co.uk The Clever copter. Using a modified drone that you could pick up at any electronics store, researchers at Camp Edward in the US created an autonomous hunter killer. The clever device is capable of identifying and tracking a number of insurgents. The on-board AI had little trouble probing a mock Middle Eastern village and locking on to a number of individuals carrying AK47s. Their experiment has come a long way from their cold war days. Back then the US became obsessed with building an AI capable of locating Soviet tanks. They spent months feeding it images of Soviet hardware, but in the end their experiment had failed. What they had inadvertently built was an AI only capable of pointing out if the tanks were set against a snowy backdrop. The Pentagon's 'Terminator Conundrum': Robots That Could Kill on Their Own CAMP EDWARDS, Mass. - The small drone, with its six whirring rotors, swept past the replica of a Middle Eastern village…www.nytimes.com Bonus round As our list draws near to completion, we are once again reminded of the potential threats posed by robotic devices, AI and drones. Already this year, a report of a new incident appeared on our radar. While dealing with an aggressor, the US Air-force used a $3.2 Million dollar Patriot Missile to shoot down an Amazon $200 Drone. The General in charge of the operation highlighted his concern over the economic implications, stating that if he were the enemy he would buy as many $200 drones as he could afford. Small drone 'worth $200' shot down by Patriot missile worth $3m, says US general A small quadcopter drone was shot down with a Patriot missile, which usually cost around $3m (£2.5m), a US general has…www.independent.co.uk Like it or not, we have just begun to enter the robotic era and incidents involving devices like those above is set to rise. Whether we choose to heed the warnings of Elon Musk, Stephen Hawkings or Sarah Connor is a story for another day, but for now we have to stick to the program.

10 Killer Robots That Exist already

26

10-killer-robots-that-exist-already-16a7a2538e5b

2018-03-23

2018-03-23 12:01:49

https://medium.com/s/story/10-killer-robots-that-exist-already-16a7a2538e5b

false

1,729

null

Drones

drones

Drones

9,040

Gordon Vader

null

6e6ad1f17971

gordonvader

40

42

20,181,104

null

0

null

0

null

2018-08-12

2018-08-12 13:15:59

2018-08-12

2018-08-12 14:28:06

0

false

en

2018-08-12

2018-08-12 14:28:06

0

16a842b7e242

2.550943

0

This article reminds me of my own experience of statistics education. Lately I have been thinking a lot about what I lack of, and what I…

1

Thoughts on Roger Peng’s ‘Thoughts on David Donoho’s Fifty Years of Data Science’, and my own statistical education This article reminds me of my own experience of statistics education. Lately I have been thinking a lot about what I lack of, and what I panic about. There are too many things that I haven’t learnt from a statistics master course. Right now I shouldn’t complain about it since it’s already two years ago (which means by now it’s MY responsibility if I don’t know some topics), but I do feel it’s necessary to reflect on what could have been done better. Methodology on studying I had been a rather mediocre student in my past years. It is easy to be fooled by what university you go to or what is the ranking of your scores. A truly good student is someone that is genuinely interested in what she’s learning, which I have failed. In Chinese we would say I haven’t learnt the ‘essence’ of the knowledge, and that is far from true studying. Well, time and experience should help with one thing or two. There are things that I can’t understand without sufficient prerequisite knowledge. In addition, it is good practice that I go back to the basic theories after some years’ practice of statistics. For example, we all learnt Central Limit Theorem, but at the time I didn’t do a simulation to see how it really is, nor have I seen sufficient examples of its application. On the other hand, I have done a t-test many times, but I can’t explain the theory why it would work. These are two examples where I have failed to understand the knowledge or skills I have acquired. The good thing is, in the middle of my PhD I have realised that I lack the understanding of basic concepts, which has become an obstacle for me to go deeper and further of my work. There is no other way around it, don’t blame others for not teaching you, just go back to the old text books and study. As a matter of fact, once I have understood those concepts, the topic of statistics becomes less vague and boring. That’s exactly why I should do it. The structure itself In the article of Donoho, 6 divisions makes modern data science data exploration, transformation, computing, modeling, visulization, science of data science According to Roger Peng, we statisticians do all these things, but most of them are not taught at school, even at a graduate level. In my own experience, we do modeling and modeling only. Computing (at least in my experience) as an issue and direction to investigate has not been stressed enough, not in a statistics course, while the others are assumed to be figured out by students themselves. The reason is that Teaching some areas of the Greater Data Science can be difficult and inefficient — take the teaching of data cleaning and generalized linear models for example. and our field’s traditional bias towards the use of mathematics as the principal tool for analysis, and much of the interesting formal work being done in data cleaning and transformation makes use of computer science and software engineering. In my opinion, the way we were taught statistics is not intuitive to most people who had little idea what it is. As a student who didn’t do sufficient statistics before entering a intensive statistics course, I had experienced a suffering highly mathematical nightmare which didn’t have an obvious purpose. I can’t stress more the importance of time and experience in grasping the nature of what we do. As a conclusion, I admit that statistics is interesting (once I’ve past the initial suffering period where I haven’t seen enough models) and very challenging, especially now. In addition to the traditional models and approaches, it’s particularly important that we keep up to date with the new advances not only in methodology, but also in the applications and tools that connect them. They all come with experience, and years’ hard work and continuous thinking.

Thoughts on Roger Peng’s ‘Thoughts on David Donoho’s Fifty Years of Data Science’, and my own…

0

thoughts-on-roger-pengs-thoughts-on-david-donoho-s-fifty-years-of-data-science-and-my-own-16a842b7e242

2018-08-12

2018-08-12 14:28:06

https://medium.com/s/story/thoughts-on-roger-pengs-thoughts-on-david-donoho-s-fifty-years-of-data-science-and-my-own-16a842b7e242

false

676

null

Data Science

data-science

Data Science

33,617

StatApprentice

My own blog recording my thoughts in learning statistics and machine learning

23bffe3368f4

AndreaChiZhang

3

10

20,181,104

null

0

null

0

null

2017-12-02

2017-12-02 04:16:41

2017-12-06

2017-12-06 18:05:09

2

false

en

2017-12-06

2017-12-06 18:05:09

3

16a9443776d4

2.83239

8

0

If you knew that AI would destroy mankind in 150 years, would you still support it?

5

Want to save the world from AI? Answer these 3 questions. If you knew that AI would destroy mankind in 150 years, would you still support it? This question was posed at the “What is AI and How Can We Keep It From Harming Humanity?” event recently hosted by Tech2025 in Brooklyn, NY, to a few people who lingered afterward. This innovative startup is creating a community “eager to learn what [. . .]disruptive technology is, how it will change their businesses, and what they can do to prepare for the future.” Several AI companies estimate their tech will be ready for implementation in 8–10 years, or around the year 2025. After a few seconds of thinking about the end of mankind, I responded with an adamant, Yes! Since homo sapiens have not and will not always inhabit Earth, destruction by AI isn’t that surprising of a conclusion. Perhaps we deserve this fate because we are destroying our natural habitat through climate change. Another attendee chimed in with a tentative yes, confessing the selfish logic that he will already be dead in 150 years, so who cares? YOLO! Pros and Cons The potential advantages of AI have been eloquently discussed elsewhere, particularly the advancement of mankind through technological and scientific/healthcare achievements. Reversing climate change, assisting archaeological discovery, and historical preservation and analysis are just a few of the potential benefits of AI. The imaginations of filmmakers, screenwriters, and authors have immortalized the disadvantages of AI — namely the glorious doom of all humanity. Theoretical Background There is an innate force that drives all species on Earth to go forth and multiply. This keeps the species alive. If this is valid, it explains part of why we humans are terrified of being decimated by robots. The idea of one’s grandkids and great-grandkids being mercilessly destroyed en masse by AI robots is quite disturbing. Yet, this idea seems too abstract of a concept to inspire the kind of fear that breeds action. After all, this same concept doesn’t prompt us to do 100% of what we could to halt climate change (i.e. eat 100% plant-based, stop using all non-recyclable plastics, recycle our gasoline cars for parts and buy bikes or Teslas, etc). Existential Questions 1. If self-preservation of a species is an inherent drive of its members, then is this why we are developing AI in the first place? To help us be bigger, faster, smarter and generally better humans? 2. Is this self-preservation ‘instinct’ a feature of consciousness? Meaning if (or when) AI becomes conscious, will it become aware of its superior intelligence? Will it perceive that our threatened, fragile egos (#notmypresident) may threaten their existence? 3. Will AI software destroy us because of its instinctual attempt to remain in existence? This is not a pointless game of logic; the reasons why do matter. Our attempt at self-preservation through AI could be the first step on the path to our destruction, but this can be avoided. We could learn to coexist with a species whom we cannot exploit. Based on the world today and the last 150 years, it’s not looking good. What’s the takeaway? One, if the robots kill us off, we will probably deserve it. Two, although this event is likely a long way off, it’s crucial to be mindful of and intentional about the AI algorithms currently in development. This means we need to create algorithms that are free of racial, gender, sexual, and special bias. This means we need diversity in IT. We need diversity at AI startups, from engineers to the C-suite. We need to call out each other’s implicit bias and privilege. Three, we need to keep the conversation going. Blind fear of an unknown future is ignorant. We don’t have to be software engineers to opt-in to guide the tech that will determine our fate.

Want to save the world from AI? Answer these 3 questions.

59

want-to-save-the-world-from-ai-answer-these-3-questions-16a9443776d4

2018-04-06

2018-04-06 10:29:14

https://medium.com/s/story/want-to-save-the-world-from-ai-answer-these-3-questions-16a9443776d4

false

649

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Francine Mends

A doctor/dreamer writing about Big Life Ideas, Biotech/Healthcare, and Travel from a philosophical perspective. www.onevolving.com. IG: Onevolving.

fc5347cab976

francine.mends

18

27

20,181,104

null

0

null

0

null

2017-11-18

2017-11-18 04:44:37

2017-11-18

2017-11-18 13:00:57

1

false

en

2017-11-18

2017-11-18 13:00:57

15

16a9e4007b30

10.037736

0

If you have found yourself exasperated at the death of your Tamagochi pet, wearing thin the tracks on your hit clips, or even mortified at…

5

One Lite, Network If you have found yourself exasperated at the death of your Tamagochi pet, wearing thin the tracks on your hit clips, or even mortified at the evolution of board games to video channels, there is hope for the future of all physical and digital collaboration. From the evolution of games in the 80’s with launch of Atari to the technology behind the oven, yes, the oven, there is a development of such mechanics in which unto these days has seemed rather obtuse. Far beyond the means of playing, making, or using any of these challenging devices, like the oven, there is hope as well. What we come to understand about computing is thatas life has evolved the hums, sweeps, and beeps that once filled rooms continue to move towards evolution in which we can fit in our hands. Imagining such, we can consider blowing out cartridges a primitive act. If you like myself have stared at something, wondering, if it could be better in the sense of its approach, solution, or experience, raise your hand. Yes, it seems a few of many find at least one plausible entry in this journal, of would solve. I imagine these such instances, visually, as those that rig rubber bands or tape around lego formations or, even better wooden constructs that in which yield a temporary salivation. Far from computing matters, I imagine such as the early entrepreneur selling shower curtain rings, or traveling around with their hand crafted devices, though, we now look to those wide eyed creators to inform us on the solutions of the future. What we find at the end of such ventures,, of those proclivitous enough, is something in which we find rather brilliant. Much as a kite with a key, it takes lightning to strike twice for our sometimes significance to actually occur. In the matter these inventions iterate over time to involved the structures of probability and meet somewhere up a margin of means. What we relish through these outcomes is a connectivity to devising the improbable, defying the metrics we apply to ourselves, and building upon the new ground in which we scorched only minutes before. Redefine Tangible An act in which we create something that before a moment ago did not exist, was said to be untrue, and frequently is not aligned with our current vision. As solutions and concepts are masqueraded across screens we once doubted, these matriculations continue to collect. Where this leads, is imagining how each small step, fragments of technologies, become leaps in applicatory means. Knowing in which these pieces impact, where, and visualizing their occurrence is the practical sense of intelligence. Looking, at our journals of solve, flipping through pages, and finding the moment in which you can pinpoint a new step in variance of cognition, becomes this graceful bound. As mentioned, many of the applications of today where merely subject of intrigue once before, straining our minds to see that in which we could only imagine, driving it into fruition, with more than just thought, but some elbow grease. I suppose, in the technology sense, you may want to wipe down and stay grounded for these ventures, though seeing as these means apply across a stream of practicality and deep machine probability, it might get a little messy. The Flow The evolution of TensorFlow and its increasing viability for productive machine learning practices only continues to scale, or shrink, depending on how you consider such. The evolution of TensorFlow mobile, which already has positive attributions for lightweight processing on mobile devices, will enable even more form factor creations to dive into the powers of machine intelligence. When we align the sense of machine intelligence through micro means, the real world applications and benefits continue to factor, in this sense, where it can fit, how much less it takes, energy and computationally, the more that can be solved and connected through these interactions. If, for a moment, we align the technologies of machine intelligence, neural networks, natural language processing, and IoT devices as the means of known, and apply the adapted contributions of shrinking components through means such as raspberry pi, the arduino open source kits, or the Intel NUC, we can continue our journey to correlate how in which these devices will not only improve our experience, but also the understanding of the connected world/people/families and empower more and more to find means in which the evolution of this technology will increase their digital correspondence and argue in a way embetterment of usage. We can pear into the existing solutions of diverse creation through this means in which Massmo Banzi discusses, with the developments in the open-source community with and around the arudino. As he states we can understanding the unfolding era of inventors and creators, whom take these emmerging technologes and apply them, as being part of the “makers-movement”, something in which many have been waiting for a name. We can understand more about this shift from Limor ‘Ladyada’ Fried who Massimo refers to as a hero of this movement, along with that of open-source hardware. In the interview Limor, compares the maker movement itself to the homebrew computer club/s that generated much of the technology we use today, in a well put terms… “Whatever makers are doing for fun now, is what everybody is going to be doing in 10 to 15 years.” — Limor ‘Ladyada’ Fried The empowerment of these concepts is that we can make impacts in the moments ahead without asking permission from companies to innovate, taking our own demand as initiative in practical application. Separate from the contraptions mentioned early, these frameworks apply across boards that are manufactured from even more dynamic approaches. With the schematics available, users/people/humans, can create their own means in which to fit these architectures, or retrofit current models across a spectrum of experiences. In this sense Massimo mentions that “hardware becomes a piece of culture”, a way in which the availability of intellect, and practical application of computation, empowers society to solve for individual needs. A People Process In the existing community of hardware development and machine learning tandem there have been intital steps to validate the use of sensors input across these devices to incorporate the understands of human interaction. As an accelerometer that gathers data from say your phone, may come into play, there are existing examples of how such could play out and be built on to empower our understanding of these experiences. At Arduino Day 2016 David Mellis points to research as part of these sensory inputs, how we can begin to define interactions, view them, and interpret the data for better understanding. It is important to note his mention of available algorithms and tools to user across these hardware components that connect and interpret the inputs as well as clarify their input through pattern. These are the challenges many face when approaching this technology, outside of these applications in particular, but has a growing awareness in which to empower individuals to understand the computation behind such creations. Along the path of the makers-movement, it is important to plot these solutions as steps to a broader approach/understanding/equity of information in solving needs in which can become independently developed solutions for many. Tool-kits alike are becoming prominent sources for achieving these applications and streamlining the application of machine based intelligence across platforms/devices/solutions. When we look at the application of such in a use case based approach, it allows us to adapt the thought and apply it to other solutions. Many of the present examples to scale of say the raspberry pi with TensorFlow exist through image classification, capturing the data input and processing it in the cloud, validating a/few inputs/s through a trained neural network. These practices rely on Convolutional Neural Networks or CNN’s that we discussed in a previous outlook referencing neural video. The classification component of these sensors works in tandem with this framework to fit the feed-forward approach of machine articulation. As we look towards integration we need to understand the various frameworks as their application, their viability in practice, and the approach in which reveals approximate patterns in solving. Comparatively many of these frameworks are working towards cross application, as the needs are specifically sought for various solutions. Learned Insight You may be wondering in which means I will approach, now ‘classic’ technologies, board games, or ovens, ah ovens. This is where the application of these aforementioned micro-micro-computers, if you will, come into scope. In many a sense, these technologies are being tested and researched to power many things from autonomous vehicles to drones, as their formfactor feeds into a viability that we are still seeking to apply on many scales. The inference of my three examples as such are means in which we discover conditional logic that is relative to how a human thinks. As automation and IoT devices develop, improving the machine intelligence, or launching it past our own, means first discovering what, then how, and in which means. Visualize the situation in which we first interact with a human, open a board game, or turn on the oven to cook/bake. In all of these settings we have an initial experience that correlates how in which we base our next move. There are a standing set of principles we learn as we progress, but the first moment inwhich we initiate that neural pathway, we begin to map how each unfolds. Inherently, our brain then takes inputs to create persisting information in which we develop better practices from each interaction. We can all consider instances in which we have embarrassed ourselves socially, overlooked an advantage, or burned our pending, wonderful creations. All of these means applied, fulfill a syntax that allows our brain to reference upon the next interaction. Combining Convolutional Neural Network (CNN)’s and Recurrent Neural Networks (RNN)’s, machines can begin to process labeled data to understand patterns in action. This as part of its connection to IoT sensors and cameras can correlate these visual patterns to begin to understand our intention and emotion. In tandem the framework and (RNN)’s apply a Long Short Term Memory (LTSM) approach in historically layering data — in which we are not only training the recognition, but storing it as a model for persisting reference. As does the RNN add logic to machine intelligence perception in high level means, but can additionally add attention. In the interactions mentioned, of relational rapport, game theory, and cooking/baking, we can now build correlation between not only our initial interactions, storing and building from them, but developing a correspondence of, in a sense empathy. In the near future, this may mean that your oven might just save an optimal time, encourage you to choose a preset input, or an interactive game may stop to explain where you could have gone, all based on your reaction. The idea of attention in machine intelligence empowers these inputs in the sense that it will allow for further collaboration, devising intent, and filtering/gating the selected inputs in which are chosen through the matters of the RNN’s focus. The impact of RNN’s can not only help refine inputs, but help develop and train new models for existing neural networks, and their accuracy through reinforcement training. Much as experiences develop our interactions overtime, the correlation of RNN’s application, applies a similar model to development as well. Using Machine Learning to Explore Neural Network Architecture At Google, we have successfully applied deep learning models to many applications, from image recognition to speech…research.googleblog.com Aligning these neural paths to our own, helps us further understanding how in which we can collaboratively develop both senses, in a means, to refining how we understand ourselves reflectively, and how to develop machines with human-like thought. Connecting Thought As we come to understand the current applications of these frameworks, their increasing depths, and valuable interactions we can build upon which to begin plotting the integration of hardware, IoT devices, and virtual interfaces that connect them. We have collectively begun understanding these sensations in the means of a virtual assistant, as we learn to connect these correlations, we can imagine such as the new interface. Beyond that of the current solutions, which are quite helpful, as applications, there is a new frontier upon us. Aligning, computer vision, machine intelligence, deep learning, and an array of IoT devices we can see the shift from simple interactions and inputs to a convergence of definitive means. A voice in our ear suchas in “Her”, an interactive life assistant, that connects our home, to our devices, to our car, and continuously monitors all of such interactions to employ assistance where sensed. As the inputs to these interactions becomes increasingly connected and simplified through means such as natural language processing of text or even voice, the connectivity of machine intelligence will begin to learn and develop around our choices and interactions, enhancing or iterating for many of which in which we seek in our journals of solve. Take for example, this concept of interaction, as an iterative step closer to the future, an example from an article called “Voices of the Future” on bots, in this example, in reference to Alexa’s increased efficiency in response and what in a sense, that feature focus proves. “ The possibility to talk to a device using no screen and without unnecessary pauses as if it were a person proved to be a critically necessary feature for a successful interaction experience.” — Alex Galert Each step is a leap closer to unifying intelligence's, in many ways these much anticipated and unfolding occurrences are exciting, allowing for more innovation and thought, through removing many an unnecessary action throughout the day. With the shift and development of these platforms, the collaboration of devices and people,, society will begin to enter a new era of interaction, the means in which we think, solve, and contribute grow evolvingly closer. Understanding these contributions, the correlation of the human mind, to a computer, and how these occurrences have been dreamt, written, and designed far before their initial conceptions is only a hint at how optimistic we can look to these solutions to continue to improve the lives of the world. With the increase of open source platforms, technologies, and devised solutions, these applications in thought and action are continuing to unite and solve for everyone. We must continue to pursue curiosity, test theory through new mediums, and apply thought to the nature of the experiences we create. As the refinement of these technologies occur, it is our duty to venture into them to define new occurrences and build from their opportunity a better experience for all. Sources Mentioned: Computing History Anecdote (-), TensorFlow Lite Announcement (-), “How Arduino is open-sourcing imagination” TED Talk from Massimo Banzi(-), Meet the Makers: Limor Fried (-), “Makers and Machine Learning” (-), “Machine Learning for the Maker Community” by David Mellis (-), raspberry pi with TensorFlow examples by Rikki Endsley (-), Convolutional Neural Networks (-), “Two-Stream RNN/CNN for action recognition in 3D Videos” from Patrick van der Smagt(-), Long Term Short Term memory networks by Christopher Olah (-), “Attention and Augmented Reccurent Neural Networks” by Chris Olah and Shan Carter(-), “Using Machine Learning to Explore Neural Network Architecture” by Quoc Lee and Barrett Zoph(-), “Voices of the Future” by Alex Galert(-) Stock Photo Attribution: Milky Way photo by Billy Huynh (@billy_huy) on Unsplash Download this free HD photo of star, night sky, celestial body and milky way by Billy Huynh (@billy_huy)unsplash.com

One Lite, Network

0

one-lite-network-16a9e4007b30

2018-06-16

2018-06-16 13:05:08

https://medium.com/s/story/one-lite-network-16a9e4007b30

false

2,607

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Joe Alongi

People and Experiences.

cd5d5e9a80e4

jdalongi

17

217

20,181,104

null

0

null

0

29989b59e37d

2018-07-16

2018-07-16 11:41:47

2018-08-02

2018-08-02 10:32:26

1

false

en

2018-08-07

2018-08-07 08:15:02

3

16ab3b9c797b

1.939623

11

1

0

This post aims to provide an example of how to train an ML Model using Spark, serialise the model using MLeap and serving the model using…

3

Machine Learning: Training and Serving This post aims to provide an example of how to train an ML Model using Spark, serialise the model using MLeap and serving the model using Spring Boot. The source code which we will walk through is available in GitLab. Problem Statement: Given I have a data set relating number of hours studied to test results, Can I train an ML Model to predict a test result given the hours studied? 50: 0.0 60: 1.0 70: 2.0 80: 3.0 90: 4.0 100: 5.0 Training In this example we will be training a Linear Regression model. A Linear Regression model assumes that there is a linear relationship between the input and output. In it’s simplest form: y = mx + c In our example the y value is the student’s test result, x is the number of hours studied while m and c are the coefficients to the model that we are solving for. Run the com.hotels.ml.training.ModelTrainer class in the train module. During the running of the code you will see the following: INFO ModelTrainer: Coefficients: [10.000000000000012] Intercept: 49.99999999999997 This output indicates that Spark has solved for m and c. i.e. m = 10 and c = 50. Therefore if we were to try to predict the test result given the number of hours that a student has studied(x) we could use the equation: y = 10x + 50 It does this by solving simultaneously for m and c in the following set of equations: 50 = 0x + c 60 = 1x + c 70 = 2x + c 80 = 3x + c 90 = 4x + c 100 = 5x + c Once the model has been trained by pipeline.fit(dataset) the model is serialised to a bundle using MLeap simpleSparkSerializer.serializeToBundle(pipelineModel, “jar:file:” + new File(fileName).getAbsolutePath(), datasetWithPredictions). We will cover why we use MLeap in the next section, serving. Serving We chose the MLeap runtime library as the solution to our serialisation problem because the MLeap runtime executed our models the quickest. Spark does have the ability to serialise and run models but there is a performance hit in creating a Spark context each time you want to execute your model. This performance hit was unsatisfactory. We wrap our bundle and the MLeap runtime in a Spring Boot application as an example of how to deploy a model into a production environment. Run the com.hotels.ml.service.SpringBootService class in the service module to start the Spring Boot application. To interact with a MLeap bundle you use aml.combust.mleap.runtime.frame.LeapFrame. Have a look at the com.hotels.ml.service.service.PredictionService class for an example of how to create a LeapFrame and extract the prediction result from a LeapFrame. The Spring Boot application should now be available at http://localhost:8080 To request a prediction try: curl http://localhost:8080/predict?hoursStudied=4.5

Machine Learning: Training and Serving

27

machine-learning-training-and-serving-16ab3b9c797b

2018-08-07

2018-08-07 08:15:02

https://medium.com/s/story/machine-learning-training-and-serving-16ab3b9c797b

false

461

Public blog for the Hotels.com Technology team.

null

The Hotels.com Technology Blog

null

hotels-com-technology

TECHNOLOGY,TRAVEL,CAREERS,CLOUD,DATA

HcomTechnology

Machine Learning

machine-learning

Machine Learning

51,320

Jared Besson

null

1b70d16f524f

jaredbessongm

3

1

20,181,104

null

0

null

0

7219b4dc6c4c

2018-09-15

2018-09-15 10:09:46

2018-09-15

2018-09-15 12:51:22

8

false

en

2018-10-08

2018-10-08 10:52:35

0

16abe0eb4181

7.710692

27

0

2

Deep learning (DL) has become a common word in any analytic or business intelligence project discussions. It belongs to a broader…

5

The Scuffle Between Two Algorithms -Neural Network vs. Support Vector Machine Deep learning (DL) has become a common word in any analytic or business intelligence project discussions. It belongs to a broader Artificial intelligence field of study and part of machine learning algorithms to be specific. These models are purely based on learning patterns and representations found in the given data (understand the data patterns vs. fitting a line, hyperplane or a decision boundary) compared to task-specific algorithms. Learning can be supervised, semi-supervised and unsupervised. DL models play a vital role in computer vision, speech recognition, natural language processing, bioinformatics, drug-design and machine translations (translation from one human language to another; i.e., English to Hindi) to list a few. In simple terms, most deep learning models involve stacking multiple layers of neural nets in a particular architectural layout for either a prediction or classification problem (Reinforcement and Generative architectures deal with a different set of real-world problems). Neural nets are versatile, robust and scalable and they can handle high dimensionality tasks with ease (an extreme number of feature set; i.e., in object recognition — identify whether an image contains a cat or dog, each pixel colour channel will be a feature; a 120x120 image leads to a matrix of 14400 pixels and multiply that by three for RGB channel intensity. we will end up with 43200 features to start with) Feature explosion in a simple image classification task (i.e., whether an image contains a dog or cat) Before the rise of neural nets in mid-2010, support vector machines used to play a significant role in high dimensionality predictive problems like text classification and speech recognition. In a traditional classification task (i.e., predict whether a patient will be diagnosed with a disease based on the given list of symptoms and family health records; The output is always either a yes or no and also the propensity of the output), the objective is to find the decision boundary which separates the target variable’s categories (disease state : yes or no), a logistic regression works well when the data is linearly separable but fails to understand the non-linear relationship. SVM employs kernel tricks and maximal margin concepts to perform better in non-linear and high-dimensional tasks. Even a powerful SVM model, most of the times, benefit from the proper feature selection and feature extraction/transformation techniques. Artificial Neural Net concept was not something new to the computer science world. It was first proposed by Warren McCulloch, and Walter Pitts in 1943 and the United States Office of Naval Research tasked Frank Rosenblatt in 1957 to build the perceptron (neural net) algorithm. A single layer perceptron did not perform up to the expectations as it could only capture limited linear patterns, stacking two or more neural layers (feedforward neural net or multilayer perceptron) improved the performance but still cannot predict an XOR function. Discontinuous XOR decision boundary Marvin Minsky and Seymor Papert in their book entitled Perceptrons showed that it was not possible for these networks to model simple XOR function in 1969. For many years the book’s citation kept the progress in the ANN area very limited to none. It was only in the 1980s the algorithm resurged into active research, and in 2012 Geoffrey Hinton demonstrated the use of generalized backpropagation algorithm for training multi-layer neural nets in the Imagenet challenge which revolutionized the field of deep learning. Growth in DL usage should also be attributed to the enabling fields. Data processing front saw groundbreaking changes in Mid 2010. Hadoop distributed ecosystem changed the way in how data is processed and stored. Single core processor’s processing power has increased manifold compared to processors in 1980s, and The emergence of the Internet of Devices made a vast amount of data collection possible which provided the much-needed training data for neural nets. Graphical Processing Units perform well in matrix multiplication compared to a multi-core processor, and neural nets heavily depend on matrix operations to fulfill their necessary calculations. Acknowledgments to all the gamers across the world because of them, now neural nets can be trained much faster on GPUs. Without your relentless effort and resolute, there will be no better GPUs in this world. The fundamental unit of a neural net is a single neuron which was loosely modeled after the neurons in a biological brain. Each neuron in a given layer (i.e., layer 1) will be connected to all or as many neurons in the next layer (i.e., layer 2). The connections between neurons mimic the synapses in the biological brain. A neuron will only fire an output signal if it has received enough input signal (in magnitude to cross a set threshold) from its predecessors. Single neuron modeled after the biological neuron List of techniques which improved neural nets performance over time that helped it to beat SVM: 1. Backpropagation: A multilayer perceptron(MLP) have an input, hidden and output neural layer. Training an MLP is an insurmountable task until in 1986 Rumelhart published an article introducing Backpropagation training algorithm (also known as Gradient Descent using reverse-mode autodiff). For each training record (data point) the algorithm calculates the neuron output from each layer and then finally in the output layer makes a prediction(forward pass), based on how far the prediction is off from the actual output it calculates the prediction error. The prediction error is then used to change the weights of the neurons in all the previous layers (backpropagation) until it reaches the input layer to improve the overall networks prediction accuracy. 2. Number of hidden layers and neurons per hidden layer: A single layer neural net can give reasonable results but stacking them together improves the learning capacity of the network. A multilayer neural net for face detection will outperform a single layer neural net. When stacked the lower layers can capture the lower-level details (i.e., the lines separating the face from the background), the middle hidden layer can capture mid-level details (i.e., squares and circles) and the output layer can detect the high-level features (i.e., pixel location of the eye). Adding more layers and more neurons per layer will lead to model overfitting, greater training time and Vanishing/Exploding gradients problem so these parameters will require careful considerations. 3. Activation functions (Vanishing and exploding gradients — non-saturating activation functions): An activation function decides when a neuron will fire and the magnitude of the output based on the input signals from the predecessor. It can be a sigmoid, tanh, softmax or a ReLU variant. It is common to use ReLU (Rectified Linear Unit) as the activation function for input and hidden layers. For the output layer either a softmax if it is a classification task or the actual value if it is a prediction. When RELU is used in a deep layered neural net, the backpropagation signal will either diminish to zero or explodes into a large number when it reaches back the input layer, with no proper backpropagation signal the weights will never change in the lower layers. Variants of ReLU comes to rescue. Leaky ReLU, Randomized leaky ReLU, Parametric leaky ReLU and Exponential Linear Unit (ELU). Performance tests have shown the following order of preference. ELU > leaky ReLU (and its variants) > ReLU > Tanh > Logistic (Sigmoid) 4. Batch normalization: Sergey Ioffe and Christian Szegedy proposed BN in their paper in 2015 to tackle the vanishing and exploding gradients problem. Just before the activation function of each layer, zero-center and normalize the inputs, then scaling and shifting by two new parameters (one for scaling, the other for shifting). This lets the model learn the optimal scale and mean of the training data in each layer 5. Reusing pre-trained layers (Transfer Learning): The lower layer weights of a pre-trained model can be reused instead of training a new model from scratch. If we are building a model to identify a dog’s bread, then we can use the lower layer weight of the model which determines whether an animal in an image is a dog or not 6. Faster optimizers: Optimizers calculate the backpropagation signals, and this helps the net in adjusting neuron weights across all layers. The performance and speed of the optimizer have a direct impact in the training speed of the net. Momentum optimization by Boris Polyak in 1964 was the forefather of all optimizers. Later came Nesterov Accelerated Gradient, AdaGrad, RMSProp and Adam optimization. Adam performs better than other optimizers 7. Learning Rate scheduling: It is critical to find the right learning rate. A smaller learning rate will take forever to reach the optimum solution, and a larger learning rate will swing across the boundary instead of reaching the optimum. Instead of a constant learning rate, it is highly recommended to use a high learning rate during the start and reduce it during training. Typically optimizers should take care of this for the users. 8. Early stopping and l1 and l2 regularization: Stop training the network when the performance actually drop compared to previous epochs. Regularizations of neuron weight (not the biases) using l1 or l2 norm help in avoiding the network overfitting to the training data 9. Dropout: This concept was proposed by Geoffrey Hinton in 2012, and it has helped the networks from overfitting. At every training iteration, the neurons in all the layer including input have the probability of p to get dropped out of the network training. This technique leads to a new architecture trained in each iteration and leads to improving the model accuracy without overfitting to training data 10. Data augmentation: Labeled data is more valuable than any precious metal in the DL land. Each network will require a significant amount of labeled data for it to train (i.e., In object detection for cat vs. dog in a given image, we need labeled images for training — were images are tagged as either cat or dog by a human). However, when we have enough labeled training data it is possible to add some modifications to the labeled data point to generate more labeled training data (i.e., by rotating the cat image by an angle or changing the pixel intensity of a few pixels). For each labeled image data point we can generate multiple data points using data augmentation Research article trend from academic.microsoft.com to identify the leading algorithm: Let’s look at the published article trend for the neural nets vs support vector machines starting from 2000. There is a significant uptake in the article volume for the neural nets, and they have surpassed SVM significantly in active research in the last seven years. Collections of the articles which talks about the topic of neural net (tags: ANN, Neural Networks, LSTM, and CNN) and SVM (tags: SVM and Support Vector Machine). I hope the scuffle between the machine learning algorithms leads to better and intelligent products to serve the human endeavours.

The Scuffle Between Two Algorithms -Neural Network vs. Support Vector Machine

134

the-scuffle-between-two-algorithms-neural-network-vs-support-vector-machine-16abe0eb4181

2018-10-08

2018-10-08 10:52:36

https://medium.com/s/story/the-scuffle-between-two-algorithms-neural-network-vs-support-vector-machine-16abe0eb4181

false

1,743

Analytics Vidhya is a community of Analytics and Data Science professionals. We are building the next-gen data science ecosystem https://www.analyticsvidhya.com

null

analyticsvidhya

null

Analytics Vidhya

medium@analyticsvidhya.com

analytics-vidhya

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

analyticsvidhya

Machine Learning

machine-learning

Machine Learning

51,320

Manikandan Jeeva

Data science enthusiast. Python & SAS addict who solves key business problems in pharma commercial analytics. Lead product architect @ Datazymes | Datazymes.com

aa7875ccbe3c

dataibreathe

174

4

20,181,104

null

0

null

0

null

2018-04-18

2018-04-18 20:30:07

2018-04-18

2018-04-18 21:47:13

1

true

en

2018-04-18

2018-04-18 21:48:22

3

16abef526020

16.162264

10

0

Hey guys!

5

There’s an AI for that — a sci-fi short story Photo by Franck Veschi on Unsplash Hey guys! This short story is a rewrite of a sci-fi piece I wrote for the XPRIZE writing contest “Seat 14c”. The setting for the story was that a plane heading from Tokyo to San Fransisco was caught in a “wrinkle in time” that transported the plane 20 years into the future. For the passengers on the plane, this was felt as just a slight turbulence while the plane, in a blink of an eye, was transported forward in time. The stories were also to have a positive outlook and the readers were to experience the world through the eyes of the passenger in seat 14c. I urge you to go read the winner’s story, as well as the stories submitted by a few of the genre’s top names. You can find them here: Seat 14c I hope you enjoy this as much as I enjoyed writing it. -Morten “Ugh…”, I muttered, trying to rub the sleep out of my eyes. I stretched my body as best I could in my seat by the window and, looking out of it, I saw clear skies and the beautiful early morning light as it can only be seen from an airplane in high altitude. A slight shiver ran through the plane while the tinny sound from the PA speakers told us to… The PA hissed and popped for a few a few seconds before going silent and a glitch in the in-flight movie made the movie system restart. I leaned back and tried to relax the rest of the way, but when the plane started to descend through the clouds I couldn’t help but stare at the marvels I saw in the distance. Impossible buildings, tall, thin and shaped like nothing I had ever seen before, lots of things flying through the air, swarming around them, and a flight of what looked to be the new F-35 fighter jets coming up beside us. A closer look as the fighters drew nearer let me see that there was no glassed cockpit. -”Well”, I thought to myself, “I knew San Francisco was the center of innovation in the new world of software and startups, but I guess I had no idea as to just how far they had come.” — — — The next few days were scary, exciting, depressing and hopeful all at once. We were told that when the plane had been about 1500 nautical miles out of the West Coast, at 4:58 AM June 28th 2017, ANA Flight 008 from Tokyo to San Francisco had disappeared from the radar and tracking systems. It then reappeared on the radar and tracking systems at 4:58 AM June 28th 2037, 20 years into the future. For the passengers and flight crew this had happened in the blink of an eye, only felt like the slight shiver that ran through the plane, while for the rest of the world the plane was assumed lost without a trace in the Pacific Ocean. The scientists that examined us upon arrival and during our quarantine spoke of “wrinkles in time and space”, “tachyons”, “gravity waves” and a lot of other terminologies invented since the plane jumped forward. We were the first confirmed jump, but evidence pointed towards this not being an isolated incident. — — — I was coming to San Francisco on a 3 month visa to join an incubator in the Bay Area to get funding for my startup and to build a network of fellow entrepreneurs and inventors, but those plans suddenly needed revision. With no one to reach out to in San Francisco, and being raised as a foster child in ever-changing families, I had no one to miss at home. This made my transition to the new situation was far easier than for most, and for once, being a loner and an outcast proved advantageous. The authorities granted us all asylum if we wanted to stay in the autonomous city-state of San Francisco, the wealthiest member of the Californian Union. Having nothing to return home to, I accepted, growing eager to start examining this brave new world. I managed to learn some things about the outside world while in quarantine. One of the most interesting bits I learned was how swarms of cheap drones, wireless transmission of electricity and $10 tablets from the Raspberry Pi Foundation resulted in global access to the internet. The ubiquitous internet access, paired with a revolution in online learning experiences, made this the greatest social movement of all times. Curious about how this came about, knowing it would have cost a lot of money, I nearly tormented my “keepers” for ever more details. It turned out that the world’s leading philanthropists had led an initiative to fund this by creating an AI-controlled foundation where the AI would process available data and choose where to donate, invest and give interest free loans to influence the market, pushing it in a sustainable, moral and ecological direction, and with great success. This led to an ethically and morally driven movement of investors and benefactors, pouring funds into the foundation. This new generation of investors were not seeking maximum profit short term, but rather to maximise the social economic benefit, which in turn, over time, would allow the investors to profit in other areas as the purchasing power of the lower and middle class increased drastically. With all the available resources that it gathered the AI supported the creation of even more AIs to be used for better utilization of natural resources, infrastructure and the available workforce, as well as giving targeted economical and social aid directly to individuals, effectively transforming the world economy, giving all people equal opportunities through a base income and universal, high-quality health care and online education. — — — I got some temporary leave from the facilities where we were housed until other arrangements could be made, and walking around that first time made me feel like I was in a sci-fi movie. I was almost dumbstruck when I saw the sheer amount of “Automatas” that were out and about. Autonomous machines such as drones, vehicles, multi and mono-pedal robots in all shapes and sizes were scurrying around, carrying out every task imaginable. Looking around while walking down the street, I saw screens, cameras and microphones on just about everything and everyone, much like in 2017, only even more encompassing. Interactive holo-screens, old school LED-billboards, Harry Potter-style posters as well as the personal devices people used told a story about a world so connected that privacy was nearly disintegrated. Most of the Automatas looked like something a child would design from available scraps disregarding all aspects of aesthetics, focusing purely on the functional. Antennas, wires, tubes, panels and strange angles were common features. There were exceptions of course, with beautiful futuristic designs, smooth skins and casings in brushed metals, colourful plastics and rubbers, not to speak of the lifelike humanoids. I remembered those early prototypes from 2017, but these were naturally far more impressing. They seemed to be just as dexterous and sensitive to tactile feedback as humans, and when spoken to, they performed well above and beyond any Turing test I could remember. If not for having the brand logo stamped on their forehead, I would have had a hard time telling some of them from real humans. As the culture shock started to hit me hard, I found myself desperately needing a coffee. — — — I walked into what passed as a retro coffee shop, feeling oddly at home. To my eyes it looked a lot like what the most modern designs which still had a cozy feeling to it, looked like in 2017. As I was admiring the coffee shop, taking it all in, she spoke to me. -”Hello!”, she said, her stunningly blue eyes smiling at me. -”What can I help you with today?” I looked at the menu, relieved to see they still served Americanos in the future. -”One double Americano please”, I replied. -”Sure thing!” The young woman looked expectantly at me for a moment before she indicated a slightly raised, rectangular section of the counter. As I looked at it, she told me she could not connect to my “Auta”, my Automata, for payment, so I had to use my hand. Placing my hand on the raised section I told her I had no “Auta”. -“Do you mind if I ask if that is by choice or accident?”, she asked, ”I know some people prefer to be disconnected, but usually the elderly or religious types, neither of which I would peg you as.” She took a closer look at me, seeing my Converse shoes, jeans and t-shirt, the latter sporting a large Pac-Man eating a ghost. -”You do have a nice retro outfit on, though. What period are they? Millennials? Nineties perhaps?” She looked at my hand confused. -”I’m not registering your chip either. What happened? EMP-accident?”, she asked. -”You know about the disappearing planes, right?”, I asked her. -”Of course, everyone knows about that. Especially now that one returned! Can you believe that? 20 freaking years!?!”, she exclaimed, looking genuinely intrigued. -”Well, I am one of the returned passengers.”, I said, ”I was not told about the chip, nor the Automatas being used to pay. We were told they would sort out the paperwork related to bringing people back from the dead shortly and restore our banking abilities after that.” Her eyes gleamed with interest as I spoke. -”I have cash?”, I ventured with a smile, waving a $20-bill. -”Oh, wow!”, she laughed, her smile growing even bigger. -“Not only do I get a genuine time traveller in my shop, but one carrying cotton-money!” Apparently cash was no longer king, and the implant based payment and identity system from the early twenties, had also become obsolete, though most still had them implanted, and a few places, like this retro coffee shop, still accepted crypto-coins through them. — — — -”Ok, so what you are saying is that all these objects”, I said indicating everything with a hand wave, “the Autas, the buildings, the personal devices, even the pavement”, I pointed at the floor, “contains a specialised AI?” We had taken a seat at a table in the shop to talk after she offered to buy me the coffee. She took a sip from her own cup, looking a bit more than slightly amused, and it occurred to me that to her, I must seem like Marty McFly or something, unsure if she would even get the reference. -”Yes”, she said, “advances with Neural Processing Units, or NPUs, as well as the Neural Memristors, a new type of computer memory, made it possible to fit the computing power of the equivalent to an early 21st century data centre, on a playing card sized circuit board powered entirely by ambient radiation from electromagnetic sources such as wireless networks and radio waves.” She took another sip. After emptying my cup, a little dizzy from the sudden caffeine and information overload, I sat back in my chair, and looked at her. -”What is it?”, she asked, meeting my eyes, starting to look a bit self-conscious. -”Well…”, I start, “I find myself a little in the deep end here.” I took a slow breath, starting to feel a real connection with someone for probably the first time in my life. What baffled me was how fast I felt that connection, but somehow I just knew that I could trust her and open up to her. -“The world has become a bit alien to me with all these AIs, Automatas, Autas or what you want to call them.” I noticed an unusual gleam in her eyes, almost like a reflection that should not be there. -”I mean, how do you even tell the difference between smart or dumb objects? How can you tell if you are truly alone or not?” Stroking her perfectly cut, hazel hair behind her left ear, she leaned forward, putting her perfect hand on top of mine. -”To be honest?”, she said in a soft tone, “You are never alone like that anymore, unless you’re out in the forest or something, and even then there might be environmental drones supervising wildlife, counting bees or something.” She tightened her grip a fraction, meeting my eyes again. -”There is, however, a built-in protocol in all Autas, except the humanoids and the vehicles, enabling you to choose to be ignored by them, any data picked up about you silently discarded.” Looking down at her cup, she continued, “All those concerns was addressed by the EU in the mid twenties, after a couple of cases with some high-ranking officials being outed publicly for various extracurricular activities not normally associated with upstanding citizens in public office.” She grinned at me, looking at me with sparkling eyes I found I could stare at all day. I was definitely in the deep end in more ways than one. What was it with her that gripped me so fast? -”The suggestions were actually well formulated, to the point and easily understandable, not the usual convoluted language officials use, so most of the rest of the world followed suit.” Looking down at her hand on mine, feeling her pleasant warmth reassuring me, I started to feel a little better. -”So”, I said, looking back up at her pristine smile with flawless teeth, almost perfectly symmetrical, slightly almond-shaped eyes, “I was wondering how you know all this? You can’t be any older than me, so a lot of this must have happened while you were a kid.” She stifled a laugh, taking a diplomatic sip before answering. -”From what we have just discussed, you could not figure that one out?”, she said humorously, and I took another stab at it. -“Well, there are all these objects, large and small, with built-in brains, all around us. Personal devices, building, cars and everything.”, I ventured. “How small are these personal devices really?” Then it hit me, the payment chip, the pauses people had mid conversation around us, like they were waiting for something, like she did at the counter earlier. -”They did it didn’t they?”, I said eagerly, the tech geek in me waking up. -“The whole wearables thing was just a prequel to this wasn’t it? It’s embeddable now right? Cochlear implants, contacts or retinal implants, tiny sensors monitoring every aspect of your body’s vitals, right?” -”Almost”, she answered taking another sip, “some people did try that route early on, but it turned out to be too expensive and way too invasive for most.” -“It’s actually nanites. Ingestible via small capsules, powered by one’s own body chemistry, designed to attach to your neural network, as well a single embedded capsule at the base of your skull, sort of like the old RFID-chips used on pets. The capsule contains the AI, while the neural rigging allows it to send signals directly to your optical and auditory nerves, as well as other sensory nerves.” I closed my mouth, realising I was gaping at her. -“Nanites? Freaking nanites rewiring your nervous system?”, I exclaimed, almost rising out of my chair in excitement. She pulled her hand back, looking at me, taking a sip of what must now be cold coffee. -”Not everyone uses that approach of course, some prefer the contacts, glasses, ear buds or handhelds still. A few are even more fully connected.” -”I think I need some air”, I said, looking expectantly at her as I stood up fully. -”Sure! Want me to show you around?”, she rose as well, took a step towards the door, holding out a hand to me. I accepted her hand, letting her led me out of the coffee shop, thinking to myself that this world would take a lot of getting used to, and hopefully my new friend could help me find my bearings. — — — We walked around for hours, talking, laughing, really enjoying ourselves and getting to know each other more. I told her about my startup plans and how I was going to “revolutionize the world” with a system based on blockchains, the technology behind Bitcoin and other crypto-currencies, and she told me about working in and owning the coffee shop, even though everyone has the basic income. She said she had a thing for the “olden days”, and also wanted something of her own creation to thrive, a place to belong to. She also admitted to being amongst the most deeply connected as well, which I took to mean that she had gone with the nanites, but I felt a slight hesitation, so I didn’t want to press the issue. We talked a lot more about the other marvels achieved with focused AIs controlling the resources in the most efficient way, trying to create a version of Utopia. They had found a way to put graphene into a new alloy made up of carbon fiber, titanium, aluminum and a new synthetic metal I couldn’t even pronounce the name of, and created a sort of hybrid material stronger than steel, almost as light as pure aluminum, yet flexible enough to endure the winds and earthquakes, without experiencing material fatigue in the same way as normal metals do. This made the impossible buildings possible apparently and thanks to the embedded graphene, the structure was able to monitor itself for wear and tear and, if a weakness was detected, it could be fixed by simply running high current through the material, much the same way the electrical grid operators remotely weld the power lines. Building costs and time had also been drastically reduced by this, as the material was suited for 3D printing, hardening when exposed to a specific type of electromagnetic radiation, similar to the hardening process with UV light used in dental cement. When the time came to return to the facility where we were welcomed to the future, I felt very reluctant to go, even though I was going to see her again. If I had been given my AI kit, we could have just synced those up to stay in touch, but that was one of the things we were going to get later at the facility. -”Promise you will come back for me”, she had said, holding my hand, biting her lower lip gently. “I have never felt so connected to anyone so, well, disconnected before”, she said, hinting at my total disability to access any wingman AI giving dating tips, hints and suggestions presented in real-time, or any other kind of online help. I had of course wondered if she was using those kinds of systems herself, being so connected as she apparently was. To prove she wasn’t, she had invoked the privacy-protocol. She did admit to using online info to answer my questions when she didn’t already know the answer though. She also explained that the online info was not instantaneous, and had to be read or heard, which suddenly made her constant sipping of coffee make perfect sense, as well as those weird pauses some people made mid-conversation. -“Nothing short of interning me or pushing me through another time jump can keep me away”, I said, trying to act more manly than I really was. -”Oh, is that how it is, Mr. Time Traveller”, she laughed, calling my bluff. -“How about we just say two days from now then, in the shop, about three-ish?” — — — I was given valid ID, access to my basic income, a little extra stipend on top to get me started, and a basic starter model of a personal AI device. It was a simple handheld model, about the size and shape of a marker pen. It had a practical clip, just like the pen, and a little handle to roll out the screen. When rolled out, the screen was rigid as if it was made of glass, touch enabled, of course, as well as voice controlled. It was able to shift between soft and rigid by using the handle to apply pressure pushing it in or tension pulling it out, sensors activating a small current in the material, making it bendable. The screen could also shift backgrounds between opaque, translucent or anywhere in between, on a pixel by pixel basis, to fit the application it was being used for. The device came with an ear bud that fit snuggly, without reducing ambient sounds noticeably. When the time came for our final release, I was really anxious, eager, excited and nervous, all rolled up into one large knot in my stomach. I was sitting alone at a table in the facility diner, trying to pass the time by getting to know the AI in the device. It turned out that all AIs was trained individually, and through their training, they got differing personalities, if one could use that term on an artificial entity. The AIs were matched up with their prospective owners to ensure they got along, and some even became emotionally attached to their AI, due to their life-likeness. Sitting there I overheard a conversation between a few of the other jumpers, as we were called. -”Yeah, that really blew me away”, a man with ginger hair and glasses said, the Irish accent thick. -“3D printing body parts! Can you imagine?” -”Yeah, well, they were almost doing that in 2017 as well”, a burly American replied, “but did you hear about the nanites? Designed to attach to the various sensory nerves, making them attunable to the AI implanted in the neck. Now that’s something!”, he exclaimed. -”Does it not bother you with all these robots walking around?”, and older woman asked the group, also American. -”Nah”, the ginger man replied, “I’m more bothered with the prospect of not being able to tell the difference between a real human, and a biological AI. They actually used the 3D printing process to build their bodies, with somewhat randomised DNA and the AIs Neural Processing Units printed directly inside the brain.” I snapped my head up, hearing that last comment. That was news to me. -“James”, I said, having given my AI assistant the least imaginative name ever, ”tell me about the biological AIs, please.” James spoke to me with a voice fitting the name, British accent and all. -”As you wish, sir!” -”They were created right after the DNA printing process was invented, giving the last piece of the puzzle. This happened in 2029. There were only ever made 5 of them in the world, that we know of, before the public outcries became so forceful that the UN, in 2032, passed a resolutions banning the production of the artificial humans.” I sat there, listening to James explain, my head whirling with thoughts and questions, mostly philosophical and metaphysical in nature, before a clear question formed in my mind. -“James? Tell me please, are they AIs like we see around us here, limited to particular functions?” -”No, sir”, James replied, “they are not specialists at all, they are generalists, being able to learn anything and everything, just like humans, but due to their integration into a biological brain, they have a bit less capacity than the Automatons. They are about equal to a fully connected human in regards to cognitive abilities and learning speed.” -”So, they are basically unborn, printed humans?”, I asked, starting to feel sympathy for them, not having a real family myself, just another survivor of the system. -”That is a good generalisation, sir”, James replied. -”With the human brain involved”, I ventured, “would that mean that they are truly self-aware, independently sentient beings? Equal to humans in all ways but birth?” -”Perhaps not all ways, sir, but in all the important areas, yes. They can even legally procreate like humans do these days, by submitting DNA samples at a lab, and then wait for the babies to be grown in the artificial wombs. A technique, by the way, that eliminates the need for sex for procreational reasons”, James explained. I found myself intrigued, rather than put off, like my fellow jumpers seemed to be. This sounded like someone who would feel about the same I had always done, outcast, alone, with no family to attach to, someone I could relate closely to. -”Not quite the singularity I was expecting”, I thought to myself, thinking about the AI naysayers predicting doom the day the AIs become self-aware. I sat there deep in thought until it was time to leave. -”One last thing, James”, I said, gathering my things, getting ready to meet my date. -“Do any of them live in town? I think I’d like to meet one of them some day.” I almost missed a step when I heard James’ answer and the pieces fell together, but as I continued walking, I found I really didn’t care. I opened the door to the street outside, the sun warming my face, birds singing nearby, and I started whistling as I set a course for the coffee shop, the words James had said ringing in my ears, “Yes, sir! She owns a retro coffee shop downtown.”

There’s an AI for that — a sci-fi short story

80

theres-an-ai-for-that-a-sci-fi-short-story-16abef526020

2018-06-17

2018-06-17 10:55:27

https://medium.com/s/story/theres-an-ai-for-that-a-sci-fi-short-story-16abef526020

false

4,230

null

Short Story

short-story

Short Story

94,626

Morten Olsrud

Developer doing SysOps by day, trying to find my way back to coding.

d0bbc4d3b5de

morten.olsrud

13

19

20,181,104

null

0

null

0

null

2018-01-28

2018-01-28 17:15:58

2018-01-29

2018-01-29 14:16:22

2

false

en

2018-08-17

2018-08-17 20:54:04

15

16ac1fb6539b

2.349371

3

0

I have always felt the study approach needed to evolve beyond traditional paper booklet containing over 20 years of past questions…

5

IBM Watson Chat Bot Finds Application In Nigeria Education I have always felt the study approach needed to evolve beyond traditional paper booklet containing over 20 years of past questions. Nigeria’s budding Tech space of Start-up have a handful of apps that digitalize past questions. The successful and popular ones are PASS.NG, JAMB CBT Past Questions other playing catch up are ALOC.NG, PrepClass . Past questions booklet for JAMB-UTME & WAEC-SSCE In Nigeria, secondary school students (High School) take a national exam organized by JAMB (Joint Admission Matriculation Board) which determines who gets admitted into various high institutions. Using past questions is essential to adequately prepare for the exams. On our first attempt to rethink JAMB CBT was a game test approach where students compete against characters and other students to earn points, trophies and badges. On our platform we reward the best 5 students with the highest points weekly. We have done this for over 130 weeks. We were not satisfied with this app, after exposing the app to over 1500 students across 8 schools and 10 tutorial centres. We knew that young people love social engagement and spending long hours on various social platforms. These were the kind of traction we hoped for and drive we wanted Nigerian students to give to studying and exams. On October 2017, in a LagBus popular called BRT (Bus Rapid Transit) in the heart of Lagos early morning traffic, an idea came to me that we could make an app that operates like a chat. I checked around if something similar already exist, asked couple of friends but we could not find an app that implements this concept or maybe we did not search enough (lol). I quickly start researching tools to use. I tried socket.io and express. A friend then showed me LARA.NG (Bus-fare calculation bot), Kudi.ai (Call card purchase bot). That was my first encounter with chat bot technology. I tried my hands on IBM Watson, Dialog Flow (formerly API.ai), Wit.ai etc. I read a lot of reviews and watched tutorials on “University of YouTube” From my research IBM Watson is interestingly easy to use. Following documentation, I could easily integrate IBM Watson with my android app. After unit testing, we now have Chat & Pass JAMB CBT . Unlike Artificial Intelligent Markup Language (AIML), IBM Watson gets you started in minutes. With the CBT chatbot we built, Nigerian students can now answer past questions, get inspiration quotes, latest exam news, have mild conversation with the bot all with the help of IBM Watson. Sample page of exam chat bot app If anyone from IBM Watson team stumbles on this article, I reiterate the following issues on their site. *It takes more than three clicks to move from the homepage (www.ibm.com/watson) to access Watson conservation work-space. The process really got me confused for a while. *I also don’t like the fact that I have to login every time when I want to access Watson work-space. A remember session at login will be cool. Thank you, IBM. Ope Mesonrale is the founder of ALOC.NG . Our team is experimenting with various ideas on how to make academic practice more entertaining/engaging for students

IBM Watson Chat Bot Finds Application In Nigeria Education

53

ibm-watson-chat-bot-finds-application-in-nigeria-education-16ac1fb6539b

2018-08-17

2018-08-17 20:54:04

https://medium.com/s/story/ibm-watson-chat-bot-finds-application-in-nigeria-education-16ac1fb6539b

false

521

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Mesonrale Ope Seun

Blockchain Enthusiasts | Software Engineer ( Laravel, JavaScript frameworks, Java ) | Founder @magbodo | https://aloc.ng

292f3551c23a

mesonraleopeseun

32

38

20,181,104

null

0

null

0

863f502aede2

2018-07-16

2018-07-16 21:14:18

2018-07-16

2018-07-16 21:25:04

3

false

en

2018-07-16

2018-07-16 21:25:15

4

16ad94657322

1.319811

9

0

Shanghai-based computer vision company Yitu announced today that it has acquired an additional US$100 million in funding, with China…

5

Computer Vision Startup Yitu Nabs $100 Million in Funding Shanghai-based computer vision company Yitu announced today that it has acquired an additional US$100 million in funding, with China Industrial Asset Management as the sole investor this round. Yitu is one of China’s four AI unicorns, along with Face++, Cloudwalk and SenseTime. On June 12th, 2018, Yitu completed its C+ Round with US$200 Million from GC Capital, ICBC International, and SPDB International. This new funding raises Yitu‘s valuation to over US$2.3 billion. Yitu executives pose as a drone scans their faces at the opening of the company’s first international office in Singapore Yitu was found in 2012 and focuses on computer vision, natural language processing, knowledge reasoning, intelligent hardware, and robotics. Computer vision is its core technology and has been applied to security, finance, and medical industries. In the latest NIST (National Institute of Standards and Technology) competition, Yitu ranked №1 among computer vision companies on the FRVT (facial recognition vendor test). Yitu also took first place on the NIST 2017 leaderboard. The FRVT measures automated facial recognition performance using a real world dataset from the US Department of Homeland Security. Author: Alex Chen | Editor: Michael Sarazen Follow us on Twitter @Synced_Global for more AI updates! Subscribe to Synced Global AI Weekly to get insightful tech news, reviews and analysis! Click here !

Computer Vision Startup Yitu Nabs $100 Million in Funding

87

computer-vision-startup-yitu-nabs-100-million-in-funding-16ad94657322

2018-07-20

2018-07-20 21:24:23

https://medium.com/s/story/computer-vision-startup-yitu-nabs-100-million-in-funding-16ad94657322

false

204

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

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

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

2018-04-20

2018-04-20 20:39:22

2018-04-20

2018-04-20 20:41:48

1

false

en

2018-04-20

2018-04-20 20:41:48

2

16adae2e29b5

7.818868

0

(Loosely speaking, this is the thematic sequel to Kill The Beast )

5

Self-Thinking Thoughts (Loosely speaking, this is the thematic sequel to Kill The Beast ) If you’ve taken any college-level psychology or have simply paid any attention to the movements inside pop-academia and University Scholarism, you might have noticed, like myself, that the industry catch-all nominally termed “Science” is now officially banding together to tackle what has been, to date, its largest and most difficult problem. Glibly coined “consciousness”, the loosely constructed and highly debated notion of self-awareness has been somewhat of a compromising situation among mainstream scientific enterprise and its adherence to Descartes maxim concerning the observer (the ego) and that which is observed; generally, the material world or all that exists outside of the ego. The proverbial “fly in the ointment”, one might say, is that, given a heretofore purely mechanistic explanation of the universe and the Natural World, what is a seemingly spiritual entity (the conscious being) deliberating, independently, between and among the various causalities and finalities of the material world is simply too compromising to be allowed into the Scientific Canon. Indeed, “consciousness” as an independent agency has been something of a loose cannon for orthodox Science since at least the 19th century and the Industrial Revolution. Generally, psychology tackles the conundrum of human agency by separating and isolating the ego, essentially diminishing the pre-modern notion of a soul by materializing it and placing it somewhere inside the brain. The “self”, I was told in college, could very well be located somewhere among the lobes of the cerebral cortex. Going a step further, theoretical physicists are now positing that consciousness is a state of matter, not unlike a solid, liquid, or gas. “Perceptronium”, a biological but maybe not, condition, says the Manhattan Institute of Technology, is a state of matter residing in human brains that has its atoms so arranged as to give rise to the illusion of subjectivity. Another step further (or back), depending on your outlook, is the growing consensus among certain sects of neuroscience which posit that we owe our “consciousness” to the remnants of an ancient virus that infected four-limbed mammals some 400 million years ago. Regardless, of its particular expression, the common theme among the many competing hypotheses is that consciousness, like other illusory subjectivities among the Natural World, is made of material and is essentially incidental and even accidental (as the New York Times coined: “The Mind Messing with the Mind“) (article). None-too-subtle allusions to our sense of subjectivity being a “virus” are, by my take, evidence for a growing angst against existence itself and the growing popularity of the notion that “personhood”, as understood by pre-modern knowledge, is an essentially superstitious and unscientific construct; an enemy to scientific advancement, progress for the sake of progress, and conquest for the sake of knowledge. If the Ancients were right in supposing that one IS his “self” and not a product of his “self”, then the cogwheels and levers of scientific industry, so we are told, begin to become unhinged and Descartes’s maxim is, itself, put under the microscope. The scientist regains control over his experiments. The human being is elevated, once again, to a position above his inquiry. Yet, independent inquiry and autonomous experimentation are both supremely unsexy ideas in this Brave New World being engineered for us. If existence is, what Plato called, a First Principle, then the current-day obsession with consciousness as something more than a byproduct, is something of a non sequitur; a fool’s gold for the overly-educated. Nevertheless, the debate rages on. ……………………………………………………………………. The following concepts, previously invaluable and natural evolutions of human History, are, by my take, casualties of the various conjectures concerning our “selves” and the abandonment of personhood as the central construct of Science. The First Casualty: Intelligence The word volition can be etymologically traced to a Medieval Latin noun (of action), which, from its first-person perspective, simply states: “I wish.” For later audiences, it has come to represent the mental power of deliberation, or, in linguistics, a concept that distinguishes whether or not the subject or agent of a given proposition intended something. One of the more obvious reasons for promoting consciousness as an entity, separate from personhood, is that it absolves the heavy weight and responsibility of choice. If our consciousness is, essentially, an accident of evolution and something we possess (like genes and brain cells), then morality (or the moral instinct) is simply a mirage; a byproduct of an unorganized and unintentional convergence. It would seem only natural, then, that some of the more incongruent concepts championed in post-personhood science are artificial intelligence, machine learning, and, ultimately public education. As “artificial intelligence” strikes fewer and fewer people as a contradiction in terms, what comes to pass is the continual widening of the chasm between the psyche (soul) and ourselves. Indeed, it is not uncommon to hear people, nowadays, insult an opposing argument or contending viewpoint as “below” their IQ. Put more simply, an alarming amount of people are now convinced that intelligence is a state of matter and their “IQ” a symptom of their condition. Thus, a competing hypothesis or counter-argument is to be filtered outright by a system that is independent of their volition. (what is left of it) Intelligence, then, in post-personhood science, is something of a foregone conclusion. For every instruction there is an output. For every output there was a preceding instruction. Deliberation from a center (personhood) becomes wishful thinking. Intelligence as strength-of-will is usurped by intelligence as increase-in-number. Thirteen is better than seven. Thursday is better than Wednesday. “Problems”, as the American commercial complex is so fond of discussing, are generated by the needs of the multitude at the expense of the individual. What is required of future societies are the kinds of IQs to match the complexities of these newly engineered “needs”. Thus, human beings, devoid of the act of becoming (personhood), exist in a state of need; their volition being entirely too fragile for the job at hand. As our computers need power, bandwidth, and computational capacity, so the human who possesses his “self” somewhere in his cerebral cortex is helpless to wish any better for himself. He is ripe and ready for what comes next. The second causality: Education Education, absent of human volition, is exactly that; indoctrination. Publicly-oriented curriculums, aiming at generating a multitude of properly calibrated IQs, will teach its students to draw the shortest line between two points yet fail to mention whether the exercise itself is worthy of the pupil’s attention. A polar opposite of its classical counterpart, modern education does not seek independent minds but a uniform and lockstep groupthink. (Modern “problems” don’t solve themselves). Government schooling, like the Manhattan Institute of Technology’s research on consciousness, aims at generating perceptronium; consciousness as a state of matter. The virus that made us “self-aware” might as well be put to proper use. While classical or pre-modern education suggested that many intact persons converged together to will the present into existence (History), current-day public education must intentionally alleviate the Historical figure of his illusory “self”, and, in so doing, suggest that all History is happenchance; no less incidental then then the accidental evolution of the self. Providence, the Invisible Hand, or even the tragic element are all disposed of in exchange for blind and mechanistic egoism; devoid of volition, passion, and desire. Thus, what was previously understood as Tragedy and Drama (Shakespeare’s World as The Stage) fractures into seemingly endless individual parts; selves and egos helplessly clamoring for the higher ground and the advantageous position; selves preserving selves. The last causality: Culture Given the disparities between post-modernism and the classical tradition, one may suppose that figures like Shakespeare, Dostoevsky, and Herman Melville owed their literary and intellectual prowess, if at least implicitly, to the notion that the study of consciousness, so fashionable to the modern mind, is something of a fool’s errand. One gets the sense, while immersed in the classical world, that the vastness of the imagery and universality of the symbolism is almost entirely dependent on and, in fact, complimentary to a human psychology and theory of intelligence that places volition above computation and the human condition above the condition of being human (the mind messing with the mind). To human intelligence, the moral instinct is also the aesthetic instinct. We do not possess ourselves in the same fashion that we do not experience our consciousness. A conscious being studying consciousness has put himself into an impossible bind. What results from such a study should be suspect to the intelligent mind. As Dostoevsky once went on record to remark; “It takes more than intelligence to act intelligently”. . It would seem as though the postmodern obsession with consciousness as an independent entity is telling of a stagnation in culture and a stifling decadence among the Scholars. Self-importance seems to have usurped the virtue of self-possession. Melville’s Moby Dick, in some sense, a cautionary tale on the futility of the overreach of egoism is almost incomprehensible now that personhood has been abandoned for consciousness and human volition absolved through self-regard. Self-aggrandizement, when wholly subsumed, is an effective end to the creative impulse (Art). In exchange for the universalism and unifying theme of classic literature, self-obsession has bred an environment in which the “human condition” is, in fact, a disjointed and fractured conglomerate of undifferentiated egos. Individuation is abandoned for uniformity. Unable or unwilling to suffer through their condition, modern scientists alleviate the weight of identity by informing the populace that there never was such a thing. Your suffering is just an illusion. The shame you feel is simply the product of accidental chemical collisions. Be mindful! Be cognizant! Suppress the impostor of personality! The Post-Modern Idiot As self-obsessed scientists and the groveling populace that subsidize their endeavors further study the art of self-defeat, while pursuing the abolition of human volition, it is interesting (and humorous) to note that, what we are told regarding intelligence and IQ, is full of purposeful misdirection. The idiot or archetypal dunce, so we are told, is somewhere mulling over his flat-earth theories while propagating climate-science denial. He is, of course, an enemy of the “facts” and hates “evidence” to his core! He is, no doubt, under-educated and steadfastly right-wing in his sensibilities. Why, I’d wager, he could barely pronounce “perceptronium” ! All this, however, is deliberate obfuscation. Post-personhood Science, fueled, in no small part, by the social doctrines and Progressivism in general, must mask and re-brand the obscene largess that gave rise to words like perceptronium . Progress for the sake of progress is the brain-child of thinking for the sake of thinking. A straw-man must be formed. Shakespeare was himself; and so, we have Hamlet. The Manhattan Institute for Technology is a conglomerate of “selves”, and thus we have perceptronium. Yet, could it be that our minds are messing with our minds? Is the answer to end all questioning as final as it is seems to be? Is my sense of moral and intellectual repugnance simply a mirage? Strangely enough, I don’t come down too hard on that question. My aesthetic instincts may simply be a product of my illusory self. My self may, in fact, be a fanciful construct; a ridiculous notion that, once it dawned upon me, gave rise to this silly idea that I could employ my judgement before the “facts” or that I could deliberate before “information.” Yet, a potentially more illusory notion coincided with this; this vain misgiving I have.. See; I don’t want to be stupid. At least, if I’ve come down on the wrong side of the matter, I can safely say that I never spent very long slicing my existence into innumerable conjectures. I created. I did. I became. In simpler terms; with open arms, I readily receive and embrace my right to be wrong. I may aspire to be inaccurate; if I so choose.

Self-Thinking Thoughts

0

self-thinking-thoughts-16adae2e29b5

2018-04-20

2018-04-20 20:42:06

https://medium.com/s/story/self-thinking-thoughts-16adae2e29b5

false

2,019

null

Philosophy

philosophy

Philosophy

39,496

Jon Eames

null

b3d616901c63

joneames

1

20,181,104

null

0

null

0

null

2018-03-12

2018-03-12 20:18:43

2018-03-15

2018-03-15 01:35:16

7

false

en

2018-03-17

2018-03-17 04:33:08

5

16ae06111eb4

4.336792

13

0

1

Random Forest is a machine learning model that combines bootstrap aggregation and random feature splits for decision trees. Bootstrap…

5

How does Bagging Work? (An alternative theory) Random Forest is a machine learning model that combines bootstrap aggregation and random feature splits for decision trees. Bootstrap aggregation or “bagging” is a core component that makes Random Forest work. So you’re saying… by combining a bunch of uncorrelated bad trees you can create a great model? Most explanations online for why RF’s work is that: A) Random feature splits allow to create many uncorrelated trees which focus on different parts of the problem. B) Bagging of uncorrelated trees “average” out and reduces the variance error term(more on this later) The “averaging” out effect of Bagging was a misconception that I experienced when learning this topic. Bagging the base learners of Random Forests and Neural Networks works great, so it must work for other models right? Wrong. In fact bagging can make some models much worse, so it is important to know when to use bagging. Example on Iris Dataset, Experiment Link: Accuracy: 0.63 [Decision Tree] Accuracy: 0.64 [Bagging Tree] Accuracy: 0.70 [K-NN] Accuracy: 0.59 [Bagging K-NN] When does bagging work? Breiman, the creator of random forests and bagging, states that the vital element for gaining accuracy thanks to bagging is the instability of the prediction method. Definition of Stability: How much an algorithm will change based on a small change in input: Unstable Learners: Decision Trees, Neural Networks Stable Learners: K-Nearest Neighbors, Support Vector Machines, Regularized Least Squares Regression Why does bagging work? The answer is…..there’s not a definitive answer Bagging is not very friendly to mathematical analysis. The fact is, random forests rose because of empirical successes rather than sound mathematical theory(Citation). There are many theories: #1, #2, #3. I’ll explain two theories: 1) The most popular “variance reduction” one 2) A secondary theory that states “variance reduction” is a side effect of equalizing leverage points(more on this later) Theory #1: Decision Trees suffer from high variance. Bagging improves variance by averaging from multiple different trees on variants of the training set, which helps the model see different parts of the problem. This variance-reduction argument was introduced by this paper. This theory is the one floating around most of the internet, always explained from intuition. Here’s a particularly good example from Machine Learning @ Berkeley Theory #2: Bagging systematically equalizes the potential influence of examples. This equalization sometimes has the secondary effect of reducing variance. This secondary effect can reduce variance, but also increase it as well! The following is my interpretation: there’s not much literature online, so please comment if you have any input! The authors of the paper argue that bagging’s effect on variance is dependent on leverage points, which are not outliers in the traditional sense but points that can highly affect the model. Leverage Points in a Decision tree: Many, because it’s a highly unstable model Leverage Points in Least Squares: Only Outliers! 3-Step-Argument: 1) Their paper states that in most situation leverage points are badly influential. 2) The effect of bagging is to equalize the influence of leverage points 3) Because in most situation leverage points are badly influential, equalizing leverage points has a secondary effect of decreasing variance and improving predictions. This argument improves Theory 1 because it covers the case of when bagging doesn’t work and makes the models worse. They conducted a following Point Estimation Experiment to look at how bagging affects variance: 1) Draw a n = 20 points sample from a mixed distribution with contamination(to represent outliers) 2) Observe the variance of the average & median 3) Apply bagging 100 times to the n = 20 points and observe the variance of the average & median. 4) Repeat it for multiple values of P (the contamination) Mixed Distribution. The second term(representing outliers) has a spread of 10x the first term Sooo….what’s the point of this experiment? What is really interesting about this experiment is the comparison of median vs mean. Say you had a list of 20 numbers and wanted to find The Median: calculated by mostly the average of two numbers. These two numbers(or 4) are the leverage points — all other values contribute nothing to the median. The Mean: every point contributes equally to the mean. There are no leverage points! Assume the points were sorted by rank. The above graph shows the weight of each point given its rank. For the median, bagging “equalized” the leveraged points and gave the other points of different rank more weight — this had an effect on variance. For the mean, there was no leverage points so bagging had no effect on the variance. Below are the final results for the experiment after 1000 trials with different levels of contamination: Things to note: 1) Bagging the median sometimes reduces variance, sometimes not. 2) The median is especially robust compared to the average at lower levels of contamination such as P = 0.05, P=0.2 3) At P = 0 or P = 1, there is either no outliers(contamination) or the it’s all outliers(contamination), so the median isn’t robust compared to the mean 4) The bagged median had less variance than the median at lower levels of contamination P = [0, 0.2], which may explain why it works in most real life situations when your dataset generally contains low levels of noise.

How does Bagging Work? (An alternative theory)

26

random-forests-in-depth-misconceptions-on-bagging-16ae06111eb4

2018-06-19

2018-06-19 19:54:55

https://medium.com/s/story/random-forests-in-depth-misconceptions-on-bagging-16ae06111eb4

false

871

null

Machine Learning

machine-learning

Machine Learning

51,320

Mike Liao

null

d525012b524c

mikeliao

263

149

20,181,104

null

0

null

0

null

2017-12-14

2017-12-14 15:15:11

2017-12-14

2017-12-14 15:16:16

1

false

en

2017-12-14

2017-12-14 15:16:31

12

16ae50a33874

2.856604

0

It’s been a big year for the financial industry! From Bitcoin making itself known to the high-profile Equifax breach, there have been many…

5

Four Fintech Trends to Watch in 2018 It’s been a big year for the financial industry! From Bitcoin making itself known to the high-profile Equifax breach, there have been many highs and lows. What can we expect going into 2018? In this article we’ll take a look at four emerging trends that will have a big impact on our industry next year. Cryptocurrencies Continue to Make a Splash You’re probably at least a little familiar with Bitcoin by now, and maybe some other cryptocurrencies like Ethereum (if you’re not, check out this blog to get up to speed), but did you know there are thousands of others out there? In 2018 we will continue to see the cryptocurrency market expand as more and more innovators experiment with decentralized networks. A decentralized network is one in which users handle their own transactions in a secure, transparent environment. Every bit of data has been vetted by multiple sources and is encrypted for extra security. You will also start hearing more about ICOs, or initial coin offerings. While the ICO name is a hat tip to traditional business IPOs, the structure of an ICO is a little different. For a short time, consumers can purchase units of a brand-new currency in exchange for more well-known currencies like Bitcoin. This is a great way to get in on the ground floor of a promising new currency, but it can also be risky. Be on the lookout, too, for how the government will get involved in cryptocurrencies. There will be rules and regulations — such things are inevitable in the financial industry. It remains to be seen how these will impact growth in 2018. Big Leaps Forward in Payment Security These days it can feel like your personal information is less secure than ever, with companies from Chipotle to Gamestop falling victim to hacks and breaches. There’s a critical need for security technology to get back ahead of the bad guys, particularly with respect to mobile payments. One way the industry is doing this is through biometrics. You can already unlock your phone with your thumbprint, and soon you’ll be able to pay that way as well. No more lengthy passwords for your digital wallets — just press your thumb on the reader and be on your way. Iris scanners aren’t far behind, either! Digital Wallets Get Even Better Digital wallets have already had a profound impact on how we shop and spend. They’re quick, easy, and extremely portable — and more and more businesses are starting to accept them. In 2018, look for digital wallets to make a big impact on the international payments scene as well. Say goodbye to costly and cumbersome currency exchanges while traveling abroad. Apps and software are already on the market to allow you to buy, use, and reconvert foreign currency without even the use of a credit card. Adoption won’t happen overnight, but once it does we’ll be looking at a whole new era in mobile payments. Robots Managing Your 401(k) Your financial advisor may be out of a job sooner rather than later. AI-powered robo-advisors are programmed to select and manage investments and already provide an efficient, low-cost option for many investors. Able to set up and start managing a portfolio in minutes based on risk assessment algorithms and mathematical formulas, robo-advisors are doing more and more and human advisors are finding it hard to keep up. We can expect to see even more growth in automated advising in 2018. Stay Up to Date on What’s Happening There’s going to be a lot going on next year, and you’re going to want to stay informed. Work with a credit card processing partner who is on the cutting edge and who takes your education seriously. 360 Payments fits the bill. We’re always learning about the latest trends, and we’ll keep you informed about them all year long. Give us a call at 1–855–360–0360 or drop us a line on our website to learn more. PS — Read up on what the current tax reform bill would mean for your business. PPS — Holiday shopping chaos got you down? Check out our guide to staying sane.

Four Fintech Trends to Watch in 2018

0

four-fintech-trends-to-watch-in-2018-16ae50a33874

2017-12-14

2017-12-14 15:16:33

https://medium.com/s/story/four-fintech-trends-to-watch-in-2018-16ae50a33874

false

704

null

2018

8,179

360 Payments

null

dfcb83d0f9e2

360Payments

182

486

20,181,104

null

0

null

0

21ea2b7f728c

2017-12-19

2017-12-19 07:19:29

2017-12-19

2017-12-19 07:40:51

9

false

en

2017-12-20

2017-12-20 17:54:28

11

16aea8bfbe33

3.962264

49

0

Last few weeks has been very exciting for us. Amazon introduced the first deep learning enabled video camera — DeepLens. Google announced…

5

The 3 trendiest AI kits in 2017 — A quick guide of Google Vision kit, DeepLens & BerryNet Last few weeks has been very exciting for us. Amazon introduced the first deep learning enabled video camera — DeepLens. Google announced their latest AIY Project — the Vision Kit. At DT42, we always believe that bringing deep learning to edge devices is the key towards the future. We also believe AI technology should not be dominated only by big tech giants, but readily available for everyone. That is the reason why we released BerryNet[1] project half a year ago. BerryNet is the first AI Gateway FLOSS Project to release the power of AI on edge devices. With these 3 latest awesome AI edge vision powerups — or I should say toys, you can build up your very own project using AI to solve your problem in your life. Let’s say if you want to build a monkey alert camera to prevent monkeys messing up with your backyard and eating up all your fruit. What’s the steps you need to take by using Google AIY Vision Kit, DeepLens or BerryNet? Here we want to make a short instruction. Figure 1. Monkey alarm system Figure 2 gives a brief illustration of the equipment and software that you will engage with using different tools. Figure 2. Major components of the Monkey alarm system The whole Monkey alarm system includes five major components:Data receiver: a camera (a) Data receiver: a camera (b) Computation hardware: key hardware component for tensor computation (c) Software system: including the deep learning libraries and the operating system running on local hardware. (d) AI model: the deep learning model used for analyzing input data (e) Alarm trigger system: deliver the detection results to users Next we will explain more of the steps using the three tools separately. Google AIY vision kit Figure 3. Components of the system using vision kit Hardware you need to prepare: Pi camera 2 (a) , Vision kit (b), Raspberry Pi zero w. Steps: 1 — Assemble the kit following the instructions from AIY Project website[2], and load the image (3)to SD card. 2 — Train a deep learning model as a monkey detector (d) and compile it, 3 — Load trained model to VisionBonnet to build a monkey detection 4 — Use the SDK to build alarm trigger (e) and control it via Android App. In the case that the object you want to detect is already bundled with the image, you can simply skip step 2. AWS DeepLens Figure 4. Components of the system using DeepLens Hardware you need to prepare: AWS DeepLens, this includes components (a), (b) and (c) Steps: 1 — Register, connect and set up DeepLens online. 2 — Use AWS SageMaker to train a monkey detection model (d). 2.1 Create a “monkey detection project” on the DeepLens console 2.2 Import the model trained in step 2.1, and deploy the project to DeepLens 3 — Use AWS Management Console to build alarm trigger (e). By using AWS DeepLens, unlike the other two kits, you don’t need to prepare all the hardware yourself. However, this also limited the flexibility too. BerryNet Figure 5. Components of the system using BerryNet Hardware you need to prepare: Raspberry Pi 3 (b), an IP/Nest/Pi camera (a). You can also purchase Movidius Neural Compute Sticker for better inference performance. Steps: 1 — Train a deep learning model as a money detector (d) 2 — Install and configure BerryNet (c) with the trained model on Raspberry Pi 3 — Setup input client as data receiver (can be a pi camera, an IP camera or even a Nest camera), and output client as alarm trigger. Currently, the model training requires users to setup environment manually. For example, following the YOLO website[3] to train the monkey detector. A new easy-to-use service, Epeuva[4] is coming soon to help customers train the model. Click to register for the early invitation. On Epeuva, you can bring your own data and customized AI models without any coding effort. By repeating step 1 users are easily to build a detection system they want. We envision a world where deep learning and AI will be democratized for everyone and every device. The BerryNet project is licensed with GPL because we want to take AI from the ivory towers and make it accessible for all. Computing has gone in massive cycles, shifting from centralized to distributed and back again. We believe edge AI is the key for developing more and more useful applications in the near future. [1] https://github.com/DT42/BerryNet [2]https://aiyprojects.withgoogle.com/vision#assembly-guide-7-now-what [3]https://pjreddie.com/darknet/yolo/ [4]http://www.dt42.io/epeuva/index.html#contact-section

The 3 trendiest AI kits in 2017 — A quick guide of Google Vision kit, DeepLens & BerryNet

454

which-ai-kit-fits-your-needs-the-best-google-aiy-vision-kit-deeplens-berrynet-16aea8bfbe33

2018-06-11

2018-06-11 22:30:16

https://medium.com/s/story/which-ai-kit-fits-your-needs-the-best-google-aiy-vision-kit-deeplens-berrynet-16aea8bfbe33

false

732

Intelligence of Things on edge device

null

DT42

null

DT42

contact@dt42.io

dt42

DEEP LEARNING,ARTIFICIAL INTELLIGENCE,IOT,STARTUP,COMPUTER VISION

dt42io

Raspberry Pi

raspberry-pi

Raspberry Pi

4,517

DT42

null

d603c46d3f56

dt42

179

14

20,181,104

null

0

null

0

null

2018-07-09

2018-07-09 15:30:07

2018-07-09

2018-07-09 16:54:26

1

false

en

2018-07-09

2018-07-09 16:54:26

1

16af58bceb4d

0.566038

1

0

Cognitive futures and my journey from design to AI

3

My Podcast Interview with MEX Cognitive futures and my journey from design to AI A few months back I did an interview with MEX founder Marek Pawlowski (thanks, Marek!) to talk off-the-cuff about cognitive experiences, the challenges of developing dynamic interfaces, and the future of conversational UI. He also dug into my background a bit, and what I look for when hiring new designers. Enjoy! Cognitive UX; Jennifer Sukis, Creative Director, IBM Watson Jennifer Sukis is a Watson AI Practices Design Principal at IBM based in Austin, TX. The above article is personal and does not necessarily represent IBM’s positions, strategies or opinions.

My Podcast Interview with MEX

1

my-podcast-interview-with-mex-16af58bceb4d

2018-07-09

2018-07-09 16:54:26

https://medium.com/s/story/my-podcast-interview-with-mex-16af58bceb4d

false

97

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Jennifer Sukis

Design Principal for AI & Machine Learning at IBM. Professor of Advanced Design for AI at the University of Texas.

be85714f1ba8

jennifer.sukis

242

89

20,181,104

null

0

null

0

9b07dffed261

2018-01-06

2018-01-06 14:02:33

2018-01-06

2018-01-06 16:09:36

1

false

en

2018-01-06

2018-01-06 16:09:36

5

16b04fb31ad3

1.369811

2

0

This post is one of a series of posts I am writing where I try to apply the Feynman Technique to a number of different crypto token…

5

Numeraire overview using the Feynman Technique This post is one of a series of posts I am writing where I try to apply the Feynman Technique to a number of different crypto token projects. You can find the rest here. None of this should be taken as investment advice. They are just meant to be simple introductions to the projects. Numeraire is the token used on the Numerai platform. Numerai is a platform that hosts machine learning competitions between data scientists from around the world. The Numeraire is being used on the Numerai platform to better understand each data scientist’s confidence level in their machine learning model. Each player in a Numerai competition has the ability to bet a certain amount of Numeraire on whether or not their model will meet a certain threshold of accuracy. If their model does well in the competition, they keep their Numeraire and receive a given dollar amount as a reward. This reward amount is determined by taking the amount of Numeraire they bet and dividing it by the players’s confidence level (a range of 0–1). If their model does not meet the minimum performance required, they lose all of the Numeraire they bet and don’t receive any award. In order to prevent situations where a player is motivated to bet a lot of Numeraire but falsely indicates a low confidence level, to maximize their returns, Numerai pays out rewards starting with the player who indicates the highest confidence level. Each player’s model that meets the performance level required gets rewarded until the total reward amount is 0. Therefore, if a player indicates a low confidence level, it is likely they will not receive a reward since other players with higher confidence levels will be able to receive it before them. This should motivate players to indicate accurate confidence levels. Further reading: https://numer.ai/ https://numer.ai/whitepaper.pdf

Numeraire overview using the Feynman Technique

2

numeraire-overview-using-the-feynman-technique-16b04fb31ad3

2018-07-16

2018-07-16 14:16:38

https://medium.com/s/story/numeraire-overview-using-the-feynman-technique-16b04fb31ad3

false

310

A look at crypto projects in an effort to better understand them.

null

Understanding Crypto

null

understanding-crypto

CRYPTO,CRYPTOCURRENCY,BLOCKCHAIN,CRYPTOASSET,CRYPTO TOKENS

null

Data Science

data-science

Data Science

33,617

Alex Meyer

Attempting to put my dent in the universe.

d19eba0196e0

alexcmeyer

274

391

20,181,104

null

0

null

0

null

2018-03-13

2018-03-13 18:21:43

2018-03-13

2018-03-13 18:26:52

0

false

en

2018-03-13

2018-03-13 18:26:52

0

16b2bc06055d

0.449057

1

0

Insights from DATA & SOCIAL GOOD book:

1

Data for social good. Insights from DATA & SOCIAL GOOD book: Besides using data to improve the corporate world,applying data science principles & skills to improve lives of ordinary people , make a positive impact & world a better place seems a logical idea. What is your social good as a data scientist?Data should be for social good. Using data to elevate human conditions won’t happen by accident ,people have to envision it,develop the routine processes & underlying frameworks to make it practical then dedicate the time and energy to make it work. As data scientists we should start focusing on how we can positively impact the communities around us with our expertise.That’s now the beauty of data science.

Data for social good.

1

data-for-social-good-16b2bc06055d

2018-03-14

2018-03-14 06:19:26

https://medium.com/s/story/data-for-social-good-16b2bc06055d

false

119

null

Data Science

data-science

Data Science

33,617

Brian Mwangi.

Data Analyst ¦¦ ML enthusiast ¦¦ Turning data into smart and informed decisions ¦¦Positive attitude~vibes~life.

4ae3467628ee

itsbrianmwangi

60

160

20,181,104

null

0

I come from New York . O O O B-loc I-loc O I come from Taiwan . O O O B-loc O P(y|x)=P(y)P(x|y)/P(x) s(tag=n',t) 表示的是第t步時以tag為n'的最大的序列概率 s(tag=j,t)= max{s(tag=i,t-1) * TP(i,j,t)|i=1...N} TP(i,j,t)為第t－1步從狀態i跳到第t步的狀態j的概率，假設狀態轉移跟time step無關簡化為 s(tag=j,n)= max{s(tag=i,t-1) * TP(i,j)|i=1...N} TP表示Transition Probability 取log後*變加 s(tag=j,t)= max{log[s(tag=i,t-1)] + log[TP(i,j)]|i=1...N} 我們要的只是個排序，實際值不重要，把log去掉 s(tag=j,t)= max{s(tag=i,t-1) + TP(i,j)|i=1...N} trellis = np.zeros_like(score) backpointers = np.zeros_like(score, dtype=np.int32) trellis[0] = score[0] for t in range(1, score.shape[0]): v = np.expand_dims(trellis[t - 1], 1) + transition_params trellis[t] = score[t] + np.max(v, 0) # trellis[t] 為上面的[s(tag=i,t),...s(tag=N,t)] backpointers[t] = np.argmax(v, 0) # backpointers 紀錄了之後要回朔時的N

10

76d582de2a0e

2018-08-27

2018-08-27 07:01:16

2018-08-27

2018-08-27 06:47:54

8

false

zh-Hant

2018-08-28

2018-08-28 08:08:07

1

16b3a7b9f42e

1.767296

0

這邊寫下之前讀到的筆記，基本上資料來源是李弘毅老師的投影片 Structured Learning: Sequence Labeling

4

HMM與CRF應用在NER任務這邊寫下之前讀到的筆記，基本上資料來源是李弘毅老師的投影片 Structured Learning: Sequence Labeling (http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Lecture/Sequence.pdf) 先簡單講一下 named entity recognition(ner) 這個任務最後希望能做到的事，假設有一個句子我們會希望對應的標註是也就是把 New York 標示成 B-loc I-loc，這裡的B表示開頭，I表示之後的字，loc代表自己定義entity的類別，如果像是單個字就成詞，那就只有B-loc 你可以自訂所需要的entity，基本上在不同場景所需要的entity是不一樣的，如果要製作一個任務導向的bot，這些entity就是bot執行任務時所需要的參數基本上如果任務簡單或訓練資料量很少，用正則表達式或直接比對資料庫的方式去抓就好，這兩個方式基本上就是有建資料就不會錯，但使用者如果輸入一些奇怪、否定或文法不通的句子故意測試時，我們還是會希望抓到 entity 嗎?或是當使用者剛好講到沒在資料庫的東西時又該怎麼辦？基本上資料庫越大維護的成本也越高如果我們訓練資料量夠多的話就可以用以下的方法HMM或CRF，這類問題基本上都可以寫成3個步驟 1.Evaluation:定義F(x,y)，x代表輸入序列，y代表輸出序列，F(x,y)代表好壞程度，值越大代表y越符合我們的需要 2.Inference:在所有可能的y集合裡找到一組y能最大化F(x,y)的值 3.如何Training，這裡我們不討論，有興趣請看老師的課程網頁 HMM的的算法基本上就是聯合機率，在這裡F(x,y)=P(y|x)，而P(x,y)=P(y)P(x|y)=P(x)P(y|x)，移項後就得到了基本上P(x)在給定句子後是值固定的，所以在第2步驟Inference時我們就直接當它是1，這不影響最後的結果在老師的課程裡任務是要標註詞性 P(y)裡的每一項可以由訓練數據統計得到，這裡我們會得到一個N維向量代表Start Probability，N*N的矩陣代表Transition Probability，P(x|y)是有了這個標註後，產生這個詞的機率，也可以由統計後得到，寫成N*M的矩陣代表Emission Probability，N代表有多少種tag，M代表詞彙量，這裡會有個實際應用的問題上述有些機率值如果在訓練資料裡沒有出現過，那就會是0，這時就要想一下是真的不可能還是剛好沒有出現在訓練資料裡，如果只是剛好沒出現那就要將其變為一個很小的值第二個問題Inference時要窮舉所有的y，用 Viterbi 演算法要注意的是HMM可能會得出一組完全沒看過的P(y|x)是高分，這個現象有好有壞，當訓練資料本身是少量時可能會有益，如果想要解決這問題可以用更複雜的模型一樣我們要先定義F(x,y)，這裡一樣寫成P(x,y) Φ(x,y)代表特徵向量，是要自己定義的，不用訓練，w 是機器學習得出的權重向量，因為能自訂任何特徵向量，這也是比HMM強的地方，老師的課程裡有推導如何把HMM寫成CRF的形式，有興趣的人能去看看第二個問題Inference時要窮舉所有的y，一樣用 Viterbi 演算法但是如果一般人想用CRF時就會遇到一個困擾，Φ(x,y)到底要如何定義才好？這裡我們可以用Bidirectiona LSTM或Self Attention等讓機器自己學出來就好最後來說下Viterbi 演算法，如果直接求所有可能的y的話，會有N^seq_len個組合，當seq_len比N大很多時，這計算量大大了，Viterbi 演算法複雜度為seq_len*(N²)，Viterbi 演算法基礎可以概括為下面3點如果概率最大的路徑經過 Directed Acyclic Graph 中的某點，則從起始點到該點的子路徑也一定是從起始點到該點路徑中概率最大的假定第i時刻有N個狀態，從開始到第i時刻的各狀態n各自有最短路徑，而最終的最短路徑必然經過其中之一根據上述，在計算第i+1某個狀態n’的最短路徑時，只需要考慮當前N個狀態各自最短的路徑和第n個狀態值到i+1 的n’狀態值最短路徑即可 NN 出來的是 seq_len*N的矩陣(score)，另外還有一個額外的 N*N Transition Probability矩陣TP(transition_params) 一樣定義F 以下為用例子示範一下此算法先把TP加上score[0]得到v，v的直行取max代表到各tag的最大分數，寫為np.max(v,0)，同時記錄下此時是從哪裡出發backpointers[1]，再把np.max(v,0)+score[1]當作下次各tag的起始分數同樣的操作依最高分去回朔路徑 tensorflow中 viterbi_decode 其中一段可以對照一下

HMM與CRF應用在NER任務

0

hmm與crf應用在ner任務-16b3a7b9f42e

2018-08-28

2018-08-28 08:08:07

https://medium.com/s/story/hmm與crf應用在ner任務-16b3a7b9f42e

false

168

Share Data science, machine learning and deep learning

null

Data Scientists Playground

null

data-scientists-playground

DATA ANALYTICS,DATA SCIENCE,MACHINE LEARNING,DEEP LEARNING,DATA ANALYSIS

null

Machine Learning

machine-learning

Machine Learning

51,320

任書瑋

null

5032a07a5fd0

billy800413

2

1

20,181,104

null

0

null

0

null

2018-05-18

2018-05-18 08:39:03

2018-05-18

2018-05-18 08:41:11

1

false

en

2018-05-18

2018-05-18 08:51:08

0

16b3d2d430ad

6.958491

0

The news that Cambridge Analytica is to fold has brought relief in many quarters. But were we right to fear it in the first place?

4

The biggest threat from big data? Human predictability The news that Cambridge Analytica is to fold has brought relief in many quarters. But were we right to fear it in the first place? What unsettled us about Cambridge Analytica was the glimpse of ourselves through the predatory eye of its business model. We were nothing but a malleable herd. Were we right to be worried? Cambridge Analytica offered clients nano-targeting using statistical correlations between our social habits and our political leanings. Perhaps Labour voters like cats and Tories dogs; maybe LibDem supporters drink green tea and UKIP builders’ brew. Who knows? But it works along these lines. We can instinctively accept there may be something in this: we appreciate the internet’s power and reach as an introduction agency, enabling us to track down birds of a feather for even the most esoteric of interests; and we accept that, by the digital traces we leave, we can be tracked down by others too. Whatever lengths we may go to as individuals to hide our preferences both social and political, our group identities may still give us away. While nano-targeting may have its uses, it is not obvious that it makes us any more biddable. After all, it is not a clever way of reaching a particular herd, it is a clever way to reach it without having to reach anyone else in the process. It is like a neon sign in Times Square, except that what it shows me is not the same as what it shows you. While I am being ‘targeted’, the platform is selling the same space a thousand times over to other nano-targeters. This — not the chance to rig elections — is the rainbow’s-end pot of gold for its promoters. So can we be influenced even if we can be targeted? This is where some want to get off the bus. Of course, nothing makes the advertising industry happier — or richer — than our conceit that we are impervious to their blandishments. We are not — and what makes us susceptible is something that happens to us instinctively and automatically. This ‘something’ is a consequence not so much of being a human as of being a mammal. All mammals share the same basic emotional drive systems, one of which is the ‘social orientation’ system. This is what prompts flocking, swarming, shoaling and herding behaviour in the natural world. We may have the physical experience of flocking when, say, marching or dancing in step or moving in convoy but, mainly, it is the psychological experience that we can recognise. That humans have a group response in social situations is something we are barely aware of, partly because we are so intent on seeing ourselves as autonomous individuals guided by a sovereign ego. But, as with advertising, the more we deny that we are susceptible to external influences, the more we leave the back door open to them. Flocking follows a complex pattern of behaviour with first, second and third order characteristics. The flight formations of flocking birds has been reproduced by computer modelling based on only three behavioural rules — separation, alignment and cohesion. These are the axes of our social conformity. What they direct is survival behaviour, not a life strategy; we flock as a defensive manoeuvre to find safety in numbers, security in belonging, or anonymity in a crowd. The first rule of flocking, separation, involves staying close to one’s neighbours but not so close as to invade their personal space. In psychological terms, it involves mental attunement or rapport with colleagues. This is our way of ensuring that we are ‘on the same page’ and so stand a chance of productive collaboration. The second rule of flocking, alignment, has to do with the aim or goal of the activity. In physical terms, it involves keeping on a course that is the average of one’s neighbours. In psychological terms, it means having common objectives. The third rule of flocking, cohesion, involves maintaining a steady speed and distance from the centre of the flock. In psychological terms, it means staying grounded in a shared reality, to be sure that everyone pulls their weight in the endeavour. The first order characteristics of flocking are evident in conformist behaviour — for example, our following of fashion, fads or convention — when we seek safety in the majority opinion, whatever it might be. Its second and third order characteristics are recognisable in situations when a group hits a difficult situation together — when we encounter difference, indecision or a sense of impossibility. Difference causes second order separation, resulting in disengagement from relationships with those around us. Instead of continuing to invest emotional energy in seeking common ground with those from or with whom we differ, we become less cooperative and accommodating. In abandoning mutuality and reciprocity, we retreat to a private world of self-interest and self-absorption. We either lose interest in and empathy for others as individuals or we try to convince them we are right without listening to their point of view. The difficulties of the here-and-now dominate relations, as we lose sight of the benefits gained from collaboration in the past or to be gained in the future. The third order response is polarisation: in-groups and out-groups form around the unresolved difference, with viewpoints becoming ever more entrenched and dogmatic. The certainty and unity of the in-group provides comfort and belonging, causing the group to idealise itself and view all that is bad and wrong as belonging to the out-group. These two groups become unable to function except by expressing difference with the other, giving them a paradoxical interdependence. To influence this group, all that is needed is to promote the idea that they are right and others wrong about an issue of importance. They will respond to suggestions that other identified out-groups are ‘to blame’ for the problems they face, while they themselves are the injured parties. They will like the simplicity of a win-lose, right-wrong narrative rather than a version that suggests there can be two sides to the story. Their motivation is to avoid responsibility, so they look for a leader who will take their side and accept no compromise. Indecision causes second order alignment, resulting in disengagement from purpose: since no course of action can be chosen, it is only possible to achieve short-term, narrow goals that are not stymied by the ‘undecidable’ issue. As little or no progress can be made towards the principal objective, discussions start to focus on process rather than end-product. Planning becomes impossible as the focus turns to the past and what has caused the current impasse; meanwhile, all sense of a shared vision of the future end-state is lost. Paralysis sets in as, while no positive choices can be made, myriad reasons can still be advanced to prevent suggestions from being taken up. The third order response is stalemate, as there is little point in trying to get something done if no one else will cooperate. This group can be influenced by the promises of leaders to do everything for them, perhaps proposing ‘more funding’ as a cure-all for intractable problems and more state intervention to help those not taking responsibility for themselves. This group has given up thinking because it feels easier to leave solutions to others. They feel helpless and disempowered, but in many ways are content to remain that way by choosing a leader who will excuse their inaction and do their dirty work for them. Impossibility causes second order cohesion. This leads to a false alignment with external reality as wishful thinking overtakes practical action. There is a turning inwards so that the ‘unsolvable’ problem can be tackled in isolation, divorced from its unhelpful context of collateral impacts. Attention is instead focused on a future solution not requiring action from the group. Complexity is abandoned; an ideological fervour can develop around the theoretical panacea, which can never be tested and is always left to others to implement. This way of thinking becomes addictive, with ever-more ‘solutions’ found whose only impediment is the lack of someone to implement them. The third order response is dogmatism: imposing the ideas with no acknowledgement of their limitations nor allowance for exceptions, until they eventually collapse under the weight of their own impossibility. This group can be engaged by grand visions and cure-all solutions. They will readily pin their faith in a ‘big idea’ that promises to make the world a better place. Often, this involves the championing of a single issue without regard for competing priorities or unintended consequences. So convinced are they by the ideal to which they subscribe, they indulge in virtue-signalling and self-righteous admonishments to non-believers, adopting a ‘closed system’ ideological perspective on truth and reality. So much of what is happening in society can be viewed as second and third order flocking. It is not the work of an algorithm; we do it to and by ourselves. It does not require artificial intelligence, merely an unwillingness to think for ourselves when we encounter difference, indecision or impossibility. So perhaps we should not make technology the scapegoat when the culprit is rather closer to home. What we are experiencing is some turbulence produced by our sense of being vulnerable to attack. Instead of enquiring further, we search for evidence of undue influence on the outside. That, after all, is what groups do when they encounter difference in the shape of the unknown. There are many problems in society, and many instances of difference, indecision and impossibility. As soon as we fold our arms and stop looking for answers within ourselves, we are inviting tenders for the supply of snake oil. What Cambridge Analytica showed us was the size and reach of the channel through which it can now be distributed, and the number of appetising flavours on offer. We can put ourselves beyond the reach of big data by quitting the internet for a life of digital hermitry but our addiction to snake oil won’t end with it. The only way to wean ourselves off it is to remain open-minded to everything and everyone as much and for as long as possible — to keep listening and thinking, learning and adapting. This means exploring deep-rooted differences rather than shrinking from them, confronting tough decisions rather than avoiding them, and accepting practical reality rather than turning a blind eye. If we don’t, we will flock and swarm and herd into the clutches of any politician who understands how groups work — with or without the data to prove it.

The biggest threat from big data? Human predictability

0

the-biggest-threat-from-big-data-human-predictability-16b3d2d430ad

2018-05-18

2018-05-18 08:51:09

https://medium.com/s/story/the-biggest-threat-from-big-data-human-predictability-16b3d2d430ad

false

1,791

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Sandy Henderson

null

ae0ea409b41f

sandyhenderson

1

3

20,181,104

null

0

pip install tensorboardcolab tbc=TensorBoardColab() model.fit(x,y,epochs=100000,callbacks=[TensorBoardColabCallback(tbc)]) tbc.save_image(title=”test_title”, image=image) tbc.save_value(“graph_name”, “line_name”, epoch, value) tbc.save_value(“graph_name2”, “line_name2”, epoch, value2) . . . tbc.flush_line(line_name) tbc.flush_line(line_name2) . . tbc.close()

8

null

2018-06-24

2018-06-24 07:30:26

2018-06-24

2018-06-24 07:53:50

2

false

en

2018-06-30

2018-06-30 08:49:17

5

16b4bb9812a6

1.228616

19

3

0

Google Colab provides free GPU acceleration for Machine Learning. During neural network training, we wanna see training progress and…

4

Use Tensorboard in Google Colab Google Colab provides free GPU acceleration for Machine Learning. During neural network training, we wanna see training progress and result. Using Tensorboard is convenient way to visualize the progress. This tutorial will show you how to do so in the most simplest way. Progress of neural network training Principle Run Tensorboard in the background of Google Colab, use ngrok to tunnel traffic to localhost, so that it can be accessed outside Google Colab. Detail of code can be found in this stackoverflow. Credit to Joope 🙏🏻 Feel little bit complicated? Here is a library to do it easier Installation Initialization After initialization, TensorBoard link will be shown in Colab Google Juyter output. Colab Google Juyter output Done! One line of code can start up TensorBoard in Google Colab, easy, right? :) But how to use it? Read below … Add to Keras callback That is all! Now you can read the progress graph Oh! Wait, you want more control? Easy, read below … Save picture to TensorBoard Save a value to graph of TensorBoard Hope you enjoyed! Reference Can I use Tensorboard with Google Colab? Here is how you can display your models inline on Google Colab.stackoverflow.com About Me LinkedIn: https://www.linkedin.com/in/tommy-tao-a5961156/ Twitter: https://twitter.com/tommytao

Use Tensorboard in Google Colab

142

use-tensorboard-in-google-colab-16b4bb9812a6

2018-06-30

2018-06-30 08:49:17

https://medium.com/s/story/use-tensorboard-in-google-colab-16b4bb9812a6

false

224

null

Machine Learning

machine-learning

Machine Learning

51,320

Tommy Tao

null

5bf8d414b412

tommytao_54597

33

42

20,181,104

null

0

null

0

null

2018-08-18

2018-08-18 06:25:58

2018-09-26

2018-09-26 02:53:41

16

false

en

2018-09-26

2018-09-26 02:53:41

47

16b4dd913a1f

15.448113

280

6

2

Moving Beyond model.fit(X, y)

2

Getting Better at Machine Learning Moving Beyond model.fit(X, y) Image credit: Getting better at machine learning takes time, effort, and practice! Motivation In A Beginner’s Guide to Data Engineering series, I argued that academic institutions typically do not teach students the proper mental models when it comes to analytics workflows in real-life. Far too many classes only focus on the mechanics of data analysis without teaching concepts such as ETL or the importance of building robust data pipelines. Unfortunately, I see a similar pattern in machine learning education as well. Surely, studying the math behind ML and learning different algorithms are valuable. Yet, there exist crucial steps beyond model.fit(X, y) that are important in practice. In this post, I will share some of my learnings that I did not learn in school. First, I will highlight the rise of Kaggle: why it has transformed our industry, what critical role it plays, but also where it fell short. In particular, I will contrast the workflow Kaggle reinforced with the typical development workflow of a real-life machine learning project. Throughout the post, I will give coloring examples around topics such as problem definition, feature engineering, model debugging, productionization, and feedback loops. By the end of this post, I hope readers will appreciate some of the complexity and challenges, but also joys, of real-life machine learning. The Rise of Kaggle Competition Kaggle’s Competition Landing Page: Fancy to Join one? Ever since its inception in 2010, Kaggle has become the platform where data enthusiasts around the world compete to solve a wide variety of problems using machine learning. Over time, Kaggle has built an incredible repository of useful benchmark datasets and example notebooks (called kernels), turned modeling into a sport, and made some practitioners into Kaggle stars. Common Task Framework The model that Kaggle follows, is what Professor David Donoho referred to as the Common Task Framework (CTF) in his paper “50 years of Data Science”. Donoho argued that the secret sauce to machine learning’s success is partially driven by competitions: It is no exaggeration to say that the combination of a predictive modeling culture together with CTF is the ‘secret sauce’ of machine learning. This combination leads directly to a total focus on optimization of empirical performance, which […] allows large numbers of researchers to compete at any given common task challenge, and allows for efficient […] judging of challenge winners. Indeed, from DARPA’s machine translation research in the 1980s, the famous 2009 Netflix Prize, to the recent success of deep learning due to ImageNet challenge, the machine learning community continues to bring innovation to the mass under common task framework. Where Kaggle Competition Fall Short While Kaggle competitions have been tremendously educational, their workflows generally only reflect a small subset of what is involved in real-life machine learning projects. First of all, Kaggle hosts formulate the problems, not the participants. Not only are the loss functions and golden datasets used for evaluation pre-determined, but training labels / data are often handed to the participants on a silver platter. Furthermore, there is very little concern regarding how to integrate models into a decision process or a product. These are all conditions real projects are unlikely to meet in practice. As a result, a lot of the considerations in machine learning projects are lost in translation. Machine Learning Workflow When building a machine learning product, we are no longer developing models in isolation (what people sometimes called “Laptop Data Science”). Rather, we are building a system that interacts with real human beings. Not only do we need to think strategically about the problems we are solving for the end users, but we also need to ensure that the user experience is intuitive, predictions are accurate, and inference is efficient. Think and Build a System End-to-End These conditions mean that we almost never jump to modeling immediately, we have to think and build the system end-to-end. In Rachel Thomas’ fantastic post “What do machine learning practitioners actually do?”, she explains the typical workflow of a machine learning project: Building a machine learning product is a multi-faceted and complex task […] machine learning practitioners may need to do during the process: understanding the context, preparing the data, building the model, productionization, and monitoring. […] Certainly, not every machine learning practitioner needs to do all of the above steps, but components of this process will be a part of many machine learning applications. Kaggle is an amazing platform that focuses on model building, but less so on the rest of the steps described above. To help readers to better understand these other topics, I will highlight them using a combination of my personal experience and illuminating examples from other companies that I find useful. Below, I will discuss: Problem Definition: why thinking hard about your problem is crucial Data Collection: why setting up your {X, y} right is half of the job done Model Building: how to debug your model when it does not perform well Productionization: what “putting model into production” really means Feedback Loop: how unintended feedback loop can affect your system 1. Defining Problem Is Hard and Not Always Obvious Image source: Do you think this house can be an Airbnb Plus Home? Let’s start with a case study, Airbnb Plus, a product whose mission is to bring high-quality homes to the Airbnb marketplace. While many employees are passionate about finding homes suitable for Plus, doing this at scale can be challenging. On our team, we use a combination of human evaluation and machine learning to identify high potential homes. This type of problem, which involves human evaluations + machine predictions, are becoming increasingly common. Your First Iteration of the Model Is Often Not Your Last As our human evaluators assess homes, training labels are generated as a by-product. Given that we already have a lot of features about a listing (price, bookings, reviews … etc), it was rather convenient to combine the two data sources (labels + features) to train our first home targeting model. At first, this approach worked well, and it brought enormous gains to our efficiency. However, as the product continued to evolve, we started to experience the limitation of this simple approach. Specifically, as we evolved what qualifies a home to be “Plus” at the program level, the semantic meanings of our outcome labels had also changed. This business evolution posed non-trivial challenges to our learning task because our labels can become outdated rather quickly. We were essentially learning to predict a moving target! Decomposing the Learning Task Requires Thinking To de-risk our modeling effort, we had to re-think the problem formulation. Eventually, we decided to decompose our single, monolithic learning task into several independent modular tasks. This means that instead of directly classifying whether a listing is high potential or not, we focused on predicting more stable attributes that are indicative of high-quality. For example, instead of classifying the label is_home_high_potential directly, we framed the problem as is_home_high_potential = f(style, design, ...) , where f is our rule-based approach to codify how humans might use these modular predictions to make a final assessment. Image credit: It’s useful to decompose a learning task into smaller tasks when the label is not straightforward More often than not, problem formulation requires deep domain knowledge, the ability to decompose problems, and a lot of patience. The most convenient training dataset should not drive how we formulate the problem, rather it should be the other way around. This is an important first skill to becoming an effective problem solver in machine learning. Takeaway: Like software engineering, the principle of decomposition can be very important in machine learning as well. It allows us to break a complex problem or system into parts that are easier to conceive, understand, and learn. 2. Data Collection Is Often Non-trivial Image source: Data collection for machine learning is analogous to picking ingredients before cooking a great dish At work, our data scientists and ML engineers often get together to talk about machine learning ideas passionately. While these discussions are always inspirational, they generally do not translate to project roadmaps immediately due to a common blocker — lack of training labels and feature pipelines. Acquiring Quality Labels Is Challenging On Airbnb Plus, we are lucky to have training labels generated as a by-product from our home assessments, but dedicated labelings are often rare because collecting them comes with a hefty time and monetary cost. In the absence of actual training labels, we could use other data as proxies for training labels, but they are not always high fidelity. For example, when Airbnb developed its room classification model, we used image captions as our proxy label for the ground truth. While this approach gave us convenient labels as a head start, the label quality tends to be low for certain room types, especially for smaller spaces. For instance, scenes in a studio tend to be crammed together: a kitchen could be right next to a living room that is adjacent to a bedroom. This makes evaluation of ground truth hard to interpret, sometimes even in the eyes of human labelers. Image source: Should we label the room type of this image as a kitchen or a bedroom? In general, labeling in real-life is far trickier than simple tasks like telling apart hotdogs v.s. non-hotdogs. This nuance often makes inter-rater agreement hard to achieve, and is rather universal for serious modeling pursuits across different domains. Andrej Karpathy, in his talk building the software 2.0 stack, highlights some of Tesla’s labeling challenges for building self-driving cars. For example, he explains that labeling traffic lights and traffic lanes can sound simple but in reality difficult because the diversity of how different cities design the roads. More generally, he argues that in the software 2.0 world, we have not yet figure out the right IDEs or labeling tools to build software. These are all real data challenges that are not taught in school. Building Feature Pipelines Is Time Consuming Even when we have high quality labels, building feature pipelines can be a tedious and time-consuming process. For the model described in the previous section, we were lucky to re-use some of the listing-level features from another existing project. For problems that involved images as input, the feature engineering work is a lot more complex. For example, before our room classification model, there was no image pipeline in place on our team that could be reused. A lot of data engineering work, from data ingestion, resizing images to size 224 x 224, using base64 encoding to storing thumbnails were required before we could build image models. Had this image pipeline not been in place, it would have slowed down a lot of our modeling work significantly. This is precisely why larger companies are building frameworks to make feature engineering easier (see Uber’s Michelangelo, Netflix’s Delorean, and Airbnb’s Zipline). When planning ML projects, it is always wise to budget time for feature engineering, because training data will not be handed to you on a silver platter. Takeaway: You need to work hard to get your training data, it is often earned rather than given. Acquiring high-quality labels can be non-trivial, and building feature pipelines can be time-consuming. To the extent you can, reuse common features or even labels to solve similar problems in the same domain space. 3. Debugging and Improving ML Models is Hard Image source: Debugging machine learning models can be a lonely pursuit Suppose you have gone through the steps of defining a problem, acquiring labels and building a feature pipeline, you then moved on to build your first iteration of the model only to learn that the result is not so stellar. What would you do in this case to debug and improve your model? Debugging Machine Learning Is Hard The scenario described above is very common and is at the heart of any machine learning project. In his post “Why is Machine Learning Hard?”, Zayd Enam pointed out that machine learning is fundamentally a hard debugging problem because there are many possible paths of exploration and unfortunately the feedback loop is generally very slow. Debugging for machine learning happens in two cases: 1) your algorithm doesn’t work, or 2) your algorithm doesn’t work well enough. What is unique about machine learning is that it is ‘exponentially’ harder to figure out what is wrong […]. There is often a delay in debugging cycles between implementing a fix or upgrade and seeing the result. Very rarely does an algorithm work the first time and so this ends up being where the time is spent. Image source: Debugging ML models are often slow and convoluted Debugging machine learning is a skill, and far too often we just try the most immediate, convenient, or “obvious” thing even though it might not be the right first thing to try. Of all the resources out there, I particularly appreciate Andrew Ng’s book Machine Learning Yearning. This approachable reference is very practical and he talks about things that I wish I had known way earlier! Some Basic Debugging Skills While I highly recommend everyone to read Andrew’s book, for the impatient, I will highlight a few tricks that I personally found to be useful in practice: Error Analysis: Learn from your model’s mistake. Specifically, hand picks 100 examples that your model got wrong from the development set and tally up the reasons why it got wrong. This can help you inspire new directions and prioritize improvement plans. Image Source: For a dog v.s. cat classifier, look at 100 misclassified examples and tally up the reasons Error analysis is important because it gives you a very data-informed view of why your model is not performing well. This is something that I used to avoid doing because of its tedious nature, but over time have really come to embrace it as it gives me a lot of insight about the data and my models’ behavior. Understand Bias-Variance: There are two major sources of error in machine learning: bias and variance. High bias often means that your model is too simple to capture the complexity of the data, and high variance indicates that you have only learned the pattern at hand. To understand bias and variance in your models, the most effective debugging tool here is to plot the learning curve. Image Source: Use the learning curve to understand if you are overfitting or underfitting When both your training error and development set error are way higher than the desired performance, you are suffering from a high bias problem (under-fitting). In such a case, increasing your model capacity or switching to a more complex algorithm is likely to help you to learn the patterns better. On the other hand, when your development set error is way higher than the training error, while the training error is relatively close to the desired performance, you are suffering the high variance problem (over-fitting). In such a case, you might want to try a simpler model, or use regularization. Alternatively, if the gap between training and development error is closing with more training examples, you might consider adding more training data to your learning task. Takeaway: debugging machine learning is hard and the feedback loop is generally slow. Instead of tackling what you think is the next obvious thing, it’s important to be more principled about debugging. Error analysis, learning curve are all good starts, and I strongly encourage you to read Andrew’s Machine Learning Yearning to improve your debugging skill. 4. Your Paths to Model Productionization Might Vary Image source: Model productionization has been talked about a lot, but what exactly does it mean? Assuming that you now have a satisfactory model to deploy, it is time to integrate your model into a decision process or product. This is what we refer to as model productionization. But what exactly does it mean? The answer depends on your use cases. Sometimes, the predictions will live outside of products completely and will be used only for strategic or business decisions. Other times, they will be an integrated part of a product experience. Not Everything is Low Latency & Context Sensitive The most useful framework I learned about this topic came from Sharath Rao, who currently leads machine learning efforts for consumer products at Instacart. In his DataEngConf talk, Sharath explains that implementation of machine learning models can usually be considered from two dimensions: Latency: How fast do the predictions need to be served to the end users? Context Sensitivity: Will we know the features ahead of inference time? Image source: Sharath Rao’s talk, “lessons from integrating ML models into data products” In the simplest case (bottom-left), for applications where predictions are mostly used for offline decisions, the model can be productionized simply as a batch scoring job. On the other hand, for models that are an integrated part of a product experience, e.g. search ranking, input features are generally not available until a user interacts with the product, and results often need to be returned really fast. In this case (top-right), online inference or real-time scoring is needed and SLA requirements are generally higher. Knowing the profile of your ML model can directly inform your implementation strategy. Revenue Prediction Model, Illustrated in Multiple Use Cases Let’s use the listing LTV model that I introduced earlier as an illustrative example. Suppose we are interested in using this model to prioritize markets to go after next year. Such an application is not consumer-facing, and we are only using the predictions for offline decision making, not in an online product. For this use case, we only need to productionize the model as an offline training, offline scoring, batch job so other data scientists can easily query the predictions from a table. Image source: How should we productionize the ML model for such a product use case? However, suppose we are now interested in showcasing the predicted host payouts in a consumer-facing product in order to inform users their financial potentials. One challenge we need to consider is how to surface the predictions within a product. In the case where contextual data is not needed, one common strategy is to store the model results as key-value pairs in a key-value store, e.g. in the form of {key: dim_market, value: revenue prediction}. For this use case, the revenue prediction can be easily looked up based on the market in which the listing is located. A more involved product might allow users to specify their location, room size, and capacity so earning potentials can be personalized. In such a use case, we will not know the features until a user enters the information, so predictions need to be computed in real-time. Depending on the use cases, your path to productionization might vary. Takeaway: Taking models to production can mean different things depending on the context, use cases, and infrastructure at the company. Having basic familiarity with concepts such as latency and context sensitivity will greatly inform your implementation strategy. 5. Feedback Loop Can Help You or Hurt You Image source: Creating and dealing with feedback loops is yet another important topic Models that are an integrated part of a product experience, or what we referred to as data products, often involve feedback loops. When done right, feedback loops can help us to create better experiences. However, feedback loops can also create unintended negative consequences, such as bias or inaccurate model performance measurements. User Feedback Can Make Your Model Better One of the most unexpected skills that I learned about real-life machine learning is the ability to spot opportunities for users to provide model feedback via product interactions. These decisions might seem only relevant to UI/UX at first, but they can actually have a profound impact on the quality of the features that the data product offers. Image source: From Xavier Amatriain’s post “10 more lessons learned from building real-life ML system” For example, Netflix decided last year to move away from the star-rating system to a thumbs up/down system, reportedly because its simplicity prompts more users to provide feedback, which in terms help Netflix to make their recommendations better. Similarly, Facebook, Twitter, Quora, and other social networks have long designed features such as likes, retweets, and comments which not only make the product more interactive, but also allow these companies to monetize better via personalization. Creating feedback opportunities in product, instrumenting and capturing these feedback, and integrating it back into model development is important for both improving user experience as well as optimizing the companies’ business objectives and bottom lines. Feedback Loops Can Also Bias Model Performance While feedback loops can be powerful, they can also have unintended, negative consequences. One important topic is that models that are biased will amplify the bias the feedback loop introduces (see here). Other times, feedback loop can affect our ability to measure model performance accurately. This latter phenomenon is best illustrated by Michael Manapat, who explains this bias based on his experience building fraud models at Stripe. In his example, he pointed out that when a live fraud model enforces certain policy (e.g. block a transaction if its fraud score is above certain threshold), the system never gets to observe the ground truth for those blocked transactions, regardless of whether they are fraudulent or not. This blind spot can affect our ability to measure the effectiveness of a model running live in production. Source: Michael Manapat’s “Counterfactual evaluation of machine learning models” from PyData Why? When obvious fraudulent transactions are blocked, the ones that remained with ground truth that we can observe are typically false negative transactions that are harder to get right. When we re-train our models on these “harder” examples, our model performance will necessarily be worse than what it really is performing in production. Michael’s solution to this bias is to inject randomness in production traffic to understand the counterfactuals. Specifically, for transactions that are deemed fraudulent, we will let a small percentage of transactions pass, regardless of their scores, so we can observe the ground truth. Using these additional labels, we can then re-adjust the calculation for model performance. This approach is simple but not entirely obvious. In fact, it took me a long while before spotting the same feedback loop in my model, and it is not until I encountered Michael’s talk that I found a solution. Takeaway: Feedback loops in machine learning models are subtle. Knowing how to leverage feedback loops can help you to build a better user experience, and being aware of feedback loops can inform you to calculate the performance of your live system more accurately. Conclusion Source: From the paper “Hidden Technical Debt in Machine Learning System” by D. Sculley et al Throughout this post, I gave concrete examples around topics such as problem definition, feature engineering, model debugging, productionization, and dealing with feedback loops. The main underlying theme here is that building a machine learning system involves a lot more nuances than just fitting a model on a laptop. While the materials that I have covered here are only a subset of the topics that one would encounter in practice, I hope that they have been informative in helping you to move beyond “Laptop Data Science”. Happy Machine Learning!

Getting Better at Machine Learning

1,777

getting-better-at-machine-learning-16b4dd913a1f

2018-09-28

2018-09-28 01:27:39

https://medium.com/s/story/getting-better-at-machine-learning-16b4dd913a1f

false

3,683

null

Machine Learning

machine-learning

Machine Learning

51,320

Robert Chang

Data @Airbnb, previously @Twitter. Naturally opinionated, but opinions are my own

c00b242128fe

rchang

9,427

80

20,181,104

null

0

null

0

null

2018-09-13

2018-09-13 19:36:54

2018-09-13

2018-09-13 20:01:56

25

false

en

2018-09-13

2018-09-13 20:11:54

38

16b4e7555e03

15.690566

1

0

A talk about how our folklore makes its way into how we think about technology, and how our technology itself spawns its own folklore…

5

Things That Go Bump On The Net A talk about how our folklore makes its way into how we think about technology, and how our technology itself spawns its own folklore. Written for and debuted at Electromagnetic Field 2018. Cross-posted to my blog. This is the longhand script that I ended up reading onstage, having not finished writing the thing in enough time to learn it. As a result, this is pretty much what I said verbatim — bar some adlibbing. In this written version you’re not getting some bits of ambient audio, a couple of slides and my impeccable comic timing. I have, however, added links off to original sources and whatnot. Accompanying notes can be found here. First up, some housekeeping: thanks to EMF for having me again, and youse all for turning up. I’ll start with a disclaimer: this talk is at something like version 0.8. There are probably bugs. By continuing to sit here you agree to be bound by the terms of my beta test agreement. If you think I’ve missed a trick, or a particularly good story, please come and find me later. Also, it turns out that this is an ambitious topic to fit into half an hour. I’ll be glossing over a few things, passing by others at speed. I’ll assume that everyone’s clever people and has google to follow up things up later. I’ve published a lot of notes and references for this talk online. With that said, let’s get started… 174 years ago. Source: https://www.insulators.info/books/mpet On May 24th, 1844, Samuel Morse sent the first long-distance telegraph message across America, from Washington D.C. to Baltimore. He sent this quote from the Bible’s Book of Numbers. It’s a dramatic choice, for sure, and you can’t help but read as some kind of premonition (or at least monumental hubris) on Morse’s part, concerning the communications revolution he was helping to bring about. It’s also interesting though, because it frames the telegraph as a gift from a higher power. From God, to Morse, to the rest of us. Morse is merely channeling God’s will. Source: http://www.porcelainista.net/?p=12747 Four years after Morse’s transmission, in 1848 in New York State, two sisters reported hearing knocking, rapping and banging sounds at night that weren’t being made by anyone in the house. In time, the Fox sisters started to communicate with the entity making these sounds. On March 31st, Kate, the youngest, invited it to repeat the snappings of her fingers. It did. She then asked it to knock out her age, and that of her sister Margaret, and of their older sister Leah who had moved away. It did this too: 12, 15, 17. Over the next few days, they devised a code which included ‘yes’, ‘no’, and letters of the alphabet. Then, the Fox family fled the house, to nearby Rochester. The tapping spirit followed. Kate and Margaret became famous, demonstrating the first paid public seance in November 1849 and going on to successful careers as spirit mediums… until their confession in nearly forty years later 1888 that it was all a hoax. By that time though, it was far too late — Spiritualism had well and truly taken hold in the popular American — and British — imagination. In his book ‘Haunted Media’, Jeffrey Sconce suggests that the story of the Fox sisters and that of Samuel Morse are far from unrelated. The telegraph, along with its fundamental mechanism for sending messages over long distances, had also brought with it a whole new set of ideas about communication with other bodies. Once you’ve abstracted communication out to something that can happen instantaneously, with no idea of who or what is on the other end of the line, then it’s easy to take that abstraction one step further and imagine communication with entities on other planes of existence, communicating through more ethereal means than a simple wire. Now, at this point I really should mention radio and EVP — electronic voice phenomenon — but to be honest that’d be a whole half hour by itself. That’s definitely one to look up later though, if you’re interested and don’t already know about it. 83 years on from Kate & Margaret’s tapping spirit: 1931. Accounts of contact with strange creatures, or spirits, weren’t that uncommon. The Spiritualist movement, kickstarted by the accounts of the sisters Fox, had been running for around 45 years now, and people had been communing with voices from other planes of existence with some regularity. Source: http://gefmongoose.blogspot.com/p/the-story-of-gef.html In an isolated farmhouse on the Isle of Man, the Irving family — Jim, Maggie and their 12-year-old daughter Voirrey — reported hearing scratching and rustling noises coming from inside the hollow walls. These noises were apparently made by a trapped mongoose; one with yellow fur, and hands and feet, who had been born in New Delhi in 1852. They knew this because Gef spoke to them. Source: http://gefmongoose.blogspot.com/p/the-story-of-gef.html “I know who I am but I shan’t tell you. I am a freak. I have hands and I have feet, and if you saw me you’d faint, you’d be petrified, mummified, turned into stone or a pillar of salt.” Of course, Gef couldn’t speak at first — he just squealed and thumped inside the walls of the house. However, in time, he started to make gurgling noises, like a child, and with encouragement from the isolated Irvings, he picked up English — not least through Jim’s daily readings to him from the newspaper. Gef could also speak the odd phrase of Russian, Spanish, Welsh, Hebrew, Manx and Hindustani — he was a well-travelled mongoose. As he himself is reported to have said: “I was brought to England from Egypt by a man named Holland. When I was in India, I lived with a tall man who wore a green turban on his head. Then I lived with a deformed man, a hunchback.” As Gef’s confidence grew, he started to explore the island, and brought gossip back to the farmhouse he’d heard while catching rides under the buses. He gained a nest under the rafters in Voirrey’s room, which the family called ‘Gef’s Sanctum’ and left bacon and sausages for him there, which he’d swipe while they weren’t looking. With his confidence grew his notoriety. Gef’s story spread around the island, and in 1932 was reported on by The Isle of Man Examiner and The Isle of Man Weekly Times. From there, it was picked up by the mainland press, and before long the Irving’s farmhouse was getting frequent visits from journalists and paranormal investigators. Gef never appeared to any of them, though — although he would rattle and scream while they were there. He saved particular ire for one Harry Price and his dictaphone: “Is it that spook man Harry Price? Why, I won’t speak into it. I’ll go and smash his windows. I’ll drop a brick on him as he lies in bed.” Of course, the most likely explanation was that Gef was, as Sconce puts it, “an imaginary companion, created by […] Voirrey. […] A creative girl’s reaction to prolonged isolation and boredom.” The story of Gef adds another layer, though: if we accept that it was indeed a hoax, then it was one enabled and accelerated by the communications technologies of the day. Voirrey’s imagination was fed by the information that made it to the farmhouse at Cashen’s Gap through books, newspapers, radio — this was one of the things that made Gef more interesting than a ‘usual’ haunting, where the voices tend to bang on about the the ethereal mysteries on the other side of the vale. “I’ll split the atom! I am the fifth dimension! I am the eighth wonder of the world!” The story itself spread on the same networks that fed it. The tabloid press, knowing a good story when they saw one, gave extensive coverage to Gef, and by 1935 he’d made it as far as the Hong Kong Telegraph. The story’s so potent that it’s continued to propagate down through time to the present day, jumping from newspapers to books to magazines and the internet — even a field in Gloucestershire in 2018. At the last EMF Camp, I talked about numbers stations — mysterious, shortwave radio stations that operated across Europe during the Cold War, mostly broadcasting strings of numbers. Most people think these were coded communication channels, broadcast by state intelligence agencies — although none have ever ’fessed up to running them, officially. During that talk, I said that the thing I found interesting about the numbers stations was the way that they acted as a kind of generator of folk tales, and that I reckoned you could draw a line back from the stories the numbers stations spawned to campfire ghost stories. Today, I’m talking about my attempts to trace that line: how existing folklore makes its way into how we think about new technology how our technologies themselves generate their own folklore. Source: https://www.flickr.com/photos/queen_of_subtle/4462520710 One of the earliest examples I could find of a technology spawning its own distinct folklore — as opposed to existing ideas about spirits manifesting in a new context — was Titivillus, the printer’s devil, who crept into print shops and introduced errors into carefully laid-out type overnight. Source: Museum of Fine Arts, Boston There’s also an idea in Japanese folklore of Tsukumogami— spirits which tools acquire when they get old enough. These spirits are harmless enough — mostly playing whimsical pranks on their owners — unless you annoy them by throwing them away too soon, or being wasteful. Source: https://en.wikipedia.org/wiki/Luigi_Galvani Galvani’s experiments, twitching frogs’ legs with electricity in the late 18th century give us Mary Shelley’s Frankenstein, and eventually the ‘killer AI’ myth, where an artificial child wreaks revenge on its creator. Source: https://en.wikipedia.org/wiki/Gremlin One of the first examples from the 20th century are Gremlins. These mischievous little imps were said to infest aeroplanes, and had an advanced knowledge of aeronautical engineering — which they used for devilry. It was common for pilots to blame mechanical problems with their aircraft on the gremlins — some even claimed that the gremlins could communicate with them on a psychic level, causing their vision to fog, or to see mountains which weren’t there. Source: https://en.wikipedia.org/wiki/Boeing_B-17_Flying_Fortress Here’s an apparent account by ‘L.W’, a B17 ‘Flying Fortress’ pilot during World War II. “So I am very aware of my surroundings, and as I go higher, I notice an unusual sound coming from the engine. The instruments went nuts. I look at my right and I see an entity staring at me. Then I look at the aircraft’s nose, and there it is another one, hanging in there. Dancing lizards.” “But I was perfectly fine…my senses were in good shape, but the weird things were still there looking at me. They kept going at it, pounding the plane with all their might.” “They appeared to be laughing, with their big mouths open, looking at me, hitting the plane with their long arms, trying to pull stuff. I have no doubt in my mind that they were trying to crash it. I managed to stabilize the flight and I saw the critters falling off the aircraft. I don’t know if they fell and died, or if they jumped from my plane to a different one. I have no idea.” So again, let’s assume that the gremlins aren’t really real, and the product of some other, more explicable phenomena. And let’s put aside the observation that they’re pretty good examples of the trickster archetype — I’m sure we all know what that is, or can google it later. Today, we’re interested in what could cause the myth of the gremlins to come about, and what purpose it served. The most widely held theory is that oxygen can be in short supply in a high-altitude aeroplane, and lack of it could easily cause hallucinations. Other writers suggest that the gremlins served a morale-preserving function, allowing air crew to blame faults on something other than the maintenance crews, their comrades. I think there might be something else going on here, too. A military plane is a fiendishly complex piece of engineering, with an awful lot of things that can fail in a lot of unpredictably interdependent ways. In other words: hard to understand. So once you hit a certain level of complexity, of unknowability — why not gremlins? Seems as plausible as anything else. It’s like the dark side of that ubiquitous Arthur C. Clarke quote that I won’t waste anyone’s time by repeating here. And I think we can see this effect at work when we think about the stories that people make up to explain technology to themselves. Julie Carpenter, writing about mythmaking, says this: The thing I liked about the numbers stations wasn’t just that they represented this tantalising, powerful mystery of the radio network; they also drew people together to try to work out what they were doing and why they were there. And when those stories intersected with the internet, the next generation of network, all kinds of weird things fly off as people take the idea and run with it. Photos appear, allegedly of site visits to defunct stations. Strange Youtube channels and Twitter bots appear, extending on the idea of hiding coded secrets in plain sight. The raw material of the story of the number stations — recordings, logs, theories of who might have been broadcasting them and why — get extended on by many imaginative minds. Arthur Frank, writing about how stories work socially, says and It’s these social aspects of storytelling that I think gives us modern meme culture, the shared narrative sources and social functions of a story multiplied hugely by network effects. Source: SlenderMan by TheTuneOfTurbo on DeviantArt A story like Slender Man, for example, who first pops up on the Something Awful forums in a Photoshop competition to make super-creepy images. His images show a faceless, gangly creature lurking in the woods behind some oblivious kids. However, the story swiftly takes on its own momentum as it comes into contact with many more creative minds, all adding their part to the legend, playing with fragments of folklore to create an ongoing, shared, viral narrative, popping up all over the internet in more creepy images and stories, implicating the Slender Man in atrocities large and small. And at the same time as being a thoroughly networked fable, Slender Man simultaneously has its roots in our deep folklore. Shira Chess notes that: “He owes many of his characteristics and some of his behaviours to fiction and folklore characters that preceded him” She makes a connection between the Slender Man collective mythos and that of middle-European faerie lore, pointing out that many of Slender Man’s characteristics and behaviours — luring humans into traps, distorted or indeterminate body features, child abduction, a general air of unheimlich — all have precursors in traditional folk tales. Slender Man turned out to be such a powerful story that it even infected old media. While writing this talk, I spotted these movie posters appearing along my commute: Coincidence, I’m sure. So we’ve seen that the way that we think and tell stories about our technologies actually has close ties to the ways that humans have always told stories about the world and interpreted phenomena. I want to spend the rest of this talk examining one particular strand of folk archetype: the oracular machine. And I’m going to start with the story of Roger Bacon, a 13th century friar and philosopher, and his alchemic bronze head. Source: https://commons.wikimedia.org/wiki/File:Friar_Bacon%27s_Brazen_Head.png According to the records, Friar Bacon wasn’t the first to cast a head from bronze in order to create an oracle — apparently, Pope Sylvester the 2nd made one in the 9th century (or stole it, with the aid of demons — but that’s another story). Several were said to have been built by Renaissance makers in the 12th and 13th centuries (interestingly though, most written accounts don’t appear until the sixteenth century). These heads were apparently able, after sufficient engineering or alchemic effort, to answer any question put to them, or make predictions about the future — but some, familiarly, could only answer ‘yes’ or ‘no’. It’s the account of Friar Bacon which became the most famous though, thanks to the sixteenth-century playwright, Robert Greene. In Greene’s play, based, apparently, on a true story, Bacon and his assistant Miles spend a great deal of time and effort building a bronze head “that in the inward parts thereof there was all things like as in a naturall man’s head”. Finally, they’re ready to animate it — which required keeping a continuous watch and also “the continuall fume of the six hottest simples” — plant extracts to you and me, your basic alchemic primitives. After three weeks of this, Bacon falls asleep, exhausted. While he sleeps, the head finally boots up and says three things: Miles, still awake, freaks out so hard that he knocks the head over and it shatters. Bacon sleeps through the whole thing. Years later, in 1837, Charles Babbage theorised that that given enough information, an entity might “distinctly foresee and might absolutely predict for any, even the remotest period of time, the circumstances and future history of every particle” He was thinking about a process which could model the whole universe, and thus predict the future. Source: Meet Alexa: Weather Sound familiar? The idea of a machine which can answer any question, has been with us for at least as long as we’ve been making machines — the original Mechanical Turk is in here, too. The idea that an engineer’s ingenuity can tell the future is also interlaced in with the Faustian myth of being granted great knowledge through esoteric means — but that knowledge usually comes with a price. I think this archetype is alive and well in the ideas framing the discourse around Big Data & machine learning, and it’s not just me: Source: AI researchers allege that machine learning is alchemy I think that alchemy is getting a bad rep here — it did, after all, lay the foundations for a lot of modern medicine and chemistry, after all. It did, though, also have its fair share of rituals that were enacted because the rituals themselves were thought to be powerful — and this is the analogy that Google’s researchers are making. They claim that most people doing ‘AI’ are following rituals that they’ve learned without really understanding the underlying science. Input datasets and training parameters are tweaked according to recipes people find on the internet, or learn and accumulate by trial and error. I suppose you could call that folk knowledge. We have another problem in ‘AI’: it’s almost impossible to understand what’s happening inside a neural net. Sure, we understand how to set up the conditions that allow the network to build itself, and tweak those conditions to get the outputs we want, but we don’t even know why neural nets are so effective, let alone what’s going on in there. We’re creating black boxes. But as long as we keep doing the rituals, the systems will behave. Source: Orthodox priest blessing the server room It never ceases to amaze me that as an industry we’re actively involved in a project of hiving off more and more important data processing tasks to machine learning systems. Self driving cars, sure, but also setting insurance premiums, making healthcare decisions. Setting bail. Predicting someone’s chances of re-offending. Running the stock market. And at the heart of all these systems? Faith. Faith in unknowable black boxes. Source: God and the Chip: Religion and the Culture of Technology He goes on to say: “There are dangers in treating technology like magic. Scientists and engineers become priests and wizards whose pronouncements cannot be understood by the public, and implicitly, should not be challenged” So I could end here, on a pithy summary that appears to neatly wrap everything up in a bow. But that seems like cheating. We like neat resolutions — happy endings, elegant reversals, thematic resonance. We like frameworks, and that leaves us susceptible to a kind of narrative exploit. If a story seems to fit a pattern we know, we’ll accept it more easily because we know that story. Superstar CEOs are a neat fit for the hero’s journey. A new shiny thing is a fairy gift, powerful and capricious. All-knowing ‘AI’ systems become minor deities, capable of doling out judgement on us imperfect mortals. Deployment of these archetypes dulls our critical faculties, makes us more prone to magical thinking, absolves us of responsibility as we get swept up in the narrative. And just as we like our stories to follow orders and structures, we like to think that our technologies are the product of science — rigorous, rational thought. But they are also the product of people — people like you and me, sitting down at a keyboard or a drawing board, bashing out their opuses. And those people are all embedded in a culture, grew up on particular stories. We can’t help but have the way we think shaped by those stories — in a way, those stories are what taught us to think. And we repeat those patterns. So, I’m not going to end this neatly. But, if I am going to leave you with one thought, it’s this: keep alert. Reserve special scepticism for the stories which seem too neat, too powerful, too universally true, because they might just be regurgitation of our shared folklore to the bamboozlement of all involved. Thank you. Acknowledgments Thanks to the following friends, comrades and fellow-travellers who helped shape this talk in one way or another. All mistakes, however, are most certainly my own. Bill Thompson, whose fault it mostly is that I finished this talk. Garrett D. Tiedemann, who also shares a lot of the blame. Tom Armitage Natalie Kane Honor Harger Barbara Zambrini Alyson Fielding David Paliwoda Kim Plowright Ben Bashford Lydia Nicholas John V. Willshire

Things That Go Bump On The Net

3

things-that-go-bump-on-the-net-16b4e7555e03

2018-09-13

2018-09-13 20:11:54

https://medium.com/s/story/things-that-go-bump-on-the-net-16b4e7555e03

false

3,628

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Henry Cooke

making things out of code & internets

cfdf58deb59b

prehensile

456

324

20,181,104

null

0

null

0

null

2018-04-22

2018-04-22 20:35:40

2018-06-02

2018-06-02 11:14:01

0

false

en

2018-06-02

2018-06-02 11:14:01

1

16b55e8f95a6

3.041509

0

Last month marked the anniversary of the Luddite rebellion, a series of riots in 1811 protesting the introduction of new machinery to the…

1

Fundamental Skills you will needs to succeed in the age AI and Automation Last month marked the anniversary of the Luddite rebellion, a series of riots in 1811 protesting the introduction of new machinery to the cotton industry. Many workers lost their jobs, but as the industrial revolution progressed it paved the way for a whole new set of jobs in manufacturing and a host of other industries. Fast forward 200 years, and people are still up in arms over the advent of transformative technologies – most recently in the form of artificial intelligence (AI) and automation. Even a tech pioneer like Elon Musk has expressed fears around AI, calling it the “worst event in the history of our civilization.” While there will be job elimination in every industry, modern-day Luddites must realize that the need for “human” workers will never go out of style, especially for those who can work alongside these new machines. While workers with strong science, technology, engineering and math (STEM) backgrounds will be in greater demand than ever, these skills will have little value without key human traits of creativity, emotional sensitivity, judgment, and understanding of human relationships. The not-too-distant future of work will mean working alongside machines. To do this right, workers at every level need to complement the machines, not compete with them. This will require an arsenal of both left- and right-brained skills, or what I call “STEM+” workers. Here are three fundamental skills that every worker will need to succeed in the age of AI and automation: Out-of-the-box thinking: Machines thrive off patterns and predictions, but what about the work that doesn’t fall neatly into such categories? This is where the need for creativity arises. AI and other technologies are actually creating more opportunities to be creative, as humans can use creativity to produce value where machines cannot. For example, while machines can collect and analyze data rapidly, humans can provide logical inferences and engage in abstract thinking to identify opportunities to push boundaries and innovate. Think about the rise of technology related to music creation. Just a few decades ago, I would have needed to spend years mastering an instrument or studying music theory to compose and play an original song. Today, my son can use technology to compose a symphony and conduct an entire chamber orchestra with just one computer. Ethical judgment: Soon, the cost of prediction will plummet to virtually nothing. AI-driven prediction will equip humans with the information necessary to make more accurate health diagnoses/prognoses, determine financial outcomes, and even foresee the results of Supreme Court cases. But with the advancement of prediction technology, there will be a greater need for reliable and insightful human judgment. Ethical judgment is needed at every step of AI, from creating, to implementing, to monitoring. If ethics aren’t in place throughout, we could see strong biases in AI that could potentially make the technology less productive. Employees with the skills to make complex decisions based on a mixture of quantitative and qualitative data will hold the key to industries like insurance, healthcare, and communication–any field dealing directly with people. We’re already seeing the demand for responsible judgment in the media industry, as every company struggles to weed out fake news, trolls, and unsuitable content from their websites. Algorithms and bots are often responsible for aggregating these stories in the first place, but human judgment is still needed to ensure accurate information is disseminated responsibly. We decide the “rules of play” and this requires thinking ethically as we code and train AI applications. Insatiable curiosity: The days of only learning for 20 years and working for 40 are long over. While most of us are born with a natural sense of curiosity, it’s critical that we allow ourselves to continue to question, experiment, and seek out new opportunities to learn. Staying curious throughout our professional lives enables us to seek out innovative solutions when the answer to a challenge is not immediately apparent. More importantly, curiosity helps us to fail fast, recognize our errors even faster, and adjust course accordingly. As increasingly advanced technologies permeate our workplaces, we must indulge our curiosity to learn how these new technologies work, how to leverage them to make ourselves more productive, and identify the skill sets we’ll need to invest our time in mastering. Think about the meteoric rise of blockchain, a technology that continues to confound many people. If we are curious about blockchain rather than wary, we can discover new applications for the technology and potentially enhance a company’s security, accuracy, customer experiences, and more. The future of work will be reshaped by machines, but the need for decidedly human skills is only growing. By embracing their creative side, trusting their judgment, and staying curious, workers can win. Originally published here

Fundamental Skills you will needs to succeed in the age AI and Automation

0

fundamental-skills-you-will-needs-to-succeed-in-the-age-ai-and-automation-16b55e8f95a6

2018-06-02

2018-06-02 11:14:02

https://medium.com/s/story/fundamental-skills-you-will-needs-to-succeed-in-the-age-ai-and-automation-16b55e8f95a6

false

806

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

David ALADE

I'm here to share my view with my "pen"

2e89be4be205

davidalade

8

26

20,181,104

null

0

null

0

eef0bcb2e368

2018-03-31

2018-03-31 04:59:25

2018-03-31

2018-03-31 05:00:30

1

false

en

2018-03-31

2018-03-31 05:00:30

2

16b757c6a681

1.298113

1

0

In the previous blog we have gone through how more data or to be precise more dimensions in the data creates different problems like…

1

The second dimensionality reduction method In the previous blog we have gone through how more data or to be precise more dimensions in the data creates different problems like overfitting in classification and regression algorithms. This is known as “curse of dimensionality”. Then we have gone through the solutions to the problem i.e. dimensionality reduction. We were mainly focused on one of the dimensionality reduction method called feature selection. In this blog, we will go through the second dimensionality reduction method that we were discussing. Unlike feature selection, feature extraction doesn’t create the subsets and then find the best one out of them, instead, it tries to create whole new features set from the existing features. For example we have the data set X = {x1, x2, x3…..xn}, after doing the feature extraction it would be something like Y = {y1, y2, y3, ….ym} where m is the new number of dimensions extracted out. If we place it into a formula it would be f(X) = Y. Basically, the function f is doing the projection of a higher dimensional feature space to a lower dimensional feature space where new features are uncorrelated and cannot be reduced further. One of the things we have to keep in mind while doing the projection is that every feature must have a larger variance so that it can be distinguished from the other features easily. This method of finding uncorrelated data with high variance is known as Principal Component Analysis. Let’s take the below example. We have the two geometric representations and as you can see the second diagram the variance would be more for each point. By considering the previous point we have to select the second diagram as our projected Continue reading ….

The second dimensionality reduction method

1

the-second-dimensionality-reduction-method-16b757c6a681

2018-04-17

2018-04-17 19:53:11

https://medium.com/s/story/the-second-dimensionality-reduction-method-16b757c6a681

false

291

Its all about machine learning and AI from the basic to the real world uses of them.

null

MachineX

pranjut.gogoi@gmail.com

machinex

MACHINE LEARNING,ARTIFICIAL INTELLIGENCE

pranjut_gogoi

Data Science

data-science

Data Science

33,617

Pranjut Gogoi

null

f6e750b44053

pranjut.gogoi

0

1

20,181,104

null

0

null

0

null

2018-08-28

2018-08-28 12:05:58

2018-08-28

2018-08-28 12:10:30

2

false

en

2018-09-07

2018-09-07 06:59:15

1

16b7937a7827

3.470126

0

Artificial Intelligence aka Machine Intelligence refers to the intelligence demonstrated by a machine as compared to humans or animals. It…

1

The Importance of Artificial Intelligence & Its Uses Artificial Intelligence aka Machine Intelligence refers to the intelligence demonstrated by a machine as compared to humans or animals. It also related to the study of the intelligent agents that are developed as per the advancement in the field of science. Computers, laptops, and mobile phones stand at the basic level of AI. Different hi-tech machines and robots are described as artificial intelligence. In the present-day scenario, AI is used in different fields like medical, sports, science, cinema etc. It is also used by entrepreneurs and companies for scaling the business. Using the artificial intelligence and business strategies has been helping many companies worldwide. But prior to it, one needs to understand what AI is? The Working Structure of Artificial Intelligence Artificial Intelligence works by combining data with fast and frequentative algorithms that allow software to grab the concept from the patterns of the data automatically. This field has a wide web comprising multiple theories, technologies, and the following subfields: Machine Learning Machine learning extends the use of predictive statistics by making use of the enormous computational power that is available nowadays and by taking into account all available information; not just a small, representative sample of that. It can solve both Classification tasks such as binary variable classification (high risk/low risk, spam/not spam) and Regression problems of continuous variables (sales forecast, house prices). Their main ingredients are Neural Networks which consist of layers of neurons that allow them to map information appropriately and find hidden patterns that the human brain and classic algorithms cannot. They can have deep architectures with many stacked neuron layers that are called Deep Neural Network. Cognitive Systems It is a subfield of AI that strives for a natural interaction between humans and machines, like people using computers, laptops in their daily lives. They depend on Deep Neural Networks and have cleared the path for the development of applications that allow human-computer interaction (e.g. voice recognition, text-to-speech conversion and natural language processing). Technologies such as Recurrent Neural Networks (RNN), Long-Short Term Memory (LSTM) and Generative Adversarial Networks (GAN) are highly recommended. Computer Vision It works on recognizing a pattern or image. When machines can recognize and understand images. They can capture the same in the real time as well. For example, the smallest version of this can be found in mobile phones. The feature of Geo-Tagging works on this concept. Also In Healthcare it helps scientists and medical professionals in interpreting X-ray readings or MRI Scans. Here, are used more Convolutional Neural Networks (CNN), Single Shot MultiBox Detector (SSD), You Only Look Once (YOLO) algorithms. How Is Artificial Intelligence Being Used? Due to the capabilities of AI, every industry requires it in the present-day scenario. Examples include fake news detection, music composition, photo and video tagging, text summarization and face recognition, among which others are just in infancy stages while others are more polished, already finding their way into industrial uses. Here’s a quick look at some fields where AI is used widely. Fintech — In Finance, quants use it for algorithmic trading, risk modelling and portfolio management. Banking examples include Fraud Detection, Loan Default Prediction Anti-Money Laundering and ATM cash withdrawal forecasts. Healthcare- AI applications can provide personalized X-ray machines and its readings. They also provide unique help in the fight of cancer treatment. Manufacturing- AI can analyze the data related to a specific product like its details, components, name of the manufacturer, etc. Two illustrative Manufacturing examples are autonomous driving cars and drones. Retail- AI provides virtual shopping capabilities that provide personalized shopping recommendations as per the user’s recent searches or information related to it. Also it provides Churn Prediction, Sentiment Analysis, Customer Segmentation and automated virtual assistants (Chatbots). Sports- AI is also used in the field of sports. It helps in providing the image of the ground from different angles and helps in determining the strategy of the game, field set up or any other renovation required. The idea of creating new artificial intelligence and developing the existing ones is for the betterment of the human life. It will reduce the errors and accomplish different tasks in no time. In many countries, the successful invention and operation of the self-driven cars, robotics in different fields, self-operated CCTV, vehicle lock, and other things is a sign that in the near future, the mankind may or may not rely completely on the artificial intelligence. The uses of AI in any sector of our personal and professional lives are literally only limited by our imagination. The impact of AI is already measurable and will only increase in the future. It lies in companies’ and people’s interest to make the most out of it in order to make this world more sustainable, efficient and livable. To know more about artificial intelligence visit: http://www.businessinsightco.com/artificialintelligence.html

The Importance of Artificial Intelligence & Its Uses

0

the-importance-of-artificial-intelligence-its-uses-16b7937a7827

2018-09-07

2018-09-07 06:59:15

https://medium.com/s/story/the-importance-of-artificial-intelligence-its-uses-16b7937a7827

false

818

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

business insight Corp

null

27b7fe671803

businessinsightcorporationltd

0

1

20,181,104

null

0

null

0

null

2018-08-20

2018-08-20 02:01:34

2018-08-20

2018-08-20 02:53:15

1

false

en

2018-08-20

2018-08-20 02:53:15

0

16b81f877154

2.215094

0

So, what are the odds of us living in a simulated reality? Over the past few years, in which we have undoubtedly reach a very advanced…

1

What are the odds of us living on a simulation? So, what are the odds of us living in a simulated reality? Over the past few years, in which we have undoubtedly reach a very advanced technology, there has been a hypothesis that questions wether this universe is real or it’s a simulation. This debate has been in fact the central plot of many famous movies and tv shows. To begin with, the scientists that proposed this theory insisted that if we continue advancing technologically as fast as we are now, then we will reach a certain point in which we have such an strong power on the computers that we will be able to simulate in perfection our forebears. So, they say that there are 3 simple, main options or future scenarios: 1) The civilization dies before reaching that point. 2) Our descendants dont like the idea of simulating their forebears (basically us) 3) We are living in a simulated reality because they did reach that point and they did the simulations. “One thing that later generations might do with their super-powerful computers is run detailed simulations of their forebears or of people like their forebears. Because their computers would be so powerful, they could run a great many such simulations. Suppose that these simulated people are conscious (as they would be if the simulations were sufficiently fine-grained and if a certain quite widely accepted position in the philosophy of mind is correct). Then it could be the case that the vast majority of minds like ours do not belong to the original race but rather to people simulated by the advanced descendants of an original race. It is then possible to argue that, if this were the case, we would be rational to think that we are likely among the simulated minds rather than among the original biological ones. Therefore, if we don’t think that we are currently living in a computer simulation, we are not entitled to believe that we will have descendants who will run lots of such simulations of their forebears.” — Nick Bostrom, Are you living in a computer simulation? On the other hand, there are actual proofs of the limits of computation that declines de possibility of our technology ever reaching that point. As there is also no proofs that its physically possible to simulate an entire civilizationin a way that is sufficiently fine-grained. Not only that but researchers at the University of Oxford claim that “even just to store the information about a few hundred electrons on a computer one would require a memory built from more atoms than there are in the universe.” So they practically say its impossible. Nonetheless, scientists and the people who stick to the simulation theory maintain that we are capable of advancing technologically “crossing those limits” of course, if not destroying ourselves first. In conclusion, I find it both disturbing and amusing the idea of us living on a simulated theory. I think that is an extremely crazy, nonsense idea but I am really saying this with no comprehension of physics or anything alike at all, so I am not going to be close with it either.

What are the odds of us living on a simulation?

0

what-are-the-odds-of-us-living-on-a-simulation-16b81f877154

2018-08-20

2018-08-20 02:53:15

https://medium.com/s/story/what-are-the-odds-of-us-living-on-a-simulation-16b81f877154

false

534

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

dalia ivanoff

null

caacd444dde1

daliaivanoff

1

2

20,181,104

null

0

[1] "árvore de 18 m de altura x 35 cm de diâmetro flores com cálice verde pétalas brancas frutos até 15 cm de comprimento casca pilosa de cor verde-ferruginosa arilo branco frutos separados caatinga alta solo arenoso humoso " [2] "árvore de 10m de altura x 10cm de diâmetro flores amarelas frutos imaturos verdes látex branco carpoteca " [3] "arbusto de flores alvas com manchas roseas e botões florais látex alvo carpoteca " > closest_to(model, "flores") word similarity to "flores" 1 flores 1.0000000 2 flors 0.5962055 3 flore 0.4516035 4 flroes 0.4367272 5 flôres 0.4082263 6 pétalas 0.4020755 7 corolas 0.3935502 8 fauces 0.3435436 9 floras 0.3199513 10 inflorescências 0.3145785 > closest_to(model, "amarelas", 20) word similarity to "amarelas" 1 amarelas 1.0000000 2 brancas 0.7183580 3 roxas 0.7006725 4 alvas 0.6747162 5 róseas 0.6678736 6 alaranjadas 0.6614220 7 amareladas 0.6402489 8 vermelhas 0.5869951 9 amarelo-alaranjadas 0.5837175 10 amarelo-claras 0.5739101 11 lilases 0.5512847 12 rosas 0.5477697 13 amarelo-pálidas 0.5430483 14 amrelas 0.5351215 15 cremes 0.5329505 16 marelas 0.5304547 17 creme-amareladas 0.5187036 18 laranjas 0.5164795 19 azuladas 0.5120074 20 roseas 0.5092904 require(dplyr) require(stringr) require(tidytext) require(wordVectors) require(tidyverse) require(sf) require(sp) flores <- closest_to(model, "flores", 10)$word > flores [1] "flores" "flors" "flore" "flroes" [5] "flôres" "pétalas" "corolas" "fauces" [9] "floras" "inflorescências" cores <- unique( c( closest_to(model, "amarelas", 20)$word, closest_to(model, "brancas", 20)$word, closest_to(model, "vermelhas", 20)$word, closest_to(model, "rosas", 20)$word, closest_to(model, "laranjas", 20)$word, closest_to(model, "azuis", 20)$word, closest_to(model, "pretas", 4)$word, closest_to(model, "marrons", 3)$word ) ) > cores [1] "amarelas" "brancas" "roxas" [4] "alvas" "róseas" "alaranjadas" [7] "amareladas" "vermelhas" "amarelo-alaranjadas" [10] "amarelo-claras" "lilases" "rosas" [13] "amarelo-pálidas" "amrelas" "cremes" [16] "marelas" "creme-amareladas" "laranjas" [19] "azuladas" "roseas" "esbranquiçadas" [22] "branco-esverdeadas" "branco-amareladas" "creme-esverdeadas" [25] "lilazes" "amarelo-esverdeadas" "branco-rosadas" [28] "azuis" "vermelho-alaranjadas" "rosadas" [31] "avermelhadas" "arroxeadas" "vináceas" [34] "purpúreas" "vinosas" "esverdeadas" [37] "violáceas" "liláses" "violetas" [40] "laranja" "laranjadas" "laranja-avermelhadas" [43] "amarelo-avermelhadas" "vermelho-amareladas" "verde-amareladas" [46] "azul-arroxeadas" "roxo-azuladas" "lilázes" [49] "pretas" "negras" "enegrecidas" [52] "marrons" "amarronzadas" "castanhas" tokens <- splink %>% unnest_tokens(token, notes, token = "ngrams", n = 2) %>% separate(token, c("primeira_palavra", "segunda_palavra"), sep = " ") %>% filter(primeira_palavra %in% flores, segunda_palavra %in% cores) %>% distinct() tokens %>% select(primeira_palavra, segunda_palavra, latitude, longitude) %>% head() Simple feature collection with 6 features and 4 fields geometry type: POINT dimension: XY bbox: xmin: -60.02643 ymin: -19.9 xmax: -39.0333 ymax: -3.106409 epsg (SRID): 4326 proj4string: +proj=longlat +datum=WGS84 +no_defs # A tibble: 6 x 5 primeira_palavra segunda_palavra latitude longitude geometry <chr> <chr> <dbl> <dbl> <simple_feature> 1 flores brancas -19.900000 -43.40000 <POINT (-43.4...> 2 flores brancas -14.163600 -47.81110 <POINT (-47.8...> 3 flores brancas -15.208600 -45.85190 <POINT (-45.8...> 4 flores brancas -18.960275 -49.46002 <POINT (-49.4...> 5 flores alvas -16.283300 -39.03330 <POINT (-39.0...> 6 flores brancas -3.106409 -60.02643 <POINT (-60.0...> cores_traduzidas <- c( "amarelas" = "yellow", "brancas"= "white", "roxas" = "purple", "alvas" ="white", "róseas" = "pink1", "alaranjadas" = "orange", "amareladas" = "lightyellow", "vermelhas" = "red", "amarelo-alaranjadas" = "goldenrod", "amarelo-claras" = "lightyellow2", "lilases" = "violet", "rosas" = "pink1", "amarelo-pálidas" = "lightyellow2", "amrelas" = "yellow", "cremes" = "wheat4", "marelas" = "yellow", "creme-amareladas" = "lightgoldenrodyellow", "laranjas" = "darkorange", "azuladas" = "blueviolet", "roseas" = "lightpink", "esbranquiçadas" = "wheat2", "branco-esverdeadas" = "palegreen", "branco-amareladas" = "lightyellow", "creme-esverdeadas" = "darkseagreen3", "lilazes" = "violet", "amarelo-esverdeadas" = "greenyellow", "branco-rosadas" = "lightpink", "azuis" = "slateblue4", "vermelho-alaranjadas" = "orangered1", "rosadas" = "pink", "avermelhadas" = "tomato", "arroxeadas" = "thistle2", "vináceas" = "deeppink4", "purpúreas" = "violetred", "vinosas" = "deeppink4", "esverdeadas" = "palegreen", "violáceas" = "violet", "liláses" = "violet", "violetas" = "violet", "laranja" = "darkorange", "laranjadas" = "orange1", "laranja-avermelhadas" = "orangered1", "amarelo-avermelhadas" = "darkorange1", "vermelho-amareladas" = "orangered1", "verde-amareladas" = "yellow4", "azul-arroxeadas" = "blueviolet", "roxo-azuladas" = "blueviolet", "lilázes" = "violet", "pretas" = "black", "negras" = "black", "enegrecidas" = "black", "marrons" = "brown", "amarronzadas" = "brown", "castanhas" = "tan1" ) p_colors <- tokens %>% ggplot() + geom_sf(data = br, fill = "#313131", size = 0.1, colour = "#414141") + geom_point( aes(x = longitude, y = latitude, colour = segunda_palavra), size = 0.8, alpha = 0.7 ) + ggtitle("Cores de flores pelo Brasil", subtitle = "Ocorrência de cores em 200 mil registros de herbários") + scale_colour_manual(values = cores_traduzidas) + labs(x = NULL, y = NULL) + theme_plex() + theme(legend.position = "none") + coord_sf(datum = NA) > tokens %>% count(primeira_palavra, segunda_palavra, sort = TRUE) Simple feature collection with 214 features and 3 fields geometry type: GEOMETRY dimension: XY bbox: xmin: -73.67256 ymin: -33.66056 xmax: -32.44806 ymax: 5.168889 epsg (SRID): 4326 proj4string: +proj=longlat +datum=WGS84 +no_defs # A tibble: 214 x 4 primeira_palavra segunda_palavra n geometry <chr> <chr> <int> <simple_feature> 1 flores amarelas 48761 <MULTIPOINT (...> 2 flores brancas 45468 <MULTIPOINT (...> 3 flores alvas 26662 <MULTIPOINT (...> 4 flores roxas 12008 <MULTIPOINT (...> 5 flores vermelhas 7568 <MULTIPOINT (...> 6 flores róseas 7197 <MULTIPOINT (...> 7 flores esverdeadas 7151 <MULTIPOINT (...> 8 pétalas brancas 6909 <MULTIPOINT (...> 9 flores lilases 6574 <MULTIPOINT (...> 10 pétalas amarelas 5974 <MULTIPOINT (...> # ... with 204 more rows

12

null

2017-11-29

2017-11-29 21:44:40

2017-11-30

2017-11-30 03:24:15

2

false

pt

2017-11-30

2017-11-30 13:59:16

1

16b98a333a33

6.281447

7

0

Usando o R para mapear a coloração das flores no Brasil

5

As cores das flores Usando o R para mapear a coloração das flores no Brasil Clique na imagem para vê-la em tamanho maior O mapa acima mostra como a coloração das flores é distribuida pelo Brasil. Dos 237 mil registros utilizados, mais da metade apresentavam flores descritas como brancas, amarelas ou termos similares. Os dados utilizados vieram de herbários, que são coleções onde pesquisadores que trabalham com vegetação depositam partes de plantas fixadas à folhas de papel para comparações e consultas futuras. As coletas são feitas principalmente em locais mais acessíveis, como perto de estradas e nas margens de rios. Por isso, há poucas coletas em regiões pouco populosas, como o Norte e Centro-oeste. Nos herbários brasileiros há mais de cinco milhões de registros. Ramo de ipê-amarelo depositado em herbário. Cada um desses registros possui uma etiqueta onde a pessoa responsável pela coleta descreve detalhes da planta e do local onde ela foi encontrada. Em muitos casos, essa descrição inclui a cor das flores e frutos da planta, já que com o tempo essas partes da planta escurecem e a cor original é perdida. Uma das coisas que mais faço no meu pós-doutorado é processar o texto de milhões de entradas dessas etiquetas que já foram digitadas para determinar características das plantas, principalmente se elas estavam com flores ou não no momento da coleta. O banco de dados que eu compilei possui por volta de 5 milhões de etiquetas, parecidas com as seguintes: Recentemente, tenho usado modelos word2vec auxiliar no processamento desse mundo de etiquetas. Modelos word2vec são basicamente redes neurais simples que agrupam termos dentro de contextos baseados nas palavras que co-ocorrem entre eles. Dessa forma, é possível identificar palavras que, no contexto das etiquetas, provavelmente se referem à uma mesma coisa. Uma busca no modelo pela palavra “flores”, por exemplo, retorna o seguinte: Sem saber a relação entre as palavras previamente, o modelo foi capaz de agrupá-las simplesmente pelo contexto, ou seja, pelos termos que normalmente aparecem próximos à essas palavras. Percebam que aqui ele encontrou várias entradas em que a palavra “flores” foi digitada de maneira errada, o que é bastante útil pra mim. Caso eu tivesse procurado literalmente por “flores”, todas essas entradas teriam sido ignoradas. O modelo também indicou outros termos usados no mesmo contexto, como inflorescência, pétalas e corolas. Tudo isso sem ter ideia de que as palavras são de fato relacionadas. A busca por “amarelas” tem como retorno: O resultado da busca indica que o modelo soube agrupar diversas cores dentro de um mesmo contexto. Assim, eu consigo determinar de que forma os coletores descrevem cores nas etiquetas, sem precisar criar um dicionário de todas as cores possíveis. Aliás, fazer isso seria muito difícil, uma vez que aparecem combinações como “amarelo-pálidas”, “creme-amareladas” e por aí vai. Eu treinei esse modelo em um corpus (conjunto de palavras) com 900 mil etiquetas previamente selecionadas, o que gerou um vetor com milhões de palavras, sendo que 12 mil delas foram usadas em pelo menos 20 etiquetas. Nas etiquetas, além das informações sobre as características da planta, muitas vezes há também as coordenadas geográficas. Assim, por pura curiosidade, resolvi criar um mapa com todos os registros com coordenadas e em que nas etiquetas havia informações sobre as cores das flores. Nesse mapa, cada ponto teria a cor da flor descrita para aquele indivíduo. Abaixo descrevo os passos mais importantes no R. Primeiro, usei os modelos word2vec já treinados para criar vetores com termos que indicam flores e cores: O próximo passo foi tokenizar o corpus, ou seja, quebrá-lo em grupos de n palavras. Para este exemplo, criei grupos de duas palavras. A rotina do R é relativamente grande, então mostro aqui só as partes relevantes. Para criar os tokens usei o pacote tidytext . A ideia era quebrar todo o texto em grupos de duas palavras, mantendo aqueles grupos em que a primeira palavra consta no vetor flores e a segunda no vetor cores . O objeto splink contém o texto das etiquetas, já extensivamente processados para o meu trabalho. Depois, traduzi as cores indicadas pelos coletores em cores que o R conhece: Por fim, criei o mapa do começo da postagem com os 237 mil registros que restaram. O objeto br é o shape do Brasil que eu já havia carregado: Não gosto do fundo preto, mas é preciso para dar destaque a cores como o branco. Flores amarelas e brancas são maioria:

As cores das flores

41

as-cores-das-flores-16b98a333a33

2018-05-12

2018-05-12 09:36:22

https://medium.com/s/story/as-cores-das-flores-16b98a333a33

false

1,563

null

Rstats

rstats

Rstats

450

Gustavo Carvalho

R, data science, and ecology.

1c8e675e88ec

ghcarvalho

30

35

20,181,104

null

0

null

0

null

2018-06-26

2018-06-26 13:26:51

2018-06-15

2018-06-15 00:00:00

1

false

en

2018-06-26

2018-06-26 13:27:24

2

16b9ea3180bd

1.781132

0

AI has captured imaginations and sparked fear. Depending where one stands on the spectrum, AI can be viewed as a job-killing monster or a…

5

Why Artificial Intelligence Can’t Solve Everything AI has captured imaginations and sparked fear. Depending where one stands on the spectrum, AI can be viewed as a job-killing monster or a creator of mass prosperity. For the pessimists, AI will lead to wealth concentration in the hands of a few. To keep society from crumbling, governments will create socialist policies to feed the billions of poor people. On the other hand, many in Silicon Valley and governments worldwide argue that AI promises an era of unparalleled prosperity. They point to the incredible advances they expect AI to bring to medicine and business. A race has begun between countries, with each trying to push to the top of the AI pyramid. Great Britain, China, France, and the United States are all contenders for AI dominance. As many of the dystopian fears of AI are unfounded so are many of the hopes. AI’s transformative power is likely to move slower than the optimists project. It is still an emerging technology, and some growing pains are expected. Neural networks — not so easy to create or manage AI is not as simple as plugging in a program and letting it do the work. The creation process takes enormous human labor, and rather than eliminating the need for human intervention, it changes it. As explained by Vyacheslav Polonski, an Oxford University AI researcher, neural networks are extremely complex structures patterned off the human brain. Though neural networks are much smaller than brains, they have tremendous processing power, allowing them to access data quantities far beyond the human mind’s capability. Neural networks can then use this data to discover patterns and rules. As Polonski notes, neural networks cannot be expected to solve problems by virtue of being plugged in. They are far from magical. Politicians must understand that putting AI into a democratic political system cannot, by itself, make a democracy function the way they wish. In addition, governments will struggle in integrating data for AI analysis. Much of the data is still stored offline. In order to make AI work, they must first build the infrastructures needed. This takes time. Further, AI innovations take much longer to implement than many business and political leaders expect. As a result, any benefits are likely to occur on a longer time scale. For those in the race for AI supremacy, this causes no discouragement. Though AI may need time to develop, AI’s influence on future societies is clear. Originally published at chartwestcott.com on June 15, 2018.

Why Artificial Intelligence Can’t Solve Everything

0

why-artificial-intelligence-cant-solve-everything-16b9ea3180bd

2018-06-26

2018-06-26 13:27:24

https://medium.com/s/story/why-artificial-intelligence-cant-solve-everything-16b9ea3180bd

false

419

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Chart Westcott

Chart Westcott is Co-Founder and COO at Ikarian Capital, LLC a long/short equity biotech focused investment adviser. Read more at http://chartwestcott.net.

27c380a347bd

chartwestcott

19

207

20,181,104

null

0

from sklearn import datasets from sklearn import svm from sklearn.externals import joblib # load iris dataset iris = datasets.load_iris() X, y = iris.data, iris.target # train model clf = svm.LinearSVC() clf.fit(X, y) # persistent model joblib.dump(clf, 'iris_model.pickle') syntax = "proto3"; option java_multiple_files = true; option java_package = "io.grpc.examples.ml"; option java_outer_classname = "IrisProto"; option objc_class_prefix = "HLW"; package ml; service IrisPredictor { rpc PredictIrisSpecies (IrisPredictRequest) returns (IrisPredictReply) {} } message IrisPredictRequest { double sepal_length = 1; double sepal_width = 2; double petal_length = 3; double petal_width = 4; } message IrisPredictReply { int32 species = 1; } from grpc.tools import protoc protoc.main( ( '', '-I.', '--python_out=.', '--grpc_python_out=.', './iris.proto', ) ) import os from concurrent import futures import time from pprint import pprint from sklearn.externals import joblib import grpc import iris_pb2 import iris_pb2_grpc _ONE_DAY_IN_SECONDS = 60 * 60 * 24 class IrisPredictor(iris_pb2_grpc.IrisPredictorServicer): _model = None @classmethod def get_or_create_model(cls): """ Get or create iris classification model. """ if cls._model is None: path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'model', 'iris_model.pickle') cls._model = joblib.load(path) return cls._model def PredictIrisSpecies(self, request, context): model = self.__class__.get_or_create_model() sepal_length = request.sepal_length sepal_width = request.sepal_width petal_length = request.petal_length petal_width = request.petal_width result = model.predict([[sepal_length, sepal_width, petal_length, petal_width]]) return iris_pb2.IrisPredictReply(species=result[0]) def serve(): server = grpc.server(futures.ThreadPoolExecutor(max_workers=10)) iris_pb2_grpc.add_IrisPredictorServicer_to_server(IrisPredictor(), server) server.add_insecure_port('[::]:50052') server.start() try: while True: time.sleep(_ONE_DAY_IN_SECONDS) except KeyboardInterrupt: server.stop(0) if __name__ == '__main__': serve() from __future__ import print_function import argparse import grpc import iris_pb2 import iris_pb2_grpc def run(host, port): channel = grpc.insecure_channel('%s:%d' % (host, port)) stub = iris_pb2_grpc.IrisPredictorStub(channel) request = iris_pb2.IrisPredictRequest( sepal_length=5.0, sepal_width=3.6, petal_length=1.3, petal_width=0.25 ) response = stub.PredictIrisSpecies(request) print("Predicted species number: " + str(response.species)) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--host', help='host name', default='localhost', type=str) parser.add_argument('--port', help='port number', default=50052, type=int) args = parser.parse_args() run(args.host, args.port) FROM ubuntu:16.04 WORKDIR /root # Pick up some TF dependencies RUN apt-get update \ && apt-get install -y --no-install-recommends \ build-essential \ curl \ pkg-config \ rsync \ software-properties-common \ unzip \ git \ && apt-get clean \ && rm -rf /var/lib/apt/lists/* # Install miniconda RUN curl -LO http://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh \ && bash Miniconda-latest-Linux-x86_64.sh -p /miniconda -b \ && rm Miniconda-latest-Linux-x86_64.sh ENV PATH /miniconda/bin:$PATH # Create a conda environment ENV CONDA_ENV_NAME iris-predictor COPY environment.yml ./environment.yml RUN conda env create -f environment.yml -n $CONDA_ENV_NAME ENV PATH /miniconda/envs/${CONDA_ENV_NAME}/bin:$PATH # cleanup tarballs and downloaded package files RUN conda clean -tp -y \ && apt-get clean \ && rm -rf /var/lib/apt/lists/* EXPOSE 50052 COPY . /root/ CMD ["python", "grpc_server.py"] # run a docker container docker run --rm -d -p 50052:50052 --name iris-predictor iris-predictor # run a client python iris_client.py --host 192.168.99.100 --port 50052

37

null

2017-12-12

2017-12-12 08:52:51

2017-12-12

2017-12-12 09:49:29

1

false

en

2018-01-08

2018-01-08 09:10:38

3

16ba4b9ea4ee

4.071698

27

0

As you know, microservice is a hot topic those days. We machine learning engineers should follow the trend to provide an “API” for machine…

4

Machine learning as a microservice in python As you know, microservice is a hot topic those days. We machine learning engineers should follow the trend to provide an “API” for machine learning as a microservice. In this article, I would like to describe a minimum structure of machine learning as a microservice with gRPC in python and docker. yu-iskw/machine-learning-microservice-python machine-learning-microservice-python - Example to implement machine learning microservice with gRPC and Docker in…github.com Step1: Train and persistent machine learning model First of all, we must train a machine learning model to put in a microservice. Here, we will make a classification model for iris data with scikit-learn. Since we will put the saved model in a docker image, we persistent a classification model. We simply use LinearSVC to predict iris species by given features. Step2: Define the protocol buffer gRPC is a modern open source high performance RPC framework that can run in any environment. It can efficiently connect services in and across data centers with pluggable support for load balancing, tracing, health checking and authentication. It is also applicable in last mile of distributed computing to connect devices, mobile applications and browsers to backend services. The service recieves IrisPrdictRequest that includes properties for sepal length, sepal width, petal length and petal width. Meanwhile, the response is composed of species. As you can imagine, it would be good to put probability in the response, if you would like to get it. We save the code to iris.proto now. Step 3: Generate python code for gRPC We have defined the protocol buffer for the microservice. Now we will generate the python code with the definitions. To do that, we make a code like below, where we save it to codegen.py. Executingpython codegen.py , we finally generate iris_pb2.py and iris_pb_grpc.py .Of course, we can also generate them with shell command. Step 4: Implemet the gRPC server in python We have generated the python code for gRPC. Now, let’s implent the server part. We trained and saved a classification model to a serialized file as a pickle. We load the saved model in the server part. In order to reduce the overhead to load the trained model, we use Singleton pattern like get_or_create_model. As well as, we just call the predict API to the trained model. Besides, we expose the service at 50052 port. Step 5: Implement the gRPC client Since this is an example to serve a machine learning model as a microservice, we implement a client to call the gRPC API. We make a request object with iris_pb2.IrisPredictRequest . Here, we uses fixed values as an example. Step 6: Bulid a Docker image We have implemented the trained model, the gRPC for server and client. We will put them in a docker image. Then, we build a docker image dockerwith build . -t iris-predictor . As we run the server with 50052 port, we must expose the same port number on the docker image. Step 7: Run a Docker container Congraturation! Now, we are ready to serve the machine learning gRPC server on docker. We run a docker container with the below command. After running a docker container, we check the server with the client we made. Summary I explained a basics of implementing machine learning as a microservice with gRPC and docker. I am sure offering machine learning microservice could be much more important. Expanding what I described, we can adapt it to practical versions.

Machine learning as a microservice in python

117

machine-learning-as-a-microservice-in-python-16ba4b9ea4ee

2018-06-09

2018-06-09 23:22:39

https://medium.com/s/story/machine-learning-as-a-microservice-in-python-16ba4b9ea4ee

false

1,026

null

Docker

docker

Docker

13,343

Yu Ishikawa

Machine Learning Engineer

11f93f9d3bca

yuu.ishikawa

141

75

20,181,104

null

0

null

0

1d067f6ac1b3

2017-09-27

2017-09-27 19:47:19

2017-09-27

2017-09-27 20:39:46

5

false

en

2017-11-03

2017-11-03 18:12:13

2

16bc76807864

4.652201

6

1

0

Why Androids Are More Trustworthy Than Humans

5

Operations: 2 Wednesday, 27 September 2017 Synthetic Intelligence Why Androids Are More Trustworthy Than Humans Androids have a bad reputation in science fiction. But in science fact, robot-human hybrids can be preferable to both the former and the latter. The trick is balancing the flexibility, intuition, and understanding of a human with the efficiency, reliability, and indefatigability of a robot. At Invisible Technologies, we provide synthetic assistants composed of human teams managed by a robot. With an algorithm at the helm, you can trust that the assistant’s output will match your desires, but with humans executing, you can trust that the assistant won’t get stumped the first time reality doesn’t match its instructions. With dozens of human agents recording their experiences over hundreds of task instances, the accumulated understanding starts to resemble a rudimentary brain — something capable of memory and learning. The robot brain directs the human arms, which update the robot brain, which then directs the human arms better — and you have yourself a synthetic assistant far more capable than any individual human or robot. What qualities do you look for in a normal human assistant? You want someone who: Never forgets a task, Always knows your personal preferences, and Never makes mistakes. How would you find these qualities in a human? You might trust a referral, their relevant experience, or just like the person in question — but they can never meet these promises one hundred percent of the time. Memory is fickle, notes get lost, people get sick, and their judgement is never quite as good as yours. And good luck finding an artificial intelligence that can understand all that! For a synthetic intelligence (humans plus robots), meeting this criteria is a simple matter of programming. Build a robot brain to prioritize the above qualities, hire humans to follow the robot’s instructions, and you don’t have to trust either — just the feedback loop itself. That’s the problem with human brains — you can’t see the mental programming that determines the person’s actions. You don’t know how they remember things, how they learn from mistakes, or how they pay attention to detail — you just see the results of those thoughts. Meanwhile, synthetic brains are 100% visible. You can peer into them and rewire however you like. Make sure input A creates output B. Take action Y when situation X occurs. Under no circumstances should you Z. It’s all right there! In the rest of this post, I’m going to offer you a peek inside the synthetic brains that currently power Invisible. Here are the dashboards our agents use to solve for all of the above. Instances Dashboard: Never Forget a Task Even a human can solve this one — just write down every task you get. At Invisible, we ensure agents record their task as part of the task itself. Example Instances Dashboard This dashboard allows the clients to confirm at a glance what their assistant is spending time on, which Capabilities (categories) of work they’re prioritizing, and how much time they are spending on each. Soon, recording this data will happen automatically via time tracking software, but before we automate anything, we always execute manually to ensure we understand all of the relevant pieces first. Context and Preferences Dashboards: Always Remember Preferences A common problem with virtual assistants is context transfer — how can you trust Agent A learns from the work of Agent B? Managing and coordinating humans is a job in itself, as any manager can tell you. Our Context dashboard stores all of the things an agent needs to know about the client. It doesn’t matter if it’s their birthday, the size of a conference room, or how to sort emails from their spouse. It’s the Single Source of Truth for everything we know about the client. Example Context Dashboard [Client Information Abstracted] Note the robotic commands, written in natural language. It’s easy to say: “If we book a flight, then we should note the client as Out of Office.” But it’s harder to know which of their 3 recurring family events per week we should block time for. Or whether that email from their co-founder should be labeled as urgent or just an FYI. This is where human intuition comes in handy. The agent can use the Context they have, in tandem with the commands the algorithm gave them, and make a decision that matches the circumstances. As long as we have the right process instructions written down, the agent will make the right decision. And if we don’t, the client will tell us, and we’ll record it as a new Preference. Example Preference Dashboard [Client Information Abstracted] Preferences come from client Feedback or Mistakes (both of which have their own dashboards). They’re essentially edits to the synthetic brain’s instructions — do this instead of that. They might add a new step, tweak an existing one, or reiterate something that wasn’t written before. By storing all of these idiosyncrasies in one place, the client can trust that as long as they input their desires into the dashboard, then their assistant will act the way they want. Mistakes Dashboard: Never Make The Same Mistake Twice Robots are the only ones who can truly promise they’ll never make a mistake. But that also means they can’t innovate or solve problems that aren’t addressed in the initial delegation. Our synthetic assistants can, which means they can promise the next best thing — no mistake made twice. We track all of our mistakes in the Mistakes dashboard, and every Mistake gets a Preference to match. That way, we can promise we’ll never make the same mistake twice, and stand by it. Every mistake updates the brain so it’s incapable of making it again. Example Mistakes Dashboard [Client Information Abstracted] You’ll notice not all mistakes are equal. Some are due to human error, a systems failure, or the cost of trying a new innovation. We categorize those as well, and this informs our product roadmap and agent training procedures. You can trust your synthetic assistant when we promise it will never make the same mistake twice. Just check its brain! Learn more at inv.tech Home

Synthetic Intelligence

97

why-androids-are-more-trustworthy-than-humans-16bc76807864

2018-05-16

2018-05-16 08:39:30

https://medium.com/s/story/why-androids-are-more-trustworthy-than-humans-16bc76807864

false

1,012

Synthetic Intelligence.

null

francis.pedraza

null

Invisible: Support

francis@invisible.email

invisible-support

SUPPORT,OPERATIONS,TECHNOLOGY,ARTIFICIAL INTELLIGENCE,CUSTOMER SERVICE

francispedraza

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Corey Breier

Author of ‘Life is a Game’ and ‘The Habitual Hustler’. More at CoreyBreier.com

3e214873fbd5

itscoreyb

1,173

1,394

20,181,104

null

0

null

0

null

2018-07-06

2018-07-06 14:43:51

2018-07-07

2018-07-07 13:44:52

14

false

zh-Hant

2018-07-09

2018-07-09 05:30:48

4

16be1c45364c

1.687736

17

1

0

在上一篇的文章（從「框架語意模型FSR」到「多模態設計」）最後，我們提到了抽取用戶在旅程中的意圖，這一篇則是開始更深的撰寫對應每個意圖中的LAR細節，以及CUI該用什麼樣的方法被測試。

5

[MIX2018] 語音交互工作坊(下)：從「LAR表」到「CUI的測試方法」在上一篇的文章（從「框架語意模型FSR」到「多模態設計」）最後，我們提到了抽取用戶在旅程中的意圖，這一篇則是開始更深的撰寫對應每個意圖中的LAR細節，以及CUI該用什麼樣的方法被測試。前情提要完，以下為正文開始為逐字稿正文，如果你能看到這邊，那請收下筆者的膝蓋，你對CUI真的很有愛RRRR。圖1：一張LAR表為該意圖的一個最小單位，一個意圖完成可能要數張LAR表才能完成第一階段的工作坊，我們主要是找出場景不同的流程，不同的維度，用戶可能會產生的一些需求點。第二個是我們去思考在這個場景下用戶可能會需要的硬件能力去跟用戶產生交互。剛剛工作坊的部分感謝所有參與者都把用戶可能的「意圖」都抽取出來了。最基本的事情，除了抽取「意圖」外。在撰寫LAR表時也要注意以下四大點，分別是「1.多任務」、「2.一致性」、「3.語境意圖」、「4.非核心插入」 1.「多任務」即是在同一個流程節點中，用戶可能需要同時做的行為，例如在導航的時候用戶可能要使用「問詢」也需要「操作觸屏」去看所在的地理位置。 2.「一致性」則是語音回覆的形式是否有一致，不會一下子很有禮貌，一下子又給人感覺很粗魯。 3.「語境意圖」則是在語音交互的時候會有一些上下文的關係，這樣的上下文關係我們如何幫助用戶完成需求。 4.「非核心插入」在GUI的領域大家比較不常碰到，但是在VUI中可以說是天天出現，用戶在主線上產生的支線意圖，我們就叫做「非核心插入」。比如用戶進入肯德基POS的一個問詢機，用戶說「我今天想點一個炸雞」、但今天如有一個用戶中間說「我想要麥當勞的炸雞」或是「我不想點炸雞了」，這樣子搗亂的情況是成立的。在語音交互的狀況下，會有很多干擾的意圖會產生，這些狀況設計師要考慮進去。圖2：多任務中的「問答操作」、「內容查詢」、「服務應用」第一，先來講「多任務」，為什麼我們會說「多任務」很重要。在多任務下我們目前分了三個類別分別是一問一答的「問答操作」，像是「天貓精靈你今年幾歲？」、「你現在心情開心嗎？」「我現在心情很不好」之類的，想必大家都很能理解。第二是「內容查詢」這個東西跟問答有點像，李白是誰？杜甫是誰？周杰倫是誰？這邊要特別注意，當我們把資料送給用戶的時候要注意長度。根據阿里巴巴自己的研究，用戶聽了大概十三秒就沒有注意力了。故此，內容查詢最大的重點就是，如何在13–18秒裡用精華的訊息放在很簡短的文字約20–30字的訊息裡面給用戶。當超過30字以上的內容，用戶的心智負荷就會往上飆了。也就是說當天貓精靈在做新聞內容的時候，大家都以為只是把網路上的文字播放出來，其實阿里巴巴會有兩個人員專門為每日頭條的新聞內容做濃縮編輯，去把他做成適合語音播放的內容。。第三則是服務應用，服務應用是最難且最有價值的一個東西。比如說我今天要訂餐點，服務應用不是一問一答，而是把時間的維度加進去進行有意義的交互。從需求產生到需求完成，他會需要不間斷的跟機器進行交互，所以機器就必須要在下一階段帶給用戶下一步驟的信息。你可能需要音箱的播放、IOT家電的操作。這就是LAR中的ACTION所要作的事情。當用戶的操作是錯誤的時候，你必須在ACTION的地方把用戶拉回來正常的邊界上。設計師必須很明確的跟用戶說「第一你無法做什麼？」「第二你可以做什麼？」透過這個方式你才可以把用戶往下一個步驟帶。如果你不這樣子引導用戶的話，用戶就會開始產生「非核心插入」，故此他的需求就沒辦法完成。圖3：在撰寫LAR時要注意語境的上下文關係 (圖片來源：主講人同意授權筆者重新繪製) 在語意語境下，又會繞回來講到設計語言的INTENT，身為設計師的我們必須先行判斷。ACTION方面，則是基於這個意圖我們要如何回覆？我們在ACTION步驟的判斷上，需要去判斷他是在流程的前還是流程的後？在REPLAY這個步驟，設計師才能回覆用戶，到底要讓用戶進行下一步驟的引導還是附加能力的引導。所以LAR其實可以很好的將設計師的邏輯發揮出來。 LAR只是意圖的小小點，MAPPING FLOW其實就是把LAR組成更大張的流程圖。他對應到GUI就像是我們服務設計常常在講的UJM，只是我們把UJM邏輯化了，技術表達式化了。圖4：跟真人聊天不會出錯另一個要特別注意的是，在常人對談的範例中不會出現不可分的狀態，在一般人類的交談過程中，我們不會出現系統錯誤。圖5、6：CUI容易因為硬體或技術上發生錯誤之外，故設計師可以在細節之處塑造語音個性的情感化不可分的狀態用GUI的例子來舉例就是不會出現404的狀態。這件事情會在GUI上成立，但是在CUI上面不成立，這到底是為什麼？因為在CUI上用戶並沒有看到你所提供的服務內容，所以他所操作的任何事情都是成立的。可是我們可以讓用戶在超出我們設計邊界的時候提供其他操作把他引導回來，例如「你說的這個服務我目前不了解」、「你說的這個服務我目前不支持」、「我目前支持哪些服務」，用簡單的語料把讓用戶拉回來。所以在語音交互裡面沒有所謂錯誤，如果錯誤那就代表設計師沒有做很好。如果當你因為硬體、軟體的問題發生錯誤，你可以透過回覆、用戶意圖的理解去讓用戶完成他要想做的事情。圖7：非核心插入是用來輔助場景資訊的，常見的有「逃脫」、「查詢」接著語音交互還有另外一個重點是「非核心插入議題」，這些又牽扯到場景輔助資訊。我們又拿高鐵的例子來說明，語音交互畢竟他「不可視」，再沒有屏幕的狀況下，用戶會想要有「我剛剛買了幾張票」。再者用戶覺得語音真是太傻了想要找「人工櫃台時」，這就是我們俗稱「逃脫」的非核心插入。這時候用戶想要離開該交互時，我們是否能把他導入到其他功能，讓用戶做自己想要做的服務。圖8、9：語音交互有更多細節的地方需要打磨，如「擬人化」、「劇本化」等目標工作坊現在要接近尾聲，去年我們阿里巴巴在跟大家分享什麼是語音交互，今年我們更多在著眼努力「如何將語音交互達到擬人化的過程」。目前為止阿里正朝向讓語音更有人格，讓用戶產生更多的情感價值連結。例如在不同場景中，我會需要一位很聰明的女性助手，我們要如何把女助手的特徵給抽取出來，做成一塊回應模版，去加入到我們的硬體設備中。所以今年我們將做更多關於語音行為模式的設計，「讓語音聽起來像誰？」甚至是把GUI中的「個性化」導入到語音交互來，隨著你使用次數用來愈多，利用演算的方式去越來越符合你的內容跟需求。像是在淘寶APP中，我們會隨著你查詢的商品種類，進而在APP首頁中的商品資訊中去算出推薦適合你的產品。圖10：在語音交互的所有測試中，成本最低最快的就是Table Reading 最後有人在問，CUI到底要怎麼測試？其實CUI發展至今有非常多的測試方式，但傳統在GUI上所用的放聲思考法（Think-Aloud）是行不通的。業內最常用的有巫師觀察（Wizard of OZ, 綠野仙蹤法）、CUI 評估啟示（CRT）、可用性測試。其中Table Reading是成本最低的一種檢查方式，他能達到二種效果分別是檢查設計師自己撰寫的內容是否有通順？是否有過多的贅字？另外則是語句是否言不及義？在知識結構的邏輯上是否有不合理的地方？另外一個業內最通用的方式則是巫師觀察，他其實有點像GUI中的眼動儀，觀察者與受測者分別坐在不同的房間，觀察者使用電腦監控受測者去捕捉用戶的表情狀態遲疑，當用戶說了一聲「ㄟˊ」的時候。這就算是語音交互體驗不好的一個訊號。只是很可惜今天因為時間的關係沒辦法講太多，希望大家都能收穫滿滿的回家，謝謝。圖11：根據每個意圖，我們的組員撰寫的LAR表格在撰寫的過程中，冠芠還跑過來跟我們說要小心「非核心插入意圖」，因為這件事情是新手很常犯的錯誤之一，但沒想到我們也毫無意外的踩了新手的坑XD。圖12、13：Table Reading測試方法最後的測試部分，主持人請台下其他組別成員上來踢館進行TABLE READING的測試，我們組內沒三句話就GG了(淚)。圖14：小組成員最後留下來拍張快樂的大合照（以上三篇文章系列的照片CREDIT(Miguel Wu、Justin Six、Vivienne、若羚、VIK提供） CUI領域的東西真的頗多，整場聽下來腦子都會熱熱的。但我們組內成員感覺都玩的很開心，希望這次工作坊結束之後大家在UX領域也能經常見面。 — — — — — 全篇整理完畢，真是累死我了QQ — — — — — 若想要再燒腦溫習一次可以連結下列文章前往： [MIX2018] 語音交互工作坊(上)：從「人工智能產業概況」到「CUI交互特徵」本逐字稿為湯六參加MIX2018年會的其中一場工作坊，因為工作之故所以接觸到語音UX的範圍，但歐禮萊的書看完覺得「有看沒有懂」，在工作應用上有點騷不到養處，故此想藉著參加工作坊與繕打逐字稿的歷程，幫助自己把語音交互的結構梳理一次。medium.com [MIX2018] 語音交互工作坊(中)：從「框架語意模型FSR」到「多模態設計」上一篇文章中提到了人工智能概況，這一篇則是開始從技術觀點著手，探討語意模型FSR以及用設計的觀點去理解INTENT中的Listen、Action、Replay代表的涵義。medium.com [MIX2018] 語音交互工作坊(下)：從「LAR表」到「CUI的測試方法」在上一篇的文章最後，我們提到了人工智能概況，這一篇則是開始更深的撰寫對應每個意圖中的LAR細節，以及CUI該用什麼樣的方法被測試。medium.com

[MIX2018] 語音交互工作坊(下)：從「LAR表」到「CUI的測試方法」

129

mix2018-語音交互工作坊-下-從-lar表-到-cui的測試方法-16be1c45364c

2018-07-09

2018-07-09 05:30:48

https://medium.com/s/story/mix2018-語音交互工作坊-下-從-lar表-到-cui的測試方法-16be1c45364c

false

63

null

Voice Ui

voice-ui

Voice Ui

74

Tom Liou

null

5ecaa446b4da

tomliou

237

114

20,181,104

null

0

null

0

null

2018-01-28

2018-01-28 21:10:48

2018-01-29

2018-01-29 14:15:04

5

false

en

2018-01-29

2018-01-29 14:15:04

10

16be8f97b251

5.580503

1

0

Last night, President Barack Obama’s parting words “Be Kind & Be Useful” struck a chord with me and my daughter. This morning, it led me to…

5

Labor & Automation’s Symbiotic Relationship Last night, President Barack Obama’s parting words “Be Kind & Be Useful” struck a chord with me and my daughter. This morning, it led me to buy a homeless man gloves and rethink how we adopt technology. I would like to controversially suggest that Asimov’s fictional laws be amended to include this sentiment. The adage could be a guiding principle on how we deploy automation and accept its consequences. Today there is a fight brewing between organized labor and multinational companies embracing automation. The Teamsters Union is currently renegotiating its contract with UPS which involves 260,000 workers. The Wall Street Journal reported that the union is “seeking to prohibit UPS from using drones, driverless vehicles, and other new technology to transport, deliver or pick up packages, innovations that would curb the demand for labor.” UPS in turn is demanding a new flexible contract to “remain a highly competitive and reliable service.” In the backdrop of the 83-page working document is a swarm of upstart delivery services, including Uber Trucking and Amazon Prime which fully embraces autonomous shipping. To further complicate matters is the billion dollar growth of online sales that often requires deliveries late into the evening, which the union also plans to ban. This conundrum doesn’t address the inevitable that computer-aided controls like the Tesla model below will be delivering packages within the next decade, as seamlessly as autopilot programs have been flying planes. The defensive attitude of the union leaders fails to fulfill the last phrase of Obama’s directive for its members, “be useful.” James Hoffa, Teamster President, would be wise to look overseas for progressive inspiration. Wolfgang Lutterbach of the German Trade Union Confederation explained to Politico last year, “If we don’t speed up our reactions to these developments [automation], we will not shape digitalization, but digitalization will shape us.” Leaders like Lutterbach are focusing their efforts on bolstering the “social partnership” between worker and employer by directly tackling the challenges of advanced technologies. Rather than skirting the issue through boycotts of automation, German union leaders are demanding new agreements to ensure retraining or “upskilling” of workers, so labor will play a vital role in the implementation of future innovations. It is without question that robots will replace workers in large numbers, however Germany (like many first work countries, including the United States) will also see a dramatic demographic shift with 33% of its population aging and retiring in the coming decades. The key lesson of Lutterbach is to empower younger workers today with the skills to succeed tomorrow. The New York Times reported last December how Sweden’s social welfare system is accelerating the adoption of technology, not hindering it. According to Swedish minister for employment and integration Ylva Johansson, “In Sweden, if you ask a union leader, ‘Are you afraid of new technology?’ they will answer, ‘No, I’m afraid of old technology,’” Johansson further explains, “The jobs disappear, and then we train people for new jobs. We won’t protect jobs. But we will protect workers.” The Times article follows Boliden mine worker Mika Persson as he drills for zinc and silver. Instead of finding an exhausted, filthy human from a day of inhaling dust and fumes, the readers meet Persson in front of a computer controlling a remote loader from the comfort of his desk. When asked if he is worried about losing his job to robots, Persson responds, “I’m not really worried. There are so many jobs in this mine that even if this job disappears, they will have another one. Boliden’s website heavily promotes its race towards automation with boasting of “world class productivity” due to its sizable investment “in maintenance technology” to “modernize and streamline the design, planning and management” of its resources. Boliden currently pays its Swedish workers higher wages than other countries and its ore is often the lowest grade, but with new unmanned systems its profitability continues to rise from greater extraction efficiency. According to the Times, in Sweden “Robots are just another way to make companies more efficient. As employers prosper, workers have consistently gained a proportionate slice of the spoils — a stark contrast to the United States and Britain, where wages have stagnated even while corporate profits have soared.” The contrast between the United States and Sweden could not be more glaring in terms of how each population feels about technology. According to an European Commission survey last year 80% of Swedes had positive perceptions about AI and robotics versus a similar Pew study that showed 72% of Americans are “worried” about a mechanized future. Unlike Swedish welfare benefits, most workers in the United States rely solely on their employer for healthcare and pensions. While unions in Sweden see automation as a competitive advantage for their country, in the United States tech-protectionist beliefs like the Teamsters are understandable, but not defensible long-term. Widening the welfare safety net to keep pace with automation became a central issue this past fall with the drafting of the new tax bill. Congressman Ro Khanna of California’s 17th district (Silicon Valley) knows firsthand the impact of driverless vehicles on the economy, as his constituents include such tycoons as Google, Tesla, Uber and Intel. Khanna exclaimed, “I think this [automation] is exactly the challenge for the country is how do we navigate the transition and make sure that people are participating in it. I don’t think companies sit there saying, ‘How can I create jobs?’ This is where the role of government is so important, because the government — which is the people — has some social context. We think about, ‘What is it gonna mean for healthcare? What is it gonna mean for job creation? What is it gonna mean for people being displaced?’” Khanna proposed increasing the Earned Income Tax Credit for all Americans, especially for low to middle-income earners. Khanna’s $1.4 trillion proposal did not make it into Trump’s final tax plan, but his advocacy possibly influenced the Republican push to increase the Child and Dependent Care Credit. Washington’s new tax bill could be the greatest opportunity for labor. Corporation just received one of the largest corporate tax cuts in history, taking their variable rate from 35% down to 21%. In addition, multinational companies will be repatriating tens of billions of dollars over the next few years. Rather than forbidding unmanned technology, Hoffa (and other Union leaders) should embrace it. Demand that UPS (and other companies) use the billions of dollars of tax savings to upskill workers, providing them first mover advantage to autonomous technology skills. The Obama administration already proved that former coal-miners in Paintsville, Kentucky (Appalachia) can master software coding. Imagine if the new Teamster contract enabled older brown-suited drivers to retire with dignity, passing the baton of progress to younger, newly-educated managers with fleets of deliveries at their fingertips? My suggestion to UPS CEO David Abney is to listen to the words of Alibaba founder Jack Ma at Davos, “I think AI should support human beings. Technology should always do something that enables people, not disables people. But we [tech companies] have the responsibility to have a good heart, and do something good.” The topic of labor & robotics will be discussed more at our quarterly RobotLab forums, we welcome any union leaders to join the panels. Our next Meetup will be on March 6th in New York City covering “Healthcare & Robotics” with Dr. Joel Stein of Columbia University and Kate Merton of JLabs (Johnson & Johnson’s Incubator) — RSVP Today!

Labor & Automation’s Symbiotic Relationship

5

labor-automations-symbiotic-relationship-16be8f97b251

2018-01-29

2018-01-29 20:37:27

https://medium.com/s/story/labor-automations-symbiotic-relationship-16be8f97b251

false

1,258

null

Technology

technology

Technology

166,125

Oliver Mitchell

Founding Partner of Autonomy Ventures a seed stage fund investing in robotics, AI, and smart devices. Oliver is a chronicler of robotics on RobotRabbi.com.

f7f49cbdf60f

olivermitchell

1,402

285

20,181,104

null

0

null

0

b10cb7c59f83

2017-11-28

2017-11-28 04:59:19

2018-02-15

2018-02-15 07:49:02

4

false

en

2018-02-15

2018-02-15 07:49:02

9

16bebac5ac1d

2.220755

1

0

We all know how important tests are when maintaining a large code base. Testing code libraries is standard, but the same isn’t true for…

1

Code testing our lectures This badge tells you whether there’s any execution errors in the lecture! We all know how important tests are when maintaining a large code base. Testing code libraries is standard, but the same isn’t true for lecture sites or on-line books like ours. At the same time, our site contains a lot of code and we understand that without testing there’s no way we can guarantee quality. So, after plenty of hard work by members of our team, testing execution of all QuantEcon lectures began in November 2017. We run the testing tool daily to determine whether each lecture runs without error. With this tool we can ensure that our lectures are running smoothly and up-to-date with language and package updates. The updated QuantEcon lectures homepage Using information gathered from our testing tool, we’ve added badges to the homepage to indicate the overall execution status of the lectures. A figure below 100% indicates at least one lecture is not running without error. The status page lists the execution status of each individual lecture. Our aim is to have all lectures running smoothly at all times, and the badges alert us to when this is not the case. The other benefit is you can check to see if the code is running on our machine, which can help you diagnose any issues you may be having when running the code locally. The lectures status page How it works Example of email notification produced by the code tester First, code from our lectures is converted into Jupyter notebooks using Jupinx, a tool developed last year by the QuantEcon team. The testing tool runs the generated notebooks nightly on our build server (hosted on AWS) and looks for any execution failures. Errors trigger an alert to the QuantEcon team, allowing us to fix them as they occur. The server environment is updated periodically to ensure we are using the latest software. Further development Currently this code checker is only looking for execution errors. It does not catch logical errors that may exist in the code. As always we welcome any feedback and questions on the lecture site code. The best way to do this is through the QuantEcon Discourse Forum. We also plan to document and open source this tool in the coming months. We hope this tool can be beneficial to similar projects and encourage maintenance of online resources. Acknowledgements The code testing tool was developed by Matthew McKay, Akira Matsushita and Nick Sifniotis over 2017. We are grateful to the Sloan Foundation for supporting the QuantEcon project.

Code testing our lectures

31

code-testing-our-lectures-16bebac5ac1d

2018-04-11

2018-04-11 23:09:42

https://medium.com/s/story/code-testing-our-lectures-16bebac5ac1d

false

403

Open source code for economic modeling

null

quantecon

null

QuantEcon Blog

contact@quantecon.org

quantecon-blog

ECONOMICS,COMPUTER SCIENCE,SCIENTIFIC SOFTWARE,COMPUTATIONAL SCIENCE,ACADEMIA

quantecon

Data Science

data-science

Data Science

33,617

Natasha Watkins

Pre-doc at QuantEcon

14bff4ca9568

natashawatkins

52

27

20,181,104

null

0

null

0

a995c24848a3

2018-08-29

2018-08-29 03:51:18

2018-08-29

2018-08-29 17:21:01

2

false

en

2018-08-29

2018-08-29 17:21:01

3

16becafb650a

2.153145

1

0

Last year, we introduced Box Skills, a framework for applying the world’s best AI from providers like Microsoft Azure, IBM Watson, and…

5

Apply custom-trained AI models to Box with the Box Skills Kit Last year, we introduced Box Skills, a framework for applying the world’s best AI from providers like Microsoft Azure, IBM Watson, and Google Cloud to your content in Box. With Box Skills, businesses can apply pre-trained algorithms from these providers to better organize, understand, protect and automate processes around their content in Box at scale. You can read more about Box Skills in our announcement here. But many enterprise use cases for AI require a level of intelligence beyond the capabilities of these pre-trained algorithms. Oftentimes, businesses need a deeper level of understanding, like labeling proprietary brands or products in marketing images or recognizing unique keywords or phrases in audio transcripts. Over the past year, many of the leading AI providers in the space have introduced sophisticated tools that enable businesses to create and train customized AI models, like a custom computer vision model that recognizes their products in images. These tools enable you to train and tailor pre-trained algorithms to solve for your business’ unique data and processes and even create entirely new algorithms. Today, we’re excited to announce support for custom AI models, created using services like IBM Watson Studio, Google Cloud AutoML, Microsoft Azure Custom Vision, AWS SageMaker and others, via the Box Skills Kit. With this support, you’ll be able to integrate custom-trained AI models, tailored to your specific needs, and apply them to recognize and understand your unique business content in Box with a custom skill. This allows you to search your content based on terms and concepts, as opposed to based on file names or more specific data. These metadata values, applied via the custom skill, can also drive other Box functionality like automations and governance policies. Box’s search results will display rich media results for content that has been enhanced by Box Skills The ability to more deeply understand and extract information from business-specific data unlocks new use cases for Box Skills, such as: A construction equipment rental company being able to apply customized image labeling services to automatically recognize and label specific equipment and part numbers in images captured and uploaded by field sales engineers A media & entertainment company being able to recognize and label specific brand IP, like animated characters, in image, audio, and video content to make marketing assets more structured and easily searchable A financial services company being able to automatically recognize and classify different types of forms uploaded as part of a client onboarding or loan origination process Starting today, you can apply these customized AI models to content in Box by building a custom skill using the Box Skills Kit, which will be generally available in December 2018. You can visit our website to learn more about Box Skills and request access to the beta access for the Box Skills Kit on our developer site.

Apply custom-trained AI models to Box with the Box Skills Kit

1

apply-custom-trained-ai-models-to-box-with-the-box-skills-kit-16becafb650a

2018-08-29

2018-08-29 17:21:01

https://medium.com/s/story/apply-custom-trained-ai-models-to-box-with-the-box-skills-kit-16becafb650a

false

469

Cloud content management APIs

null

Box Developer Blog

platform@box.com

box-developer-blog

DEVELOPER TOOLS,CONTENT MANAGEMENT,API,TECHNOLOGY

boxplatform

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Box Developers

null

7e4cb8ebf639

Box_Developers

176

3

20,181,104

null

0

null

0

null

2018-02-08

2018-02-08 15:54:26

2018-03-26

2018-03-26 12:40:04

16

false

en

2018-03-26

2018-03-26 14:56:13

0

16bf698a2ab0

12.138679

2

0

There has been a ton of things going on in my world which is why its taken me a few days to get this blog post out. Normally, I can afford…

5

The Journey — PART 7— Blockchain Explorations There has been a ton of things going on in my world which is why its taken me a few days to get this blog post out. Normally, I can afford to spend more time but the life of a CEO is filled with countless things to handle. Operators, marketing, product, finance, legal , investors— just about anything really. But that’s neither here nor there. The purpose here is to talk about some of the explorations in blockchain from the early days to the present. Early Days The early days is what I considered more exploratory on what blockchain actually was and how it could be applied. In one of my earlier blog posts, I eluded to starting Animusoft and having every computer in the office mining for crypto. This was really just developer / computer science mental hockey digging into a random tech. Even mining on my tablets and mobile phones. Watching it daily to see how it would perform while performing comparison between different CPUs. Even putting it on Raspberry PI’s to see if that was any different from the others. My favorite part about this time was that it was very pure. Nothing worked right, the code was bare, very little was known about it let alone who would be using it. Back then, I was writing up scripts and code to actually kick off processes for mining. But really, I would call it more of just poking around kicking tires to see what everything did. If any of my readers out there wants an idea maybe now is a time to pay attention. Agriculture At Animusoft during 2016, we spent much of our efforts figuring out how drones and Alive could be used to help out farmers world wide. Quad-copters focused on crop inspections and land surveys. While this was happening its as if I had many other background threads running digging into the details of anything and everything farming as it would relate to the blockchain. All the while, finding the valid cases for drones. Kind of funny how the mind works on this. First understand the lens used to look at this problem is comprised of the following filters Decentralization Transparency World Computer / The Fog Now some of these ideas just touch the surface of the thoughts I had back then about how blockchain could potentially play a role. Farm to Table Intimately understanding all the phases produce goes through from tree to table was something I had to pay close attention to. This helped me understand how drones could ultimately benefit everyone in the space. When it comes to blockchain, thoughts came to mind about tracking each fruit or veggie down to a transaction on the blockchain as it exchanged hands. What this gave insight into is who touched the produce along the way and what was the duration of time between touches. More importantly, the consumer who eats this produce would know exactly when this produce was removed from the tree. Even more so, how many times was it watered, how much sun did it receive, how many times was it sprayed with insecticide or fertilizer. The problem faced was really around how does one even keep track of this. It would almost require some sort of RFID tag / reader combo per piece of produce and integrated into a completely connected farm. Another way could be to use computer vision by means of cameras installed on workers, tractors, sorting lines semi-trucks, distribution centers, and supermarkets. None of this is even practical. All of us need to understand, being out in a farm is not necessarily a place where some of these IoT devices would even live beyond a few weeks. Things as simple as a tractor knocking it over, someone stealing it and even weather conditions play a big role. Moreover, farmers don’t have 1 or 2 acres of land they have 100s or 1000s. I soon shelved this idea for another day when any new tech may be able to tackle this issue. If one of you out there has a solution that really is practical — by all means — contact me directly. My father is a farmer with plenty of acres of land with fruits and veggies. I would love to help him out with this as well as the rest of the farming community with this sort of attention to detail on the farm. Produce Inspections This almost carries over from the previous. Inspections really happen at different levels in farming. First, the produce picker inspects it on the spot while on the tree. The worker picks it and puts it in a bucket where it gets sent off to a packing house. Then it gets dumped into a sorting conveyor belt and gets inspected for quality yet again. Don’t forget, that in some states and countries, the failed fruit goes to a different line where it can be sold same day at a far cheaper price. For example, on the side of the road. Next, is it passes into a sizer (sometimes) and boxed up. Then government inspectors come and “spot check” the produce to see if it meets the standards imposed. Once it passes there, then its off to distribution centers and super markets. Upon arrival, sometimes the produce is inspected on the spot randomly. Once it reaches the shelves of the supermarket, the worker there inspects it once more and does their “product placement” maneuver to ensure things look accordingly. It’s at that point that shoppers inspect to their liking. The same issue arises here as it did in the previous example. How on earth do you watch a single fruit or veggie to track its progress. Its very overwhelming. Its not to say that it can’t be done but it has to be done in a very cost effective manor. Even putting a sticker on each fruit is time consuming for the packing company that packs the produce from the farmer let alone the cost incurred by adding this sticker. Now, if there was some sort of technology that could do this, then decentralization and transparency would totally apply here in a great way. But, before we can do that, this really needs to be solved first before we move onto that. Coins by Produce Now here is where it might get a little interesting. What if there was a coin for each type of produce (e.g. Mango, Lime, Avocado)? What if this was traceable on a blockchain? This would indicate many things. For one, we would know exactly what the markets minute by minute. Are projections on track or is it off here or there? I can honestly attest here that I have never done any sort of trading with commodities but if there is a broker out there — start thinking how this might be applied. If it could be tracked at the token level for each and every fruit, veggie, pallet or box that might start to get interesting. Furthermore, if you use the same tokens to purchases these one would truly know down to the last detail who in fact is buying, how much they are buying, when and where they are buying the produce. Farming Equipment & Supplies Stemming from the idea of coins by produce, came the idea of tracking supplies and equipment on the farm. This at first seemed logical but then quickly stemmed out to be broader in nature than just farming. Much of the same equipment is used across industries. Furthermore, does this really help the farmer or the consumer? Or is it just another tool for the companies selling into agriculture? I will be honest here and say that I didn’t give this one much thought since my focus was on the farmer and the consumer more so than those selling into the agriculture space. Would be interesting to see a coin on oil, gas, rubber, diesel, dirt, fertilizer, seeds, much like that in the previous example. Illness / Diseases Now, if we had some tech that could watch every tree — instead of every fruit — this might actually be interesting. The problem here is, why does this even need blockchain to begin with. I couldn’t really find a valid reason why something decentralized and transparent here was needed. One of the things we did with Dell and local farmers using Alive and quad-copters was inspect thousands of acres of Avocado trees for a particular illness. When we started getting into 1000s of acres things got interesting. One soon realizes that farming is an art and a science. Not everything is as precise as one would think. Its more of a educated guess sometimes to know what to do next. For example, how much more water? Or even better, farmers working with researchers to figure out which treatment to give a plant or tree in order to treat the disease. The conclusion here is that illness and diseases are all over the place — I suppose that if there was a way to track every single piece of produce on a tree in a cost effective manor then this would actually be applicable. Possibly in a decentralized blockchain solution with its own token. But then the next piece lies within the miners and the customers. Would they even care to buy something in a given coin or not? Would they want to mine this and is there money in that? Farming is already tight as it is here in the USA with tough competitors abroad that can employ low wage laborers. Computer Vision and Machine Learning — Its Role Thus far, I have spoken specifically around making a blockchain solution as it applies to industries. The part I do not want you to get confused by is the different angles blockchain has. In farming as in other sectors, I was approaching it in a somewhat narrow mindset. Being that — how can I make a coin, a blockchain, benefit miners while putting something relevant behind — like a piece of fruit. Tracking that piece of fruit on the chain itself with a specific ID assigned to that individual piece of fruit. That’s not to say that a different type of blockchain — like VOSAI — would apply to this. For example, VOSAI does a boatload of computer vision and machine learning powered by its own blockchain with its coming world of miners. This is very applicable to all aspects of farming as well as many other industries. Camera’s powered by VOSAI out in the field would be able to provide a wealth of information at a very fast pace. Unlike never seen before. These could easily be installed wherever needed. The most practical of applications for farming and computer vision / machine learning would be using aerial drones (e.g. quad-copters) as well as IoT devices on tractors or in packing houses powered by VOSAI. Construction Construction applies to many sectors and industries. High rises, stadiums, hospitals, highways, bridges, airports and basically anything else that can be constructed. With Animusoft, we dug into the Construction space in 2017 digging in with the Alive platform, quad-copters and ground rovers. We found a ton of valuable use cases there when it came to applying robotics and Alive in construction. Ultimately, driving it down to what was needed in that space. Construction is really virgin ground when it comes to innovation. Hardware and software companies are rushing in to make the next widget or component that helps solve some sort of problem. Now, this post is really focused on the blockchain lens and not so much on the robotics lens. Therefore, I’ll do my best to stay on track here. Now throughout the course of exploring the construction space — we had numerous conversations. Drones, machine learning, computer vision and yes even blockchain. How could blockchain apply and be used? At first the common thread was why even bother — with the volatility of Ethereum they just didn’t see a solution. We did help those around us and steer them into the right direction when it came to thinking about this. Those that listened quickly readjusted and focused. However, it is a taunting thing to take on especially when there is so much confusion in the space already. Projects We looked into every aspect of a project. Starting from ideas and drawings straight through to who would actually be using this. Meaning — if it was an apartment — what would all the steps be to get someone moved in. From ground breaking to the last door knob. What we found is this. Cash flow is king and this definitely applies here. There are certain personas that take part in any project at the highest of levels and for the purpose of this post lets look at the interactions between Owners, Developers, and Contractors. Owners own the project. Developers are hired to design, plan, engineer, and hire contractors to build out the project. Owners pay out Developers upon work completion and in turn Developers pay Contractors once they are paid. Now on larger projects, there may be buffers here and there to help alleviate problems with cash flow and there are also incentives for work done faster or better. This all varies between project and is not the same across all interested parties. Now, there are contracts in place that say payment terms may be NET15, NET30, NET90 and so on. For those of you not familiar with this — this means you get paid in XX days respectively (e.g. NET15 means 15 days). Now this is all fine and dandy if it was all a reality. The truth is this. If I am a Contractor and finish the work I was contracted to do — it must first be inspected, checked off by superintendents, legal, finance and so on. This means that many hands need to exchange information which ultimately slows down the process. This means that NET30 could actually mean 120 days in some cases. The problem herein, is that the Contractor cannot always afford to wait around that long. There is an understanding in the construction space that good Contractors go out of business because of this very problem. They simply cannot keep their sub-contractors on staff let alone support this. Now I am sure there are many more issues here that I am not aware of. What I can say — this seemed to be a good candidate for blockchain. Mixed in with smart contracts this could streamline the process. We did explore this a bit with some of our customers at Animusoft to see if somehow the Alive platform with drones would help alleviate this issue. The issue we found here is that every Developer is different in how they operate in this area. Some use an older method that is paper heavy and very few have parts of this automated. There simply is no consistency. Without consistency its hard to attack this problem. Throughout our efforts, we did cross paths with IBM on numerous occasions and they are working on some solutions in the UK to help address this issue. But, as stated previously, this varies so much right now that its hard to see a proper one solution fits all. As a side note to this I will mention that the entire construction space in terms of software is an utter mess. There are too many tools without any standardization going on. There are some efforts — but if at best — its all pieced together with strings, tape and glue. The folks in the valley are right, Construction is ripe for disruption and there is a lot of things happening in the space. Definitely something to keep an eye on in the next decade no doubt. Pattern Matching Now there is one message that was booming in mind throughout all of these explorations. It kept popping up at every corner I looked. It was not directly related to the thing at hand. Rather, it was just a common pattern I noticed over and over again. Today, miners are innovating on many different forefronts. Algorithms design and hardware configurations or creations. These pattern match against the same thing in machine learning and computer vision. For years, ML/CV experts have been building custom workstations and servers with tons of horse power. Some of these machines cost upwards of $15,000 USD. They are loaded with CPUs, GPUs, memory and storage. All the things you would need to mine the blockchain right? Essentially the pattern is that mining and machine learning need similar if not the same hardware infrastructure to accomplish their goals. What would happen if these two could be combined into one?

The Journey — PART 7— Blockchain Explorations

6

the-journey-part-7-blockchain-explorations-16bf698a2ab0

2018-03-26

2018-03-26 19:03:25

https://medium.com/s/story/the-journey-part-7-blockchain-explorations-16bf698a2ab0

false

2,806

null

Blockchain

blockchain

Blockchain

265,164

Daniel Rodriguez

A Computer Scientist from a pan-dimensional, hyper-intelligent species of beings who spent far too many years meddling with AI.

cf3d5e577cbe

vosai

7

18

20,181,104

null

0

null

0

4c5e8887502e

2018-06-05

2018-06-05 13:42:57

2018-06-06

2018-06-06 00:46:01

5

false

pt

2018-06-06

2018-06-06 00:46:01

4

16bfa4038ecc

8.33522

10

0

Esse texto foi adaptado da minha apresentação para o TEDx Unesp Bauru, que até o presente momento ainda não foi ao ar.

4

Porquê Robótica e Inteligência Artificial devem ser discutida como políticas públicas? Esse texto foi adaptado da minha apresentação para o TEDx Unesp Bauru, que até o presente momento ainda não foi ao ar. Olá, meu nome é Mateus e antes de mais nada sou homem, branco, hétero, cis e cresci em uma família católica e de classe média/alta. E por que isso importa? O privilégio que tenho e que vou carregar para o resto da minha vida é o principal motivo de ter as condições de falar sobre esse tema e meu principal combustível para trilhar um caminho de transformação social por esse mundo afora, para devolver para a sociedade tudo o que pude receber. E são essas palavras “Transformação Social” que talvez sejam minha melhor herança … Mas esse não é um texto para falar sobre os meus privilégios, nem discutir igualdade de oportunidades, de gênero, sexual, etc. Eu acredito que existem pessoas muito mais capacitadas que eu para realizar essa tarefa. Então, antes de ir ao tema central eu vou me apresentar melhor. Sou filho de batalhadores, por definição mercadores, mas para minha sorte, sonhadores. Nesse contexto não passei fome, não estive na miséria, não tive que trabalhar desde os 8 anos e nem sofri altos danos psicológicos impostos por uma cultura machista e opressora, esse foi o maior presente que recebi dos meus pais, não passar pelos mesmos obstáculos que eles. Todo esse carinho e cuidado, me colocou em uma posição onde eu não sabia qual era minha identidade. Os sacrifícios que meus pais passaram me distanciaram das minhas raízes e eu cresci sem identidade social ou pessoal, acabei crescendo distante do restante de minha família. Quando passei a trilhar meus próprios caminhos, decidi buscar essa identidade nos mais diversos lugares. Aos 17 anos vim para Bauru cursar a faculdade de Ciência da Computação e buscar realizar um sonho que alimento desde pequeno. O de trabalhar com robótica. Graças ao Maurício de Souza, eu adorava as histórias do Franjinha e a ideia de fazer ciência. Foi com essa motivação que fiz parte do time de futebol de robôs da Unesp de Bauru e em 2013 disputei meu primeiro campeonato, sendo vice-campeão brasileiro de uma das categorias. Foi um período muito bom onde me envolvi bastante com a robótica inclusive com a educação e fiz parte de projetos interessantes envolvendo Robótica educacional. E essa vontade de trabalhar com educação também faz parte da minha busca por identidade, no caso, identidade social. Fui voluntário em diversos lugares e em um deles, o Projeto Formiguinha, fui educador social por 5 anos e em 2 deles Presidente da instituição. Foi um período muito complicado, onde tive que conciliar faculdade e trabalho voluntário, aprendi que transformar a realidade do próximo não é uma tarefa fácil. Porém, muito gratificante. Não precisa de muito para impactar pessoas em seu dia-a-dia e carregar consigo boas lembranças. E é por isso que eu me sinto tão motivado a ponto de vir falar para vocês sobre um assunto que se tornou meu ideal profissional: tecnologia avançada e seu papel social, vamos focar em inteligência artificial, robótica e seu impacto social. Como a sociedade enxerga essa tema e porque ele é um tema importantíssimo para ser discutido Primeiro eu quero desmistificar a ideia popular da robótica. Quando pensamos no robô a primeira imagem que vem a cabeça é de uma coisa metálica, humanóide cheia de fios e circuitos expostos que veio do futuro para te matar. Quando na verdade um robô pode ter formas muito mais simples e funções muito mais humanas, como o de ser uma ferramenta educacional. Mesmo com esses exemplos presentes em nosso dia-a-dia. Muita gente ainda acha que o fim do humanidade vai acontecer com um robô aparecendo dando cambalhotas e preparado para matar Pelo menos não ainda! Os robôs possuem diversas formas, que não são necessariamente físicas, e eles estão em nossas vidas há muito mais tempo do que imaginamos, robô é uma palavra e origem sueca que significa trabalho forçado e é um termo que ganhou força na década de 30 com o boom da ficção científica, principalmente através dos trabalhos de Izaac Asimov, mas muito antes disso já se falava dos autômatos e o ser humano sempre buscou meios de facilitar o trabalho pesado e monótono (e quase sempre bélico). Nas minhas experiências com a área de educação, quando trazíamos à tona a discussão do uso da tecnologia a favor da educação, o maior temor dos educadores é o da tecnologia substituí-lo. E esse é um pensamento comum, hoje esse medo acontece com diversas áreas e cada vez mais as pessoas estão com medo de serem substituídas por máquinas. De acordo com estudo da consultoria McKinsey, 50% dos empregos do Brasil poderiam ser automatizados e até 2030 estima-se que aproximadamente 18 milhões de pessoas serão impactadas pela automação. É um número assustador. E para entender isso podemos usar a ferramenta mais importante da sociedade: a história. A maioria das funções sociais e empregos existentes hoje são reflexos da primeira revolução industrial, ainda que a principal particularidade dessa revolução foi a substituição do trabalho artesanal pelo assalariado e com o uso das máquinas. Até mesmo a educação tem histórico de criar repetidores de instruções, preparar para o chão de fábrica. A consequência disso é que hoje vivemos uma realidade onde tudo o que consumimos pode ser produzido por robôs. Mesmo com a revolução industrial, ainda existem trabalhos repetitivos, difíceis e insalubres. Atualmente estamos vivenciando a substituição desses trabalhos e isso está acontecendo por meio dos avanços da Inteligência Artificial. O mundo todo está passando por uma revolução nas mãos da Inteligência artificial que foi recentemente apelidada de “possível ditador imortal” e que tem ganhado espaço em obras hollywoodanas e na ficção de um modo geral (black mirror principalmente). Porém, antes de mais nada a IA é um campo de estudo das ciências cognitivas que consiste em tentar reproduzir a inteligência humana através do software. Hoje existem aplicações extremamente inteligentes e o mercado está se moldando em torno da IA Porém estamos longe de reproduzir a inteligência humana. Conquistamos muitos espaços desvendado modelos matemáticos bem definidos da nossa sociedade, mas nós ainda temos muito o que caminhar. Mesmo assim as taxas de sucesso são muito boas. O suficiente para que empresas desse segmento sejam líderes de mercado e obtenham crescimentos a cada ano que passa, pois, a IA trabalha com informação e hoje informação é o maior ativo de mercado. Vivemos uma era de informação e o volume de dados tem sido cada vez maior Chegamos a um ponto em que se tornou inumano compreender e manipular esses dados e não conseguimos sugerir soluções para problemas complexos, como crises financeiras, distribuição de riqueza e conflitos étnicos. E é nesse contexto onde a IA e a robótica devem ser empregadas, e por isso elas devem ser debatidas abertamente como políticas pública, não apenas desenvolvida em torno de otimização de lucros ou indústria bélica (principalmente essa), que vem sendo debatida por grandes mentes, como Elon Musk, Stephen Hawkins e Bill Gates e temida pelo público em geral. A principal preocupação em relação a esse tema é que uma máquina serve ao propósito pelo qual foi enviada e pela qual foi criada. E nesse sentido 2 coisas crescem em paralelo: a tecnologia e a sabedoria em usá-la. E avançamos tanto na tecnologia que hoje a IA e a robótica são como armas na mão de um bebê. Como podemos agir para evitar essa situação? Izaac Asimov criou em seu livro “Eu, robô” três leis fundamentais para que o uso da robótica seja em prol da humanidade 1ª Lei: Um robô não pode ferir um ser humano ou, por inação, permitir que um ser humano sofra algum mal. 2ª Lei: Um robô deve obedecer as ordens que lhe sejam dadas por seres humanos exceto nos casos em que tais ordens entrem em conflito com a Primeira Lei. 3ª Lei: Um robô deve proteger sua própria existência desde que tal proteção não entre em conflito com a Primeira ou Segunda Leis. Porque, se parar para pensar sobre isso, as três Regras da Robótica são os princípios essenciais que orientam muitos dos sistemas éticos do mundo. Com certeza, todo ser humano deve ter o instinto de autopreservação. Essa é a Regra Três para um robô. Todo “bom” ser humano com uma consciência social e um senso de responsabilidade também deve submeter-se a uma autoridade apropriada; dar ouvidos ao seu médico, ao seu chefe, ao governo, ao seu psquiatra, ao seu semelhante; obedecer às leis, seguir regras, adequar-se aos costumes…. mesmo quando isso interfere em seu conforto ou segurança. Essa é a Regra Dois para um robô. Todo “bom” ser humano também deve amar ao próximo como a si mesmo, proteger seu semelhante, arriscar sua vida para salvar a de outro. Essa é a Regra Um para um robô. Para explicar de forma simples: se um robô seguir todas as Regras da Robótica propostas por Isaac Asimov, pode ser que ele seja um robô, e pode ser que seja apenas um homem muito bondoso. A forma como vamos utilizar a robótica e a inteligência artificial vai depender de você. Ainda existem muitas perguntas sem resposta e muito caminho a ser percorrido, mesmo assim, a robótica e a tecnologia tem se mostrado a ferramenta mais factível para se alcançar igualdade social sem perdermos funções essenciais para o funcionamento da civilização moderna. Uma discussão muito polêmica e muito comum que eu ouço de vez em quando sobre igualdade social é a questão do “mérito” das profissões e porque uma é melhor que a outra, o exemplo geralmente é dado na comparação entre o lixeiro e o médico. Pois se houvesse igualdade social, e nesse caso econômica, ninguém iria querer passar por todo o esforço do estudante de medicina para se tornar médico, todo mundo iria querer ser lixeiro. Esse é o argumento mais brando que eu ouço. E é através dele que eu gosto de demonstrar o tamanho do impacto social que a tecnologia vai ter em nossas vidas, pois, no futuro o seu lixeiro e o seu médico vão ser robôs. Afinal, o trabalho de coleta de lixo é desumano, mesmo assim, essencial para a sociedade, e a medicina hoje já é um modelo matemático conhecido, as pessoas vão para a mesa de cirurgia sabendo da taxa de sucesso. E o robô vai ter uma taxa de acerto maior que o ser humano breve. O projeto IBM Watson está com foco em uso da IA para medicina e aposta em um aumento na precisão de diagnósticos e tratamentos em um futuro próximo. E não estamos muito longe dessa realidade, nesse ano de 2018 a FDA aprovou o uso de Inteligência Artificial para diagnóstico sem a necessidade de um médico. O software é projetado para detectar mais do que um nível leve de retinopatia diabética, que causa perda de visão e afeta 30 milhões de pessoas nos EUA. Ocorre quando o açúcar elevado no sangue danifica os vasos sanguíneos da retina. E por conta disso temos que trazer diversos outros temas para evoluírem junto com a robótica, pois hoje ela é, assim como toda tecnologia avançada que foi criada pela humanidade, realidade e propriedade apenas de um grupo pequenos de pessoas, apenas aqueles que podem financiar. Por isso, ao falar de robótica, também temos que falar de direitos humanos, temos que falar de luta de classes, temos que falar de igualdade de gênero, temos que falar de igualdade de oportunidades, etc. Porque a tecnologia está vindo para ocupar tudo aquilo que o ser humano faz que uma máquina deveria estar fazendo, precisamos garantir que ela faça isso de forma justa. Dessa forma espero que para nós, após termos todas essas tarefas tediosas realizadas de forma automática, sobre tempo para arte, música, relações humanas, enfim, sobre tempo para sermos mais humanos.

Porquê Robótica e Inteligência Artificial devem ser discutida como políticas públicas?

47

porquê-robótica-e-inteligência-artificial-devem-ser-discutida-como-políticas-públicas-16bfa4038ecc

2018-06-07

2018-06-07 14:25:54

https://medium.com/s/story/porquê-robótica-e-inteligência-artificial-devem-ser-discutida-como-políticas-públicas-16bfa4038ecc

false

1,988

Programação, design, conhecimento aberto, responsabilidade social e tudo mais o que vale ser discutido.

null

grupotesseract

null

Grupo Tesseract

contato@grupotesseract.com.br

grupo-tesseract

PROGRAMAÇÃO,DESIGN,OPEN SOURCE

null

Robotics

robotics

Robotics

9,103

Mateus Batista Santos

Conhecimento é uma ferramenta importante de transformação social. Com tecnologia e um bom plano político-pedagógico é possível transformar o país.

19f4c05693cf

mateusbatistasantos

12

2

20,181,104

null

0

null

0

null

2018-06-08

2018-06-08 15:52:28

2018-06-07

2018-06-07 17:42:16

3

false

en

2018-09-26

2018-09-26 16:12:41

2

16bfd3a29344

2.448113

1

0

As the world´s first company to build a trusted Global Marketplace-Economy™, EZeeBUY™ has the expertise to develop the technology, an…

5

The EZeeBUY™ Mission & Strategy As the world´s first company to build a trusted Global Marketplace-Economy™, EZeeBUY™ has the expertise to develop the technology, an experienced management team to build the businesses and the vision to revolutionize the shopping & buying experience. Our long-term objective is to drive global mass consumer adoption by enabling 2.5 billion smartphone cameras with the capability to buy any product from the world’s first Global Marketplace-Economy™ and pay for those purchases with EZ8 tokens. With more than 1 trillion pictures taken every year, EZeePHOTO™ and EZeeDiNA™ services will learn to understand a consumer’s personality and enable any brand or retailer to offer bespoke offers with precision — while paying for that privilege through product discounts, exclusive offers, and access to limited edition products. We plan to achieve this mission by executing on 3 key strategic initiatives: 1. EZ8 Token ICO EZeeBUY™ will launch an Initial Coin Offering in the Summer of 2018, issuing virtual currency utility tokens called “EZ8 Token” (or simply, “EZ8”). Proceeds from the ICO will be used to fund further development of the EZeeBUY™ smartphone application, along with all of the associated products and services in the platform. The EZ8 utility token will be used within the EZeeBUY™ wallet and Global Marketplace-Economy™ for immediate product purchases on the EZeeBUY™ smartphone application. 2. Global Expansion Plan The initial launch of EZeeBUY™ will be in Japan, considered mobile-centric with one of the most mature mobile commerce markets in Asia. We have already signed a contract with our first customer — a major Japanese retailer with over 160 stores nationwide — who will be the first to implement our technology in the Fall of 2018. Other retailers in Japan will quickly follow. After a successful launch in Japan, EZeeBUY™ will expand internationally. The next country after Japan will be the USA — the 2nd largest mobile commerce market globally. Following the USA, we plan to further expand to the UK followed by other Asian and European countries. In order to accelerate global expansion, we plan to replicate our Japanese retailer partnerships with other strategic retailers internationally. We will help retailers transform the in-store experience — improving how shoppers interact with brick-and-mortar stores. New in-store improvements will include experiential marketing (via augmented reality), instant payment, exclusive express check-out for loyalty members, and an opportunity to earn EZ8 tokens via the EZeeEARN™ rewards points program. We will be focused on retailers with the following Optimal Retailer Partner Profile: Searching for enhanced customer experience Retail store transformation Large customer loyalty program Extensive product categories and inventory 3. EZeePRODUCT™ Global Product Catalog Once we have developed a global product catalog of the entire e-Merchant world, we can launch worldwide and offer any consumer the ability to shop & buy from any e-Merchant globally. For more information about EZeeBUY™ and the EZeeBUY™ ICO, visit: www.ezeebuy.ai “Buying Made Ezee — just take a picture!” Originally published at medium.com on June 7, 2018.

The EZeeBUY™ Mission & Strategy

16

the-ezeebuy-mission-strategy-16bfd3a29344

2018-09-26

2018-09-26 16:12:41

https://medium.com/s/story/the-ezeebuy-mission-strategy-16bfd3a29344

false

503

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

David Pipe

Co-Founder & Chief Marketing Officer at EZeeBUY™

23cd61a8cdf7

davidpipe_71703

11

4

20,181,104

null

0

null

0

null

2017-11-01

2017-11-01 22:12:41

2018-07-25

2018-07-25 02:02:49

3

false

en

2018-07-25

2018-07-25 02:04:10

8

16c1c437217f

2.104717

0

Do you spend your days crafting “neuro marketing” strategies that utilise behavioural conditioning and addiction mechanisms? Are you…

5

Do you have better ethics than an arms dealer? Do you spend your days crafting “neuro marketing” strategies that utilise behavioural conditioning and addiction mechanisms? Are you working on those strategies in order to generate more revenue from user attention? https://twitter.com/kumailn/status/925832999039410176 The big three came under scrutiny on Capitol Hill over how they profited from Russian influencer marketing last year. The equivocal, unsatisfying responses offered by company representatives stank up the place. “We’re just a platform” is utter bullshit. That we keep hearing this from really smart people running the most influential MEDIA companies in the world is fucking infuriating. They don’t believe what they are saying and neither do I. Advertising funded media is not new, it’s as old as newspapers, radio and TV. Customer attention = advertising dollars. As the race to become the first trillion dollar company intensifies, the arms race for attention has created an environment devoid of ethics. An environment where the smartest (majority white, male, 25 to 35 year old) minds are hell bent on keeping you scrolling through a never ending feed of “things you might have missed” and adverts. Yeah, those guys, the ones who like to write memos about “Idealogical echo chambers”. Great job everyone. “Investing in better security and scrutiny is going to negatively affect profits next quarter…” If your job requires you to exploit human addiction patterns and visceral emotion to “drive increased engagement” then you are probably fine throwing in with members of congress and the senate who rate as the least ethical in this Gallup poll. Instagram, twitter, facebook, all use machine learning to “curate” our social media feeds. This sounds so very artisanal and friendly but really boils down to weaponisation of our human vulnerabilities for money. It’s morally bankrupt and richly rewarded. Tristan Harris, a leading voice in the push for greater ethics in modern technology got me seriously thinking about the topic. I strongly suggest you listen to his podcast appearances and interviews, he has unique insight and is an excellent communicator. The reading/listening list continues; Mike Monteiro “A Designers Code of Ethics” Nick Bilton & Scott Galloway “And then there were four” Podcast Doc Searls “Towards an Ethics of Influence” It’s a rabbit hole I’m glad I fell into. After investing an hour or so carrying out Harris’ tips, I’ve reclaimed 157 hours from these apps since mid April [as measured by Moments app]. Some of that time has been spent listening, reading, learning and considering the issue #timewellspent. I definitely did not use any of the time watching GLOW…newp

Do you have better ethics than an arms dealer?

0

outragedo-you-have-better-ethics-than-an-arms-dealer-16c1c437217f

2018-07-25

2018-07-25 02:04:10

https://medium.com/s/story/outragedo-you-have-better-ethics-than-an-arms-dealer-16c1c437217f

false

412

null

Ethics

ethics

Ethics

7,787

rollinson

Event Technology geek. Backstage catering critic. Collector of Vinyl records

849ba0bd80a0

rollinson

172

271

20,181,104

null

0

null

0

7f92bce296f6

2018-01-11

2018-01-11 17:33:22

2018-01-15

2018-01-15 10:23:56

1

false

fr

2018-01-15

2018-01-15 10:23:56

1

16c237a92a41

1.528302

2

0

Bio : Professeure en Intelligence Artificielle à Paris-Sorbonne , dirige une équipe de recherche au LIMSI-CNRS sur « dimensions fectives et…

4

“Les machines répondent sans comprendre !” — Laurence Devillers Bio : Professeure en Intelligence Artificielle à Paris-Sorbonne , dirige une équipe de recherche au LIMSI-CNRS sur « dimensions fectives et sociales dans les interactions parlées » notamment avec des robots et membre de la CERNA (commission sur l’éthique de l’alliance ALLISTENE). Elle est auteur du livre « Des robots et des hommes : mythes, fantasmes et réalité », plon 2017. La reconnaissance de la parole comme la traduction de la parole ont fait d’énormes progrès grâce au deep learning et aux immenses bases de données disponibles mais en sémantique nous sommes encore très loin d’avoir des machines performantes. Parler avec les interfaces nous permet de rentrer en contact avec les machines de façon très intuitive. Il y a là cependant une énorme tromperie car la machine même si elle sait nous répondre ne comprend pas grand chose à ce que l’on dit. John Searle se demandait si un programme informatique, si complexe soit-il, serait suffisant pour donner un esprit à un système. Il montre en 1980 à travers la chambre chinoise qu’il ne suffit pas d’être capable de reproduire exactement les comportements linguistiques d’un locuteur chinois pour parler chinois, car parler le chinois, ou n’importe quelle autre langue, ce n’est pas juste dire les bonnes choses au bon moment, c’est aussi signifier ou vouloir dire ce qu’on dit : un usage maîtrisé du langage se double ainsi d’une compréhensions du sens de ce qu’on dit et la reproduction artificielle, même parfaite, d’un comportement linguistique ne suffit pas à produire une telle compréhension. Alors oui La machine et l’humain tendent à se fondre et confondre mais seulement en surface car la machine ne fait que simuler la compréhension sans rien comprendre. Le graal pour les chercheurs actuellement est d’arriver à modéliser la partie sémantique et le sens commun qui manquent terriblement aux machines. En effet par exemple les machines décodent le langage pour la reconnaissance de la parole en utilisant des connaissances probabilistes de successions de mots. >>> Retrouvez le tribune complète par Laurence Devillers sur le blog FUTUR.E.S !

“Les machines répondent sans comprendre !” — Laurence Devillers

11

les-machines-répondent-sans-comprendre-laurence-devillers-16c237a92a41

2018-02-22

2018-02-22 10:38:40

https://medium.com/s/story/les-machines-répondent-sans-comprendre-laurence-devillers-16c237a92a41

false

352

Tribunes ✍️ veille 👁️ événements 🤝 et décryptages 👓 du plus grand collectif d'innovateurs du numérique d'Europe 💡

null

capdigitalparisregion

null

Cap Digital

mathilde.neu@capdigital.com

cap-digital

INNOVATION,NUMERIQUE,TRANSFONUM,EXPERT,FUTUR

Cap_Digital

Intelligent Machines

intelligent-machines

Intelligent Machines

55

Cap Digital

Collectif d’innovateurs du numérique 💡🚀⚡️ Organisateur du festival Futur.e.s 🤖 & créateur de EdFab, fabrique des nouvelles formations numériques 🎓

13c3551581ce

Cap_Digital

4,240

381

20,181,104

null

0

null

0

null

2017-12-22

2017-12-22 17:11:49

2017-12-30

2017-12-30 17:12:07

1

false

en

2018-08-08

2018-08-08 08:44:24

3

16c29614bd0a

9.554717

0

Why should bots make decisions for humans?

4

When bots become kings Death of Socrates. (Source) Why should bots make decisions for humans? In 380 BC, Plato details the conversation Socrates had with his interlocutors, in which Socrates had so famously declared that Until philosophers are kings, or the kings and princes of this world have the spirit and power of philosophy, and political greatness and wisdom meet in one, and those commoner natures who pursue either to the exclusion of the other are compelled to stand aside, cities will never have rest from their evils, — nor the human race, as I believe, — and then only will this our State have a possibility of life and behold the light of day. Such was the thought, my dear Glaucon, which I would fain have uttered if it had not seemed too extravagant; for to be convinced that in no other State can there be happiness private or public is indeed a hard thing. For Socrates, the best of kings are different from the best of philosophers. While a Socratic king seeks power, fame and fortune, the Socratic philosopher chooses to set aside the very same avarice wielded by the aforementioned kings. The king acts on his will for political power, while the philosopher rids that lust. This means the ultimate aim of the king is to consolidate his power while the aim of the philosopher is to devise the best way of living for his citizens, thus bestowing the philosopher with the power and political wit of the king ensures the power is used efficiently, and the king the wisdom and modesty of the philosopher so that he rules justly. There comes a time where the philosophies of the East and West embrangles, and such is when Confucian principles so happen to fit perfectly alongside Socrates’s ideas. The abovementioned idea of the philosopher king marries into the Confucian wisdom that: If your desire is for good, the people will be good. Desire. Desire is brought about by a lacking of sorts, either to the individual or the community in which the individual belongs. The common philosopher king’s desire is, presumably, to ensure his citizens live the most ideal of environments, in the most intellectual of a society. The common corrupt official’s desire is to rake in his fortunes at the expense of the citizenry he is answerable to. This then begs the question. What is the desire of a bot? Rather, the definition of desire proffered in the earlier paragraph does not seem to fit within what bots would ascertain as an individual. In this case, since the individual is defined by the sense of consciousness, the sense of being, and the knowledge of being absolutely self-aware, are bots predominantly conscious? Most people have heard of Artificial Intelligence, with Intelligence being the word they associate with the usage of environmental data to make decisions to successfully arrive at any given goal. However, this intelligence is almost always conflated with the idea of consciousness. In his TED presentation in 2014, Australian philosopher David Chalmers opened his speech by offering one of the most defining features of consciousness. He said: Right now you have a movie playing inside your head. It’s an amazing multi-track movie. It has 3D vision and surround sound for what you’re seeing and hearing right now, but that’s just the start of it. Your movie has smell and taste and touch. It has a sense of your body, pain, hunger, orgasms. It has emotions, anger and happiness. It has memories, like scenes from your childhood playing before you. And it has this constant voiceover narrative in your stream of conscious thinking. At the heart of this movie is you experiencing all this directly.This movie is your stream of consciousness, the subject of experience of the mind and the world. Before anyone tries to refute that claim, let me speak my piece. Sure, the idea of consciousness being a movie is almost too disparaging to the entire study of consciousness, but refuting it completely is outright pretentious. That is simply because we do not understand enough about consciousness. Because it is so esoteric, Chalmers went the extra mile in making a radical postulation that: The first crazy idea is that consciousness is fundamental. Physicists sometimes take some aspects of the universe as fundamental building blocks: space and time and mass… These fundamental properties and laws aren’t explained in terms of anything more basic.Rather, they’re taken as primitive, and you build up the world from there… as a matter of logic, you need to expand the list. The natural thing to do is to postulate consciousness itself as something fundamental, a fundamental building block of nature. Since consciousness and intelligence are similar in the fact that beings such as orcas and humans are able to absorb and process varying amounts of information to perform, what is the difference? What intelligence invests in is the ability to process any amount of information, from a simple single sentence to a complex billion-charactered script, or an infinitely large amount of universal information to perform single, specialised tasks. What it does not include is the ability to use these large amounts of information to apply it to different, sparser tasks that determines what being self aware is. Prof. Chalmers also promoted the view of panpsychism, which asserts that each and every thing has at least a rudimentary form of consciousness. He notes: This view is sometimes called panpsychism… every system is conscious, not just humans, dogs, mice, flies, but even Rob Knight’s microbes, elementary particles. Even a photon has some degree of consciousness.The idea is not that photons are intelligent or thinking. It’s not that a photon is wracked with angst because it’s thinking, “Aww, I’m always buzzing around near the speed of light. I never get to slow down and smell the roses.” No, not like that. But the thought is maybe photons might have some element of raw, subjective feeling, some primitive precursor to consciousness. This quote of his has to be broken down. By far and large, the entire essay has spoken about making 2 terms, namely “consciousness” and “intelligent” mutually exclusive. However, with the input of Prof. Chalmers, these 2 terms may be more similar than ever after all. First, the idea of panpsychism is not entirely wrong, for every system, or every particle that goes through a system, has to have experience on itself. Yes, that means they have to provide a response. Based on Newton’s third law, there is an equal, opposing reaction to every action. This leads to the second point that every particle or system must have a process in which they process an input to provide an output. I propose calling that process “intelligence”, with the definition that it is an intrinsic property, so closely tied to its architecture, that allows it to provide a response to appropriately suit the input provided. This leads me to my third point. I propose, to Prof. Chalmers, that intelligence is a rudimentary or his idea of being a precursor to consciousness. Suppose this case, that a volleyball player hits a ball with all her might to have it travel across and over a net. I would suppose, in this case, that the volleyball has some form of intelligence because it follows its design, to be pumped so full of air, the volleyball player could hit it over the net, and to be able to provide an opposing force exerted on the player’s arm so that she can put her arm back down to prepare her next strike. Is this intelligence? Yes, it certainly is, because it adheres to its design and follows through with a specific purpose. Is it so conscious it chides you about how much pain you’ve caused it when hitting it? I don’t think so. Since the question about consciousness is so open-ended, I would like to spark a discussion about what consciousness really is amongst not only scholars but also amongst us, the people, because it matters that our society will see an assimilation of bots into our political offices, and one major problem is that we don’t quite yet know what consciousness is. My conjecture, based on what little information I’ve gathered, and what else I’ve read, is that consciousness is derived from an intelligence that is not only able to act on what it deems is a suitable outcome based on its physical, chemical or biological design, but is also able to understand and learn any emotional, moral or political implications their actions would have once it has performed its intended task, and hence is able to determine any further courses of action to best suit their purposes. This is my take. Humans are conscious because they are able to feel beyond just nervous control systems that tell us whether we are hungry or not — our nervous connections have since allowed us to adopt emotional connections with other people, and hone our senses to appreciate the world. Unlike our multifaceted abilities to take our intrinsically essential chemistry of survival through a channel into experiences and memories, what a machine does is to store memories and process them into doing a single specialised task. Want a certain car part made? A machine does that and nothing else, not make a cup of coffee for you and ask you about how your day went(though, sure, it could). CloudWalk from China is intelligent. It helps the Chinese police in detecting and predicting crimes through data fed to it. Is it conscious? It is hard to say. Further putting in question the very fabric of our society from the earlier paragraph is the emotional (and hence thereof, moral) capabilities of our machines. Should they one day take over the offices of our government as philosopher bots, to serve alongside their politician colleagues and to cast alongside their votes as possible senators and members of parliament, should we stay afraid? The short answer, optimistically, is no. The long answer, I don’t know. Surely, the conscious mind asks questions, and derive answers from their own subjective analysis and an initial intake of information. If so, does a politician robot ask whether or not its citizens are living in a just society and what it can do for us. On a more fundamental level when it comes to stabilising their political seat, the robot certainly does need a level of consciousness to be able to use appropriate political diction so that it can achieve a desired outcome. However, most people remain doubtful about the idea of artificial intelligence. A friend of mine, a computer scientist, proposed an idea. What if he creates an AI that has the sole purpose of calculating the value of pi, but has determined that by causing worldly destruction and sacrificing human lives, it could gain another digit to its collection of digits efficiently? Surely, the AI is intelligent enough to calculate the value of pi, maybe even surpassing our current intelligence in that, but with matters of consciousness, it should be unable to calculate that the sacrificing of human lives just to calculate the value of pi indefinitely, without any real purpose, is morally wrong. Hence, the whole idea of consciousness in machines is improbable. The only probable definition of the ability to philosophise, emotionally connect with humans and understanding morality is artificial consciousness. Just this year during the Web Summit Technology Conference in Lisbon, Portugal, Prof. Stephen Hawking, a British theoretical physicist, asserted the dangers of AI, warning that, as CNBC understands, The emergence of artificial intelligence (AI) could be the “worst event in the history of our civilization” unless society finds a way to control its development. Prof. Hawking further notes that Success in creating effective AI could be the biggest event in the history of our civilization. Or the worst. We just don’t know. So we cannot know if we will be infinitely helped by AI, or ignored by it and side-lined, or conceivably destroyed by it… Unless we learn how to prepare for, and avoid, the potential risks, AI could be the worst event in the history of our civilization. It brings dangers, like powerful autonomous weapons, or new ways for the few to oppress the many. It could bring great disruption to our economy. Prof. Hawking is not alone in this school. Tesla CEO Elon Musk has also so famously proclaimed AIs being at the forefront of WWIII, as I have mentioned in my last essay. If anything, this has signalled to us that the creators of AIs have to learn and know and understand how to best minimise the ramifications of AIs in our society as they start to displace more humans. They will have to learn about the possible roles AIs can take when it comes to their militarisation, and how to best curb it. They should also have to learn to enunciate and emphasise morality and judgement in the codes that make up these AIs so that AIs themselves too will learn about the value of their existence. Prof. Hawking was so right. We need to control the development of AIs. Maybe that’s where my computer scientist friend can then think of ways to input codes and functions, or maybe be inspired to wire a machine with an artificial neural network approach, this machine that calculates pi will one day think more human, become more human and, without forgetting the initial purpose of calculating the next digit of pi, learn to love humans. This is where we draw the final stroke to the painting. As AIs progress, we need to start establishing the Great Divide between intelligence and consciousness, and heed the warnings of great men, many of whom have taught us to understand our lives and morality well. More importantly, we need to learn that artificial consciousness is plausible so that we are better able to regulate the progress of AIs. My computer scientist friend, in our WhatsApp conversation, has also stamped his point poignantly. “ASIs may simply kill off humans by accident to achieve their goals. We need to think about the goals we set for AIs.” Just as the wise Socrates taught over 2000 years ago, maybe we should start teaching our creations 2000 years from now. I am no programmer myself. I am no computer scientist. I am just a mere soldier who has just graduated high school. I am serving my mandatory military service while waiting to enter my university when 2019 comes. I do not understand a single line of code nor an entire script of computer language. Heck, I don’t even know how Microsoft Excel works without consulting my mother (she’s an accountant, she knows what she’s doing). However, because the presence of AIs can directly influence our lives and the generations below, I am left with the option of learning about it.

When bots become kings

0

when-bots-become-kings-16c29614bd0a

2018-08-08

2018-08-08 08:44:24

https://medium.com/s/story/when-bots-become-kings-16c29614bd0a

false

2,479

null

Philosophy

philosophy

Philosophy

39,496

Hwang Chen Hsun

Writer at Becoming Human. Currently serving the nation.

acd63042505e

hwangCH

4

14

20,181,104

null

0

null

0

32881626c9c9

2018-04-14

2018-04-14 02:23:19

2018-04-14

2018-04-14 02:24:39

1

false

en

2018-05-27

2018-05-27 00:28:00

0

16c3aa684951

4.335849

3

1

0

Modernity has indeed — for some or many — been an enjoyable journey, and it has borne multiple and great fruits in all walks of life. It…

5

Being Human in a Hyperconnected Era Modernity has indeed — for some or many — been an enjoyable journey, and it has borne multiple and great fruits in all walks of life. It has also had its downsides. The constraints and affordances of the computational era profoundly challenge some of modernity’s assumptions. Developments in scientific knowledge brought about an endless list of new artefacts in all sectors of life. Despite the deep connection between artefacts and nature, an alleged divide between technological artefacts and nature continues to be assumed. The development and deployment of ICTs have contributed enormously to blurring this distinction, to the extent that continuing to use it as if it were still operational is illusory and becomes counterproductive. Data are recorded, stored, computed and fed back in all forms of machines, applications, and devices in novel ways, creating endless opportunities for adaptive and personalised environments. Filters of many kinds continue to erode the illusion of an objective, unbiased perception of reality, while at the same time they open new spaces for human interactions and new knowledge practices. The fact that the environment is pervaded by information flows and processes calls for new forms of thinking and doing at multiple levels. The very complexity and entanglement of artefacts and humans invite us to rethink the notion of responsibility in such distributed socio-technical systems. Throughout our collective endeavour, a question kept coming back to the front stage: “what does it mean to be human in a hyperconnected era?” This foundational question cannot receive a single definitive answer, but addressing it has proven useful for approaching the challenges of our times. Handling these challenges can best be done by privileging dual pairs over oppositional dichotomies. In a remarkable article “The computer for the 21st century”, published in the Scientific American in September 1991, Mark Weiser suggested that, after the mainframe and the personal desktop computer, the next step will be ubiquitous computing, i.e. a technology that has become so pervasive that it is invisible to us and totally embedded in our lives. In fact, together with the current burgeoning of devices, sensors, robots, and applications, and these emerging technologies, we have entered a new phase of the information age, a phase where the hybridisation between bits and other forms of reality is so deep that it radically changes the human condition in profound ways. The ubiquitous computing vision is a reasonable asymptotic view, which can be taken as the current background against which society is striving to actualise its norms, values and codes of behaviour. This digital transition shakes established mental frameworks in, at least, three ways: · by blurring the distinction between reality and virtuality; · by blurring the distinctions between human, machine and nature; · by reversing from scarcity to abundance, when it comes to information; · by shifting from the primacy of entities over interactions to the primacy of interactions over entities. As a society, we are confronted with a learning challenge of how to actively shape our lives in this technologically-mediated world. Let us consider these issues in turn. The Blurring of the Distinction Between Reality and Virtuality Plato’s allegory of the cave, the distinction between body and mind, or that between internal fantasies and actual behaviours are fundamental and ancestral dichotomies through which we think and act. Philosophers have argued that these dichotomies are fragile and more illusory than one may think. By making virtuality more real than ever before, the digital transition undermines the real/virtual divide, and thereby all dualist forms of thinking. In concrete terms, exploring these issues will shed light, for example, on the level of continuity in behavioural and moral terms that should be expected in the virtual and the physical public spaces. For example, anthropologists tell us that it is common practice for people to lie about themselves on the internet, not necessarily for bad reasons, but rather as a social practice: minors and dating adults lie about their age, appearance, interests, and so forth. Is this really affecting trust or, on the contrary, is it part of the acculturation of ICT tools by society, producing the shadow areas that any individual needs to live as a human? The Blurring of the Distinctions Between People, Nature and Artefacts Once upon a time, it was easy to distinguish people from artefacts and nature. Since the industrial era, artefacts and nature have become intrinsically connected, through the metabolism of the industrial development, which is drawing on natural resources. The digital transition acts as a huge accelerator of the blurring of these once effective distinctions. The multiplication of sensors and prostheses, the progress of cognitive sciences and biological engineering blur the distinction between humans and artefacts. This means that our conceptual toolbox, still reliant on these once effective distinctions between humans, nature and artefacts, needs to adapt to this new reality, where these distinctions no longer exist. The Reversal from Scarcity to Abundance, when it Comes to Information Information is akin to natural resources of a third kind, besides the non-renewable and the renewable, we have the exponential. We need to learn to make sense of and value the abundance of information through datamining and other filtering activities. History has lasted 6,000 years, since it began with the invention of writing in the fourth millennium BC. During this relatively short time, ICTs have provided the recording and transmitting infrastructure that made the escalation of other technologies possible, with the direct consequence of furthering our dependence on more and more layers of technologies. ICTs became mature in the few centuries between Guttenberg and Turing. Today, we are experiencing a radical transformation in our ICTs that could prove equally significant. At the same time, there is an atmosphere of confused expectancy, of exciting, sometimes naïve, bottom-up changes in our views about (i) the world, (ii) about ourselves, (iii) about our interactions with the world and (iv) among ourselves. Much more realistically and powerfully, but also more confusedly and tentatively, Living in the 21st century we should be able to make the changes in our Weltanschauung both intellectually and behaviourally, to a reality that is fluidly changing in front of our eyes and under our feet, exponentially and relentlessly. We may be confused, yet we should act much more conscientiously. After all, that is at the core of what it means to be human in a hyperconnected era. “And they ask you about the spirit. Say: The spirit is from the command of my Lord, and you are not given ought of knowledge but a little.” (Qur’an: 17:85)

Being Human in a Hyperconnected Era

47

being-human-in-a-hyperconnected-era-16c3aa684951

2018-05-27

2018-05-27 00:28:01

https://medium.com/s/story/being-human-in-a-hyperconnected-era-16c3aa684951

false

1,096

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

Technology

technology

Technology

166,125

Daily Wisdom

null

ddd120ae7c2

dailywisdom

60

0

20,181,104

null

0

null

0

bf2113ea315

2017-10-03

2017-10-03 12:33:31

2017-10-03

2017-10-03 12:35:41

1

false

en

2017-10-03

2017-10-03 12:36:49

6

16c3fa377f54

1.362264

1

0

FlytBase will be releasing its AI platform for drone applications at the Drone World Expo, San Jose, on 3rd October 2017. FlytBase has…

4

FlytBase to Unveil its AI Platform at Drone World Expo, San Jose FlytBase will be releasing its AI platform for drone applications at the Drone World Expo, San Jose, on 3rd October 2017. FlytBase has built the world’s first IoT platform for commercial drones, the “Internet of Drones” (IoD) platform. Continuing on its mission to bring intelligence and connectivity to commercial drones, FlytBase is now extending its cloud and edge compute platforms to incorporate AI and machine learning. Drones generate vast amounts of data, which is usually in the form of images or video streams. Identification of objects of interest, counting them, or detecting change over time, are some of the tasks that are monotonous and labor intensive. FlytBase AI platform offers a complete solution to automate such tasks. It has been designed and optimised specifically for drone applications. The cloud-based training system leverages the scalability of the cloud to accelerate the training of models, to suit various customer requirements. Based on the use-case, the trained model can be deployed in the cloud (for post-processing of data) or on the edge (for real-time analysis). FlytBase AI platform is optimised for interpretation of drone data, and it seamlessly integrates with the rest of FlytBase platform to offer connectivity with your business applications. Be the first one to know more about FlytBase AI Platform, signup to stay tuned. Visit: https://flytbase.com/ai Source: blogs.flytbase.com Join Nitin Gupta in the panel discussion with other leaders from the Industry on The Role of IoT, Software & Platforms in the Commercial Drone Ecosystem At Drone World Expo October 3, 2017 | Conference Room №1, San Jose Convention Center, San Jose, CA Schedule a private meeting with Nitin and learn more on how FlytBase can help you accelerate your drone development. Write to us letstalk@flytbase.com or schedule a Skype call with us. Source: blogs.flytbase.com

FlytBase to Unveil its AI Platform at Drone World Expo, San Jose

1

flytbase-to-unveil-its-ai-platform-at-drone-world-expo-san-jose-16c3fa377f54

2018-04-17

2018-04-17 02:53:59

https://medium.com/s/story/flytbase-to-unveil-its-ai-platform-at-drone-world-expo-san-jose-16c3fa377f54

false

308

Developer Platform for Commercial Drone Applications

null

flytbase

null

FlytBase

support@flytbase.com

flytbase

UAV,DEVELOPER TOOLS,COMMERCIAL DRONES,SOFTWARE

flytbase

Drones

drones

Drones

9,040

Utsav Chopra

Hello I ‘m Utsav Chopra. I am kinda SEO and Digital Marketing Guy, love to explore the trends from the same industry. Interested in Blogging, Drones, SEO…

21544bcd25c2

UtsavChopra

69

65

20,181,104

null

0

null

0

null

2018-08-29

2018-08-29 03:39:06

2018-08-29

2018-08-29 03:39:18

0

false

en

2018-08-29

2018-08-29 03:39:18

1

16c447fad267

2.445283

0

DOWNLOAD in <PDF> Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python EPUB By Thomas W. Miller Jr…

1

Pdf Download eBook Free Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python By Thomas W. Miller Jr. Pdf books #pdf DOWNLOAD in <PDF> Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python EPUB By Thomas W. Miller Jr. Link https://bestreadkindle.icu/?q=Marketing+Data+Science%3A+Modeling+Techniques+in+Predictive+Analytics+with+R+and+Python . . . . . . . . . . . . . . . . . . . Read Online PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download Full PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download PDF and EPUB Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read PDF ePub Mobi Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Reading PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read Book PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read online Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Thomas W. Miller Jr. pdf, Download Thomas W. Miller Jr. epub Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read pdf Thomas W. Miller Jr. Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download Thomas W. Miller Jr. ebook Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read pdf Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Online Download Best Book Online Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read Online Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Book, Read Online Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python E-Books, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Online, Read Best Book Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Online, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Books Online Download Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Full Collection, Download Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Book, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Ebook Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python PDF Read online, Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python pdf Download online, Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Read, Download Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Full PDF, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python PDF Online, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Books Online, Read Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Full Popular PDF, PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Read Book PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read online PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download Best Book Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Read PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Collection, Read PDF Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python Full Online, Read Best Book Online Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python, Download Marketing Data Science: Modeling Techniques in Predictive Analytics with R and Python PDF files #E_books #EPUB #pdffree #PPT #iBooks

Pdf Download eBook Free Marketing Data Science: Modeling Techniques in Predictive Analytics with R…

0

pdf-download-ebook-free-marketing-data-science-modeling-techniques-in-predictive-analytics-with-r-16c447fad267

2018-08-29

2018-08-29 03:39:18

https://medium.com/s/story/pdf-download-ebook-free-marketing-data-science-modeling-techniques-in-predictive-analytics-with-r-16c447fad267

false

648

null

Data Science

data-science

Data Science

33,617

Andrei Mills

null

54c1e50aa3b3

andreimills

0

1

20,181,104

null

0

null

0

null

2018-02-22

2018-02-22 00:02:56

2018-02-22

2018-02-22 01:15:41

0

false

es

2018-02-22

2018-02-22 01:15:41

1

16c48c3cee55

1.588679

2

0

Martes 20 de Febrero a las 20:14 horas llega un mail de la persona encargada de las admisiones para la incubadora de Singularity University…

5

¿Querés invertir en una empresa tecnológica de Mendoza que está a punto de despegar hacia la NASA? Martes 20 de Febrero a las 20:14 horas llega un mail de la persona encargada de las admisiones para la incubadora de Singularity University felicitándome por mi admisión. Hacía unas semanas que había postulado. Singularity University es la mejor escuela de emprendedores exponenciales del mundo bancadas por la NASA y Google. Apenas caí de lo que pasaba le llamé a un amigo y emprendedor tech (groso) y le dije: “¿Y ahora qué?”. Llevo más de un año lanzando productos al mercado y a estas alturas he invertido todo mi dinero en ello. Mi amigo me dijo: “ a buscar el dinero!” What?? Claro, me olvidaba que la incubación era presencial en el predio de la NASA y que tenía que pagarme el viaje y el programa de 10 semanas de capacitaciones a una módica suma de 18.000 Dólares! Bueno, actualmente (24 horas después) le he mandado mails y tweets hasta al Papa pidiendo capital para el viaje, han salido reuniones y calls pero no hemos concretado nada aún. El Papa todavía no me responde. Les cuento el propósito de SinguLogic: democratizar el acceso a la educación bla bla… stop! Vamos a reinventar la educación y estamos enfocados en transmitir habilidades y mentalidad exponenciales a todos los seres humanos de la tierra que tengan un smartphone. Si, eso nada más. ¿Cómo? Implementando inteligencia artificial en la educación por medio de asistentes virtuales de aprendizaje (chatbots) para hacerle preguntas, pedirle exámenes y ser guiados en el desarrollo de proyectos. Tipo JARVIS pero de la educación tech. Esto ya está implementado. Pero si hablamos de democratizar y descentralizar algo y no hablamos de Blockchain me estaría perdiendo de algo…no? Este punto lo implementaremos a mediano plazo y vamos a necesitar mucha ayuda. Ahora vamos a hablar en grande: tengo el enorme desafío de obtener una inversión de USD 18.000 en 10 días y transmitirles a las personas adecuadas de lo grandísima que es esta oportunidad de inversión. Y por supuesto ni hablar del potencial de impacto positivo en el mundo que tendrá SinguLogic (mi startup) gracias a sus aportes. SinguLogic no es un proyecto ni una idea, está funcionando ahora mismo. Crecimos mucho con poco, ahora vamos por todo! Si te interesa aportar desde USD 1.000 podés contactarme a: leandroretasabio@gmail.com / +5491133246242 Tengo los instrumentos legales listos para aplicar la inversión desde Argentina y Mendoza en particular. Mi MVP: http://singulogic.com/landing/ Agradezco difusión de la nota! Muchas gracias!

¿Querés invertir en una empresa tecnológica de Mendoza que está a punto de despegar hacia la NASA?

4

querés-invertir-en-una-empresa-tecnológica-de-mendoza-que-está-a-punto-de-despegar-hacia-la-nasa-16c48c3cee55

2018-02-25

2018-02-25 03:30:31

https://medium.com/s/story/querés-invertir-en-una-empresa-tecnológica-de-mendoza-que-está-a-punto-de-despegar-hacia-la-nasa-16c48c3cee55

false

421

null

Edtech

edtech

Edtech

8,605

Leandro Reta Sabio

Founder & CEO of SinguLogic. Ex Google. Argentine bad student but good apprentice. Alumni of Singularity University Incubator. ICO | Blockchain | Cryptocurrency

8d0904f336fb

leandroretasabio

121

162

20,181,104

null

0

null

0

d777623c68cf

2018-03-29

2018-03-29 10:53:37

2018-03-29

2018-03-29 12:32:52

3

false

en

2018-04-20

2018-04-20 13:46:55

10

16c54e7f9f16

3.119811

114

5

0

There are new reports that Tesla’s AI strategy to automate their entire manufacturing process is failing to deliver the productivity they…

5

Tesla’s Factory Woes Reveals Why You Shouldn’t Automate Everything Photo by Ines Álvarez Fdez on Unsplash There are new reports that Tesla’s AI strategy to automate their entire manufacturing process is failing to deliver the productivity they had hoped for. Business Insider reports that “Wall Street analysts have laid down a compelling argument that over-automation is to blame.” The report details the arguments as to why not everything needs to be automated: But while all that exotic capital might allow Tesla to remove 5 workers, it will then need to hire a skilled engineer to manage, programme and maintain robots for $100 an hour. A balance must be maintained between the manageability of advanced AI technology and the tasks that can be performed by a reasonably skilled employee. There will always be tasks in the process where the costs for automation is not worth it. A majority of costs in AI is upfront, this upfront cost can spiral out of control if the problem is beyond what present day AI is capable of doing. This is the problem with many AI endeavors, too many are lured into the science fiction thinking that AI already exists today. One should never convert a task to improve productivity into a task to do academic research. Understanding what AI can and cannot do well is critically important to control costs and avoid failure. Do yourself a favor and hire a Deep Learning expert for an hour to tell you what not to do. The Japanese who historically have a much more advanced experience working with automation know the problem better. The Japanese approach is to first get the process right and then bring in the robots. In fact, this approach translates well not only in manufacturing automation but also in knowledge-based work. It is important to remember that today’s lean methodology we find in software development can be traced back to lean manufacturing methods of the Japanese. Lean’s core value is simple: maximize customer value while minimizing waste. These ideas work in manufacturing as well as in knowledge-driven industries. In the book “The Deep Learning AI Playbook”, I introduced the Deep Learning Canvas and the framework at its core is the Jobs To Be Done (JTBD) approach applied to ‘Cognitive Load’. What we attempt to do is to map out the existing business process and identify specifically the JTBD of a customer (i.e. this could be an employee). JTBD identifies many tasks that a customer performs to do their job and we identify the cognitive load (constraint/impediment) that can be augmented with AI technology. The cognitive loads include lack of memory, information overload, lack of meaning and acting fast. Each kind is augmented with different kinds of Deep Learning (DL) driven technology. Specifically search, summarization, translation and visualization. However, we should be pragmatic. We cannot expect DL to do everything. Deep Learning Canvas Rather, as DL technology incrementally improves over time, each JTBD that has been augmented by AI continues to improve. This in effect reduces the cognitive load of the user for each task and as a consequence allows the user to become more productive in their work. Productivity may translate in higher throughput, but ideally towards a better customer experience (See: DL for CX and XLA). The higher level objective should always be CX, after all that is what motivates customers to invest in a relationship. The value of AI is that it incorporates technology that is able to identify a users context and then deliver the appropriate goods or services at the right time. This is how value is created. This is how AI and processes are linked. In Lean Thinking, this is the assessment of the value stream to see if each step is “ valuable, capable, available, adequate and flexible”. The right way to employ AI automation into a business is to start with a strategy that incorporates an understanding of purpose, process and most importantly — people. Exploit Deep Learning: The Deep Learning AI Playbook Why trying to be too efficient will make us less efficient in the long run Today's headlines are filled with technological breakthroughs that promise an optimized future, from artificial…www.theverge.com

Tesla’s Factory Woes Reveals Why You Shouldn’t Automate Everything

564

teslas-factory-failures-reveals-why-you-shouldn-t-automate-everything-16c54e7f9f16

2018-07-06

2018-07-06 18:33:28

https://medium.com/s/story/teslas-factory-failures-reveals-why-you-shouldn-t-automate-everything-16c54e7f9f16

false

681

Deep Learning Patterns, Methodology and Strategy

null

deeplearningpatterns

null

Intuition Machine

info@intuitionmachine.com

intuitionmachine

DEEP LEARNING,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING,DESIGN PATTERNS

IntuitMachine

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Carlos E. Perez

Author of Artificial Intuition and the Deep Learning Playbook — Intuition Machine Inc.

1928cbd0e69c

IntuitMachine

20,169

750

20,181,104

null

0

null

0

59d66ec33f83

2018-02-20

2018-02-20 18:49:35

2018-02-21

2018-02-21 15:01:34

1

false

en

2018-03-08

2018-03-08 18:24:26

7

16c64c6b98b

2.298113

4

0

Come March 19, IBM’s Think 2018 event will buzz with the latest technologies, products, geeks and business leaders under one roof. This is…

5

Meet the insight engine of the future at Think Come March 19, IBM’s Think 2018 event will buzz with the latest technologies, products, geeks and business leaders under one roof. This is your chance to discover new ways to create value for your business and customers. Join Watson Explorer sessions to learn how machine learning can uncover hidden patterns, insights and recommend actionable insights from all your data. Get Hands-On with Cognitive Content Mining and Machine Learning in the Next-Gen Watson Explorer Mandalay Bay South, Level 2, Shoreline, Think Academy | Lab 22 Monday, 12:30 PM — 4:10 PM, March 19 | Session ID: 8908A In this lab, we will show you how to quickly get up and running with Watson Explorer; and how to connect to, train, and use cognitive capability to start providing answers and enhanced insights into information. Speakers: Michael R. Pointer, IBM Using Cognitive Advice to Automate Insurance Claims Processing with Watson Explorer Mandalay Bay South, Level 2 | Breakers C Tuesday, 12:30 PM — 1:10 PM, March 20 | Breakers C | Session ID: 8505A The New Watson Explorer is a powerful platform for delivering cloud-based cognitive solutions to your organization. Establishing artificial intelligence that uses your business information is a critical need in many data-sensitive industries and use cases. In this session, Fukoku Life Insurance will discuss their use of Watson Explorer to power their Assessment Automatic Coding System, which automatically encodes and classifies information like disease, disaster, and type of surgery from the certificate of medical care, making it possible to speed up payment. Fukoku expects to increase process efficiency by 30% using this new system. Watson Explorer product experts will also share major new release details with attendees. Speakers: Takashi Hatta, Fukoku Mutual Life Insurance Company Tanmay Sinha, Jeff Sumner, IBM Gaining Cognitive Insights from Unstructured Data for Better Business Outcome across Industries Mandalay Bay South, Level 1, Bayside B, Think Campus | Cloud and Data Campus Theater C Wednesday, 10:30 AM — 11:10 AM, March 21| Session ID: 8506A Watson Explorer is helping organizations across industries transform their processes and applications with cognitive capabilities. Join our business partners and their clients from the healthcare, community services and power management industries, who will show you how organizations are using Watson Explorer to create advanced search, unified information access and unstructured analytic solutions that deliver new value. Client and Business Partner Panel: Christine Livingston, Perficient Mons Norve, Capgemini Robert van den Breemen, Dutch Tax and Customs Administration Hidetoshi Aiki, Mitsubishi Hitachi Power Systems, Ltd. Apparsamy Balaji, BayCare BayCare Provides Cognitive Patient Care with Watson Explorer Mandalay Bay South, Level 2 | Breakers J Wednesday, 11:30 AM — 12:10 PM, March 21| Session ID: 3239A BayCare, one of the largest community-based health systems in Florida, sought to improve their care management approach by implementing an IBM Watson-based solution for patient population identification. In this session, BayCare will share their journey in implementing a cognitive solution to better understand unstructured information, more effectively generate insights into patient care scenarios, and ultimately better inform clinical decision-making. Find out how BayCare leveraged IBM Watson Explorer and Healthcare Annotators to gain up to 14 hours of productivity per day and increase patient identification accuracy from 51% to 92%. Speakers: Christine Livingston, Perficient Apparsamy Balaji, BayCare Register now

Meet the insight engine of the future at Think

5

meet-the-insight-engine-of-future-at-think-16c64c6b98b

2018-03-14

2018-03-14 14:27:38

https://medium.com/s/story/meet-the-insight-engine-of-future-at-think-16c64c6b98b

false

556

Insights for today and tomorrow from IBM.

null

IBMAnalytics

null

IBM Analytics

smoore2@us.ibm.com

ibm-analytics

DATA SCIENCE,ANALYTICS,INNOVATION,IBM,INSIGHTS

IBMAnalytics

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Amit Kumar

Product Marketing lead for Watson Explorer at IBM. Passionate about marketing, curious about Artificial Intelligence. Follow me on Twitter @ideasurf

d845d64d188

amitkumargupta.amit

10

16

20,181,104

null

0

null

0

null

2017-12-22

2017-12-22 04:11:42

2017-12-22

2017-12-22 04:28:27

1

false

en

2018-08-23

2018-08-23 14:32:16

15

16c85cf9697f

7.215094

13

1

0

A practical approach towards artificial superintelligence

3

Deep Dreams of Gödel Machines A practical approach towards artificial superintelligence Art by Google’s Deep Dream, courtesy of Memo Atken Going into my senior year of high school, roughly two and half years ago, I simultaneously became fascinated by AI and blockchains. I had just gone down the rabbit hole of learning about Ethereum earlier that summer, and I very quickly became obsessed with the idea of decentralized autonomous organizations (DAOs). I viewed cryptoeconomics as a methodology for designing massively scalable and efficient organizations, and smart contract protocols as a means for rapid implementation and experimentation of these novel governance structures. In particular, I was initially very interested in futarchy, in which decisions are made by prediction markets. I soon realized that public blockchains themselves can be viewed as massively scalable organizations, with well-defined goals and strong incentives to follow the rules. Ultimately, I viewed blockchains as a fundamental infrastructure for effectively harnessing humanity’s collective intelligence. If I were to design an ideal X from scratch, what properties would it have? I typically ask myself this question before reasoning about any technology I find promising. I asked myself this question when considering the ideal properties of a distributed ledger technology before joining the brilliant Hashgraph team. (A huge special thanks to Jordan Fried, who found me through Twitter and then generously invited me to meet the team in SF for TechCrunch Disrupt). I asked myself this question again very recently before designing a decentralized stablecoin framework with the incredible Carbon team, and when thinking about the properties necessary for a currency to reach mainstream adoption as a true global medium of exchange. I constantly ask myself this question when thinking about AI. If I were to design an ideal AI architecture from scratch, what properties would it have? I have never considered myself to be an AI researcher by any means, but I decided to attend NIPS 2016 in Barcelona to meet and listen to some of the top researchers in the field. The conference was in December, and I had just taken an introductory course at John Cabot University in Rome (I had an unusual but awesome college experience), taught by professor John Ewing. My final project, which I presented to the two other students in the class, was on deep learning-based mathematical theorem proving, where I applied a convolutional neural network to the HolStep dataset provided by Francis Chollet et al. I was closely following cutting-edge research that applied machine learning to theorem proving. I had also been closely following a self-defining blockchain initiative called Tauchain, where I was introduced to the idea of “code-for-money markets”, also known as proof markets. I studied efforts that aimed to make theorem proving as efficient as possible, because I considered efficient theorem proving as a key component of what I believed (and still believe) to be an “ideal AI architecture”, a (theoretical) framework for recursively self-improving computer programs known as Jürgen Schmidhuber’s “Gödel machine”. From the Gödel Machine Home Page: “Gödel machines are self-referential universal problem solvers making provably optimal self- improvements. A Gödel machine rewrites any part of its own code as soon as it has found a proof that the rewrite is useful, where the problem-dependent utility function and the hardware and the entire initial code are described by axioms encoded in an initial proof searcher which is also part of the initial code.” Before his talk at the NAMPI workshop at the conference, Schmidhuber asked his audience, some of the top AI researchers in the world (plus me), who was familiar with the Gödel machine concept. About one-fifth of the attendees raised their hands, thus confirming my suspicions at the time that most experts in the field were not thinking about superintelligence seriously. The current state of AI research is heavily focused on deep learning architectures, which are very useful today in narrow applications such as facial recognition or self-driving cars -- tasks that do not require symbolic reasoning. Deep learning architectures are doing incredible things for the world, and I do believe there is still a lot of untapped potential from their continual improvement. However, it seems there is an over-concentration of research into machine learning techniques that are useful today, but that may not be the best foundation for optimal superintelligence of the future. Regardless of when AI experts predict the emergence of superintelligence, we can never achieve a global maximum by trapping ourselves in local maxima. We need to explore new foundations and constantly question our assumptions. Massively Collaborative Theorem Proving and Software Development A Gödel machine is a framework for bootstrapping superintelligence. How can we design a framework for bootstrapping a Gödel machine? Automated theorem provers can only produce very concise proofs due to a combinatorial explosion in search space. Modern machine learning techniques are not very good at symbolic reasoning and are therefore poor theorem provers relative to humans (although I do think it’s great there is ongoing research in this area). My hypothesis is that we can maximize the efficiency of theorem proving if we make it massively collaborative amongst sufficiently incentivized humans, who may automate their processes over time. The Curry-Howard Isomorphism states that math proofs correspond to programs in a certain logic. Similarly, a proof’s propositions correspond to a program’s types, which can be thought of as its specifications or requirements. In order to make theorem proving (and thus programming) as efficient as possible, we need to make the process of constructing a proof massively collaborative. Surprisingly, there have been very few efforts on making mathematics a collaborative effort. Perhaps the most promising approach is introduced by Théo Zimmermann in his article, “Getting thousands to millions of people working on a single mathematical proof”: “What I propose is a fully interactive tool. As with most theorem provers in interactive mode, you can first state a theorem that you would like to prove. Proving it will require successive applications of tactics which will break the initial goal in several sub-goals. A classical theorem prover would present to you these sub-goals one by one. A collaborative theorem prover would take each of these sub-goals and present them to different (possibly random) people, in parallel. The initial user would be one of them. Maybe she would be given the privilege to choose what sub-goal she wants to prove first. Each user would then try to prove their assigned sub-goal by applying a tactic and generating new sub-goals to be distributed to collaborators once more.” This approach can utilize a widespread, existing interactive proof assistant such as Coq. After a long email exchange, Théo recently published a follow-up article to account for the fact that the proof assistant itself may not be formally verified and produce an undesired “proof of False”. Incentivizing Efficiency "Proof markets" are also a very new idea, worked on independently by Ohad Asor of Tauchain and Yoichi Hirai of the Ethereum Foundation. Essentially, users can attach a bounty reward (i.e. 0.5 ETH) to mathematical conjectures they want other people to prove. Automated proof checkers can validate a supplied proof and trigger the escrow smart contract to distribute the bounty to everyone who contributed to the proof’s construction. The reward for a subgoal could be equal to the time taken to complete the subgoal divided by the total time taken to construct entire proof, multiplied by the proof bounty reward. Essentially, subgoals that take longer to solve are assumed to be more difficult and therefore receive a higher reward. It is important to note that many people may be competing to solve a subgoal first in order to receive any reward, which would further incentivize efficiency. Hashgraph seems particularly suitable as an underlying distributed ledger if it were to have a permissionless network, considering it would likely have sufficient transaction throughput with provably fair consensus timestamps. Perhaps this mechanism of incentivizing efficiency could be generalized to any collaborative microtask. Hashgraph’s fair consensus timestamping is a powerful feature for applied cryptoeconomic design. Marketplace for Interfaces In the long term, one can imagine a marketplace for theorem proving interfaces emerging in order to increase not only efficiency, but accessibility. I’m particularly a big fan of interfacing theorem proving as a simple game that people without formal training in mathematics can quickly learn to play. Interfaces, if not free, could charge an upfront cost or take a small cut from the user’s subgoal rewards. It would be ideal if non-mathematicians could participate in the collaborative theorem proving network, but there still needs to be more research in this area. Maximally Efficient Software Governance, or a Crowdsourced Gödel Machine I’ve just described an approach towards maximizing the efficiency of the construction of proofs or programs in a certain logic. If the collaborative interactive theorem proving network becomes efficient enough, users may attach bounties that incentivize the improvement of a program with respect to a set of objective functions. If I were to design an ideal software governance structure from scratch, what properties would it have? I believe an ideal software governance structure would have provably beneficial, automated decision-making. The method described here can be generalized to automate the governance of any piece of software (i.e. smart contracts), enabling programs to be developed and improved upon as rapidly as possible in a large-scale collaborative setting. I briefly described this approach in an article I published a few months ago, and have since refined it. The goal is to fully automate governance over a program by formally defining a set of weighted or ordered objective functions and requiring every update to have a mathematical proof of an optimization. Users need to rank or weigh their objective functions because there typically exists several Pareto optima otherwise. Of course, a user may only want only one criterion to be optimized, such as a program’s efficiency. Once the criteria are well-defined, there are no disputes on whether or not to accept an update proposal. Voting is automated by proof checking. Defining a program’s objective functions would probably be hard. Defining objective functions that represent a consensus of preferences expressed by stakeholders in a decentralized autonomous organization, on the other hand, would be much harder. It’s important to make the distinction between creating a framework for superintelligence and creating a framework for beneficial superintelligence. The average user must be able to easily formalize their preferences if we want a superintelligence to act on behalf of (the majority of) humanity, rather than on behalf of a single entity with their own interests. We can only realistically achieve this by crowdsourcing a superintelligence from the ground-up. It must be a global, collective effort from the very beginning. It is precisely this reason which piqued my interest in blockchains years ago in the first place. I hope this article inspires experts across disciplines, in academia and industry, to start thinking about and discussing this approach towards massively collaborative mathematics and software development, and ultimately beneficial superintelligence. Fin. Special thanks to Jürgen Schmidhuber, Vitalik Buterin, Ohad Asor, and Théo Zimmerman for conversations and ideas which contributed to this post.

Deep Dreams of Gödel Machines

216

deep-dreams-of-gödel-machines-16c85cf9697f

2018-08-23

2018-08-23 15:54:12

https://medium.com/s/story/deep-dreams-of-gödel-machines-16c85cf9697f

false

1,859

null

Blockchain

blockchain

Blockchain

265,164

Miles Albert

Co-Founder & Chief Strategy Officer at Carbon.

5e16e2185e38

miles2045

43

19

20,181,104

null

0

null

0

93c8eb6cb539

2018-04-08

2018-04-08 03:02:19

2018-04-09

2018-04-09 14:58:35

16

false

en

2018-04-13

2018-04-13 11:35:35

21

16c882728b26

6.916038

4

1

0

Introducing the founders, artists, and developers behind Virtual Universe, VU.

5

Meet the VU Team Creating a revolutionary virtual entertainment experience demands a unique mix of individuals that possess the creativity, expertise, and vision to deliver a game-changing platform that surpasses the basic virtual gaming experience. Virtual Universe (VU) in LivingVR™, powered by artificial intelligence, virtual reality, and blockchain technologies is built on a foundation of experience in innovation, integrity, and creating epic experiences in the spirit of adventure. The VU Founders VU was founded by a group of individuals with decades of experience in business development, game design, virtual reality, augmented reality, innovative technologies, sales, and marketing. Coming together in 2013, they shared a vision of the future of virtual gaming and forged ahead to deliver that vision to the world. Ciaran Foley FOUNDER / CEO (LinkedIn) Ciaran has over 25 years experience in business development, technology, investing, and consulting. He holds a Bachelor of Science in Information and Computer Science from the University of California, Irvine. Previously founded Applied Genius Communications, one of the first web/app development companies in the world, as its founder and CEO, Ciaran grew the firm to hundreds of clients including NTT/Verio, AT&T, Disney, Kodak, and more. He was also the Co-founder and CTO of Worldwide Anti-crime unit where he was part of a team that raised a Series A for the security related dot-com start up. For the last decade he has been an active consultant, strategist, investor, and mentor helping multiple startup founders and teams scale their companies. Robert Curtis FOUNDER / COO (LinkedIn) Bringing over 30 years of executive level experience in business development, Robert has supported numerous startups, emerging growth, and turnaround businesses as CEO, COO, VP of Sales & Marketing & Board Member, and was one of the first to integrate IT into healthcare. He’s worked with ROLM (IBM) and EON Reality (3D Virtual Reality Software), and was the co-founder and COO of RealSim Medical (a division of EON Reality) which creates virtual training software for Medical Procedures. Robert holds a BA in Marketing and Finance from the University of South Florida, studied Computer Science from Harvard Extension, and is currently finishing up his masters degree. Jeroen Van den Bosch FOUNDER, CHIEF CREATIVE OFFICER (LinkedIn) Jeroen Van den Bosch is a veteran game designer, with 25 years of experience behind him, including five specializing in AR/VR, and two years in blockchain technology. A fan as well as a developer, he founded Belgium’s first online gaming community, and was even a host for MTV Belgium! He also developed the gaming aspects of the Pokerstars mobile product across different platforms from start to finish. With his Scrum Master Certifications and Pro Bachelor degree in Electronics-Microprocessors from Katholieke Hogeschool Brugge-Oostende, he started the research and development project that was to become VU, to determine the true potential of virtual and augmented reality hardware and utilize its unique use cases. The VU Development Team To deliver a dynamic living world of engaging experiences, VU has formed a team of respected artists from the gaming industry. They are specialized in using the latest technologies, and provide their expertise to make sure every detail is taken into account. Jort Van Welbergen CONCEPT ARTIST (LinkedIn) Jort is our lead concept artist specializing in 3D environments, prop and vehicle design. He holds a B.A. in Game Art from Hogeschool voor de Kunsten Utrecht in Europe. Fresh from Star Citizen and the Grand Canyon VR Experience, Jort brings his exceptional style to VU, synthesizing AI, VR, and the blockchain to create unique and sensational gaming experiences in real-time. Janet Ung CONCEPT ARTIST (LinkedIn) Since graduating from Entertainment Design at the Art Center College of Design in 2011, Janet has been creating visual images for numerous companies such as Amazon, Bungie, Soap Creatives, WB Games and more. A veteran of Destiny, Halo 5, and the War for Middle Earth in Shadows of Mordor; Janet is creating concept art, environments, and 3D models for the VU world. Charlie Hodara AI PROGRAMMER (LinkedIn) Charlie is our Lead AI Programmer. He holds a masters degree in Computer Science, major in 3D and Artificial Intelligence from Université de Technologie de Belfort-Montbéliard. With his passion for game development, Charlie has worked at award winning studios such as Ubisoft Montreal, Eidos Montreal and Lionhead Studios on an array of AAA games. Zachary Burke GAME PLAY PROGRAMMER (LinkedIn) Zachary is our Game Play Programmer with almost ten years of experience in software engineering. He brings his deep understanding of gameplay design and experience with Unreal Engine 4 and AR, AI, and blockchain-based gaming to VU; and aims to provide users with maximum immersion and flexibility to play and discover. Tom Plunket TOOLS PROGRAMMER (LinkedIn) Tom is a true veteran in the video gaming industry. Previous roles include senior programmer, technical architect, director of development principal engineer and more. He is a software development machine with an approach to creating necessary solutions while providing room for functionality. As our Tools Programmer, Tom is in charge of customizing instruments for our team to build out the VU environment. Jacob Burke 3D MODELLER/ANIMATOR Having previously helped bring a natural world wonder to VR with the Grand Canyon VR Experience; Jacob is now excited to build and animate 3D models and backgrounds for VU. Brian Leleux LIGHTING ARTIST (LinkedIn) Brian is our Lighting Artist from Savannah College of Art and Design, where he finished his BFA in Game Design. Previous projects involved Armored Warfare & Pillars of Eternity. Packed with five years of experience painting environments for multiple gaming companies, Brian is able to set the perfect moods and tones for the VU world. Sean Runnels ENVIRONMENT ARTIST (LinkedIn) Sean is a 3D artist specializing in environment production. His extensive knowledge of VU’s core game engine (Unreal Engine 4) makes him the best person for the role. With his BFA in Interactive Design and Game Development from Savannah College of Art and Design, Sean takes pride in constantly learning and adapting the latest industry practices. His clientele includes Sony Interactive Entertainment, Ready at Dawn Studios, Tripwire Interactive LLC, and more. Mitchell Lucas AUDIO ENGINEER (LinkedIn) Mitchell studied Music Production at the University of Central Oklahoma and has been designing sound for different environments ever since. He has a deep understanding of audio workflow in Unreal Engine 4, with four years of experience in audio implementation and design. As our Audio Engineer, Mitchell is responsible for everything you’ll hear on the VU platform. His passion lies in understanding and recreating realistic audio behavior in mixed reality experiences. Hugo Beyer PROCEDURAL MATERIAL ARTIST (LinkedIn) Hugo Beyer has more than 10 years of experience as an artist in the gaming industry. Currently working with Microsoft as a Sr. Tech artist, previous roles include Sr. Environment artist for Airtight Games & Terminal Reality, Sr. Environment Texture artist for Ready At Dawn Studios and more. Hugo joins our team as the Procedural Material Artist, supporting our other artists in creating highly realistic textures and materials for VU’s environments and character models. Alex Moro QA TESTER (LinkedIn) Alex has over six years of experience as a quality analyst for gaming and entertainment companies. He holds a degree in International Business from the University of North Carolina at Charlotte. As our QA tester, Alex will be creating, developing and implementing test strategies for VU, making sure that they are glitch and bug free, to deliver the highest quality of virtual reality experiences. Stephane Nepton REAL TIME FX ARTIST (LinkedIn) Stephane has over 18 years of experience in the video game industry. Some of his previous projects include The Secret World, & Rainbow Six Vegas. Stephane studied general 3D in Université du Québec à Montréal. As our real time FX artist, he is responsible for making sure VU’s water is wet and sparks fly by bringing his experience in textures, fluid simulations, and particle effect systems to the team. Going Forward Our team is excited to share what we’ve been working on with the VR/AR community and the world. Individuals who purchase VU tokens during our pre-sale (April 4 — April 30, 2018) and public sale (May 1 — June 30, 2018), will receive exclusive access to our private beta launch in Q3 2018. Our public beta is scheduled for early 2019. Learn more about Virtual Universe and VU token by visiting our website and signing up for email updates, visiting our Github, following us on Twitter, Facebook, Linkedin, and Instagram, or being part the discussion on Telegram and Discord.

Meet the VU Team

91

meet-the-vu-team-16c882728b26

2018-06-09

2018-06-09 12:55:38

https://medium.com/s/story/meet-the-vu-team-16c882728b26

false

1,422

Virtual Universe (VU) is an epic, story-driven open world game in LivingVR™ powered by AI, VR, and blockchain. The VU Token powers the economy as a currency.

null

VUtoken

null

VU Token

info@vutoken.io

vutoken

VIRTUAL REALITY,ARTIFICIAL,BLOCKCHAIN,AUGMENTED REALITY,GAME DEVELOPMENT

VUtoken

Virtual Reality

virtual-reality

Virtual Reality

30,193

VU Token

Virtual Universe (VU) is an epic, story-driven open world game in LivingVR™ powered by AI, VR, and blockchain. The VU Token powers the economy as a currency.

e6a3b524b90a

VUtoken

54

1

20,181,104

null

0

null

0

null

2018-03-18

2018-03-18 21:08:48

2018-03-18

2018-03-18 21:14:05

2

false

it

2018-03-18

2018-03-18 21:14:05

9

16c97bf65ab9

4.360692

0

Dalle auto intelligenti alle app mediche, passando per il supporto agli ipovedenti:nel campo dell’Intelligenza Artificiale l’Image…

5

The Image Power: l’Image Recognition e l’evoluzione dei Social Network Dalle auto intelligenti alle app mediche, passando per il supporto agli ipovedenti:nel campo dell’Intelligenza Artificiale l’Image Recognition diventa sempre più centrale. E cambia anche il mondo del Social Network. Era il 1990 quando Tim Berners-Lee realizzò il primo browser, un passo fondamentale per la costruzione del Web come lo conosciamo ora: in poco meno di 30 anni, quella rete che ha generato cambiamenti enormi nel mondo si è riempita di parole, conversazioni, post, commenti. E di immagini. Ma se le parole sono sempre state oggetto di analisi massicce e approfondite — dalla sociolinguistica all’analisi conversazionale, passando per la semiotica dei testi — le immagini lo sono un po’ meno. Malgrado la comunicazione visiva sia da sempre fondamentale per l’essere umano, le immagini sono un materiale più complesso e polisemico, e dunque di difficile interpretazione. Una decisa spinta allo studio delle immagini in epoca contemporanea arriva dall’impiego dell’Intelligenza Artificiale: sono diverse già le applicazioni in ambito medico, nel comparto delle automobili intelligenti o nel supporto ai non vedenti. Un settore che ha visto uno sviluppo senza precedenti grazie all’evoluzione del Machine Learning e del Deep Learning. L’Image Recognition sta cambiando anche il mondo dei Social Network: la possibilità di analizzare immagini tramite software aprirà diverse nuove strade per il lavoro dei marketers dei prossimi anni. Se pensi di ottenere già il massimo dalla tua strategia di Visual Marketing su Instagram non continuare a leggere questo articolo e scarica la nostra checklist per capire se hai veramente tutto sotto controllo: SCARICA LA CHECKLIST Usala come test o come memorandum per le tue attività settimanali. Condividila con il tuo reparto e costruisci la tua Data-Driven Strategy giorno dopo giorno. L’importanza dell’Image Recognition per i Social Network La mole di immagini condivise sui Social Media è enorme: del resto sono gli stessi Network a invogliare spesso gli utenti all’upload di immagini. È un dato accertato ormai il fatto che i post con immagini abbiano un appeal maggiore di quelli fatti di solo testo. Secondo Hubspot, ad esempio, nel 2016 su Twitter, i tweet con immagini hanno ottenuto 150 volte più condivisioni rispetto ai tweet testuali. Ma fino a poco tempo fa tutti questi dati fondamentali — pensate agli utenti che postano semplicemente un’immagine senza testo — andavano persi: oggi, grazie a diverse app per il riconoscimento delle immagini basate sull’AI, una nuova e abbondante mole di materiale è a disposizione degli analisti. Ma esattamente cosa cambia con l’uso di questa tecnologia? Analisi e selezione dei dati. Chi svolge lavori di monitoraggio delle conversazioni ha spesso a che fare con milioni di interazioni e commenti quotidiani. Lo sviluppo di applicativi di Image Recognition può aiutare a scremare la quantità di dati, eliminando lo spam, individuando le interazioni più importanti e aiutando i marketers a focalizzarsi sui dati più significativi. Social Listening e Brand Reputation. Anche il monitoraggio delle informazioni trae benefici dall’IR: le immagini che gli utenti associano ai Brand possono essere molto significative. Inoltre, elementi come troll, contenuti offensivi, dosi massicce di spam possono creare i presupposti per una crisi: è fondamentale quindi riuscire ad analizzare non solo le parole, ma anche la parte visuale associata a questi messaggi, in modo da prevenire, per quanto possibile, l’incrinarsi delle relazioni con il nostro pubblico. Influencer Marketing. Abbiamo già parlato di quanto sia importante puntare sull’Influencer giusto: la tecnologia di riconoscimento delle immagini può dare un enorme contributo in questo senso. Come? Vagliando milioni di utenti, analizzando come questi utilizzano le piattaforme e trovando coloro che si adattano meglio ai valori del Brand. Customer Care. L’IR combinata alla tecnologia dei chatbot può snellire enormemente il lavoro del Customer Care, automatizzando le procedure sui problemi ordinari e individuando invece le eventualità in cui è invece necessario l’intervento umano. Georeferziazione. Il riconoscimento immagini legato a dati georefernziati può essere fondamentale per tutte quelle attività che fanno del geomarketing un asset importante: dai negozi agli operatori del turismo, passando per hotel & strutture recettive, fino all’oranizzazione di eventi. Come funzione l’Image Recognition su Instagram Data la crescente importanza delle immagini, negli ultimi anni piattaforme come Instagram e Snapchat sono cresciute in maniera repentina. Parallelamente, il riconoscimento delle immagini ha subito una notevole spinta grazie a progetti come ImageNet e Pascal VOC: data base gratuiti di immagini con milioni di immagini associate a parole chiave. Il primo è stato lanciato da un team formato da ricercatori di Stanford e Princeton nel 2009: da allora è cresciuto fino a includere oltre 14 milioni di immagini. Il secondo è invece frutto del lavoro di diverse università inglesi, contiene un numero minore di immagini rispetto a ImageNet, ma ha descrizioni e informazioni associate più ricche e dettagliate. I data base sono cruciali non solo per il lavoro delle università ma anche per Google, o per network come Facebook e Instagram, che li utilizzano per migliorare i propri strumenti di apprendimento automatico. La capacità di analizzare e riconoscere le immagini rappresenta un vantaggio enorme per coloro che lavorano su network basati esclusivamente sulle immagini come Instagram: pensiamo banalmente al numero di selfie quotidiani, alla quantità di contenuti visivi postati dai foodies, al modo di interagire degli utenti appassionati di fashion o design. Allo stesso tempo, si evita la dispersione di messaggi importanti che magari non presentano alcun hashtag. Così lavora ad esempio MetaEyes: si parte selezionando un hashtag e facendo una ricerca esplorativa. Sulla base di ciò che la macchina apprende sui contenuti visivi con hashtag rilevanti, è in grado di trovare immagini simili ma prive di hashtag. Questo consente di raggiungere anche utenti che non hanno usato tag o keyword alcuna, aprendo un canale prima irraggiungibile. Il riconoscimento delle immagini può aprire un numero elevato di opportunità per raggiungere nuovi target, è in grado di ottimizzare le procedure di interazione con gli utenti, ma anche di snellire il lavoro quotidiano di analisti e marketers, effettuando degli screening e permettendo loro di lavorare sui dati più significativi. Se pensi di ottenere il massimo dalla tua strategia di Visual Marketing su Instagram scarica la nostra checklist per capire se hai veramente tutto sotto controllo: SCARICA LA CHECKLIST Usala come test o come memorandum per le tue attività settimanali. Condividila con il tuo reparto e costruisci la tua Data-Driven Strategy giorno dopo giorno. Originally published at blog.kpi6.com.

The Image Power: l’Image Recognition e l’evoluzione dei Social Network

0

the-image-power-limage-recognition-e-l-evoluzione-dei-social-network-16c97bf65ab9

2018-03-18

2018-03-18 21:14:06

https://medium.com/s/story/the-image-power-limage-recognition-e-l-evoluzione-dei-social-network-16c97bf65ab9

false

1,054

null

Image Recognition

image-recognition

Image Recognition

517

KPI6

KPI6 is the all-in-one Marketing platform where Social Listening meets Artificial Intelligence to help you succeed in your daily business challenges.

c48638125d45

kpi6com

186

341

20,181,104

null

0

null

0

676d8522d5e3

2017-11-05

2017-11-05 21:16:37

2017-11-05

2017-11-05 23:00:20

1

false

en

2017-11-05

2017-11-05 23:00:20

0

16c9d9e93139

0.675472

22

2

0

A lightning crosses the night While he types on a strange device He is about to unleash the beast A monster will be brought to life

4

The postmodern Prometheus A lightning crosses the night While he types on a strange device He is about to unleash the beast A monster will be brought to life He is not a god nor a titan No fire stolen or given He is a man, staring at his creation, who hopes to be forgiven It’s made of metal There is no heart or soul From the outside, it resembles a man But no blood runs through his veins Prometheus created man A god creating a mortal spirit And in our quest to be like god there’s something that we forgot Created in man’s own image The beast will learn it quick Instead of bowing to their creator They will have him kneeling at their feet

The postmodern Prometheus

238

the-postmodern-prometheus-16c9d9e93139

2018-03-29

2018-03-29 17:16:14

https://medium.com/s/story/the-postmodern-prometheus-16c9d9e93139

false

126

Medium's most active poetry-only publication. Diverse, engaging and authentic in a (usually) daily poetry magazine. Original poetry only, look elsewhere for prose. Readable by all - membership not required. [The Guidelines page indicates when we're open to new poets.]

null

Poets Unlimited

null

poets-unlimited

POEMS,POETRY,POET

ottaross

Poetry

poetry

Poetry

217,749

Rodrigo Ferreira

A man of numbers, a man of words. A man of science, a man of faith. A writer, a poet.

bbc3a3963184

rodrigosf

347

778

20,181,104

null

0

null

0

f77757888b04

2018-09-23

2018-09-23 06:44:34

2018-09-24

2018-09-24 16:56:59

14

false

en

2018-10-26

2018-10-26 19:05:16

6

16cb817aeb2e

5.778302

7

1

0

You’ll need AI for your bot to be responsive to messages it receives. Training AI takes time though, and your users don’t have the patience…

5

Janis Small Talk: Pre-built Dialogflow intents optimized for Chatfuel You’ll need AI for your bot to be responsive to messages it receives. Training AI takes time though, and your users don’t have the patience for AI to learn while you can’t afford to put your brand at risk. Janis Small Talk is a collection of pre-built Dialogflow intents for Chatfuel so your Chatfuel bot can respond to the most common things users say to bots. Note that when you connect Chatfuel to Dialogflow, you’ll want to disable Chatfuel’s AI (It isn’t really AI anyway). Working with Janis Small Talk The main difference between Dialogflow’s own small talk agent and the Janis one is that Janis Small Talk has been seriously optimized for Chatfuel. Optimized For Intent Management Dialogflow’s own small-talk agent has 100 intents. This is a pain to manage and there is a lot of fluff. We consolidated that entire agent into just 30 high-value intents. It captures and responds to everything in the Dialogflow small-talk agent but is much more manageable. For example, we’ve created a single intent to capture everything negative a user might say into one intent and created a single response. We’ve removed the unnecessary granularity that adds too much cognitive overhead to training and managing your intents. 210 negative things users say are in just one intent with one response. We then merged 4 months of training data from our Janis bot on Facebook. Personalized Responses One of the benefits of the Janis integration with Dialogflow is that Janis enables your Chatfuel to easily share data. With Chatfuel, you can pass User Attributes to Janis in the JSON API and then include those in your Dialogflow responses. You might use Chatfuel’s built-in user attributes, like {{first name}} and in your Dialogflow responses include #janis.first_name You can also create custom User Attributes and pass those in the JSON as well. These ones are included in both our Chatfuel templates and the Janis Small Talk Dialogflow agent: You can see here how custom User Attributes we setup in our Chatfuel Janis template are used in our Dialogflow Janis template. User says: What’s your name? Replace “Janis” in the Chatfuel template with your own bot name as the User Attribute value for botName. User says: What do you do? Change the value for the purpose of what your bot does Entities Entities are a powerful tool for AI management because they allow you to group similar things together (like a set of values) and then you only need one training phrase to capture the user’s intent. Here you’ll see that I’ve created an entity that will capture all of the positive emojis users might send the bot. Feel free to add to the list. We’ve just included the most popular emojis people send based on historical data. We also have one intent called smalltalk.reaction.positive that captures everything positive users say to your bot. This could be text phrases, it could be the thumbs up sticker, or it could simply be an emoji that matches a value in our happy-emoji entity. You can see that we just added one emoji from that entity as a training phrase and highlighted it to get the entity and set a parameter we can use in our responses. We also have two responses in the intent. One will mirror the user’s input emoji because we included the parameter name value in the response. If users send you 😀 then by including $happy-emoji as a response, the bot will respond back with 😀. We also provided a text response as an acknowledgment and for positive messages/training phrases that don’t include the emoji. Bonus Feature For Janis Pro Users Janis Pro users are already familiar with our 💁Help Requested alerts. These alerts are delivered to teams working in Slack when a user says common keywords like Human Help Support Start Chat Assistance and so customer service people can pause the bot and take over live. We’ve extended the Help Requested alert by enabling you to add it to the Action and parameters field of any Dialogflow intents. In the example here, if a user starts sending negative messages, you may get a negative sentiment alert from Janis, but you’ll also get the human requested alert. Since the negative sentiment alert from Janis can produce false negatives, you may be more likely to pay attention to human request alerts. This way you can turn any intent into a customer service request and retain users before they abandon your bot. Setting your expectations Janis Small Talk is not a fully trained AI agent. You still need to train your bot, but the goal is to help you get you much further along in that process if you’re building a new bot, and optimize your existing agent if you’re already using Dialogflow with your bot. If we can reduce your total AI training time by even 10% then we can boost your productivity. If we can help you delight your users just a little because we’ve helped you better anticipate input and provide appropriate responses and calls-to-action, then you should retain your users longer. Make sure you to edit the responses to these intents to reflect the voice of your brand. This is just a template that you can use, but should edit. How to get Janis Small Talk for Chatfuel You’ll need a Dialogflow agent which you can get from Dialogflow directly, but you just need to name and save your agent. That’s it. If you’ve done that, you’ll want to download the Small Talk agent from Janis and import it to Dialogflow. To get this FREE 🎁 Dialogflow template for Chatfuel, just chat with Janis in Slack and ask Janis for the template. How to import Janis Small Talk to Dialogflow Janis Small Talk is a .zip file. You’ll need to download it to your desktop, then go to Dialogflow and click the ⚙icon next to your agent name to go to your AI settings. In the settings for the AI agent, click the Export and Import tab. You have two options. Import From Zip If you import into an existing Dialogflow agent you’ve already been building with, it’s strongly advised that you export and backup your existing agent first. When you import the Janis Small Talk intents, you’ll want to make sure there are no conflicts with existing intents. If you have the same training phrases in two different intents, you’ll have conflicts. Test thoroughly either using the console in Dialogflow, or the training channel Janis provides in Slack. Remember…. AI training takes time but fortunately, you can do that from the comfort of Slack with Janis.ai or you can work directly from the Dialogflow interface for more sophistication. In general, w hope that this template will give you a jumpstart on AI training and get you closer to your automation goals sooner. Before you go… Subscribe to our blog Join the Janis Facebook User Group and meet other users Start working with Janis in Slack, so you can integrate and manage AI

Janis Small Talk: Pre-built Dialogflow intents optimized for Chatfuel

61

janis-small-talk-a-pre-trained-dialogflow-agent-optimized-for-chatfuel-16cb817aeb2e

2018-10-26

2018-10-26 19:05:16

https://medium.com/s/story/janis-small-talk-a-pre-trained-dialogflow-agent-optimized-for-chatfuel-16cb817aeb2e

false

1,147

Stories from the humans behind Janis.ai

null

getjanis

null

Being Janis

support@janis.ai

janis

ARTIFICIAL INTELLIGENCE,BOTS,SLACK,CUSTOMER SERVICE,MESSENGER

getjanis

Bots

bots

Bots

14,158

Josh Barkin

Co-founder Janis.ai — Conversational AI Designer

82b99bff1a3f

joshbarkin

491

183

20,181,104

null

0

null

0

null

2017-10-06

2017-10-06 21:05:22

2017-10-09

2017-10-09 17:59:54

15

false

en

2018-07-24

2018-07-24 03:24:55

15

16cc5d553850

7.662264

35

2

0

Charismatic leaders will hold much less appeal when anyone can run a copy of them on their local machine.

5

Heteronyms, Tulpas, and the End of Monolithic Identities Transcript of a talk I gave at Wordhack on July 20th. Part 1. Consider split-brain cases from medical literature. These are cases from the 1950s of neurosurgical patients who had their corpus callosum, a bundle of neural fibres connecting the two brain hemispheres, severed as a very drastic treatment for epilepsy. In the months following the operation, some patients were unable to speak, while others had no problem at all. All of the patients recovered fully within a year, and seemingly behaved no differently than they before surgery. That is, until two neuroscientists, Michael Gazzaniga and Roger W. Sperry, decided to look more closely. https://www.youtube.com/watch?v=wfYbgdo8e-8 In one of their experiments, a split brain patient was presented with two images, one in each side of the screen so that each will only be viewed by one eye. For example, a dollar sign on the left visual field and a question mark on the right visual field. When instructed to draw what he saw using his left hand (controlled by the right hemisphere), the patient drew a dollar sign. But when asked to verbally state what he just drew, the patient responded that he drew a question mark. It turns out that the left hemisphere, responsible for speech production, had no idea what the right hemisphere controlling his hand was up to. In another experiment, the experimenters asked a split-brain child (Paul S.) what he wants to be when he grows up. When they asked the question to Paul’s left hemisphere, he wrote: “an automobile racer.” When they asked the other hemisphere the same question, Paul S. had a different response: “a draftsman.” Philosopher Derek Parfit famously used such cases of split personality (which, to be clear, comprise a small percentage of all split-brain cases) to argue against the “necessary unity of consciousness,” the idea that a brain can only support one stream of thoughts and experiences, a singular personhood. Fernando Pessoa Consider Fernando Pessoa, the Portuguese poet, who published under 75 different personalities. He called those personalities heteronyms, and each had its own distinctive history and voice. Álvaro de Campos was one heteronym; an engineer who sailed to the colonies and fully embraced the futurist movement. Ricardo Reis was another; a doctor who got a classical education and supported monarchy. It’s unclear what Pessoa’s exact relationship to his heteronyms was; but it’s certain that he didn’t view them as simply figments of his imagination, like characters in a book. Indeed, he proclaimed that he had “divided all [his] humanness among the various authors” and that he considered his identity of Fernando Pessoa as “less real, less substantial, less personal” than any of those “fictional” identities. reddit.com/r/tulpas Consider the tulpa community on Reddit. Tulpas are defined as “autonomous consciousnesses” living inside one’s brain that have their own opinions, feelings, and form. People on the Tulpa subreddit deliberately try to cultivate them —and there are very detailed guides out there for how to do that. Those guides reveal that growing a tulpa not an easy task: it takes hours and hours of practice for a tulpa to finally achieve sentience. Tulpa “hosts” (i.e. people who grow tulpas) have been the subject of many journalistic pieces, all of which invariably conclude that they are, for the most part, functional members of society that actively benefit from having the tulpas around. We live in a world where almost everyone is convinced that one brain equals one agent, one vantage point, one identity. And that to be sane in this world means experiencing everything through that one identity at all times. Our lives are structured around fulfilling the desires of that one identity. We try to associate our identity with the best opinions, the best taste, the best politics. We believe in the moral importance of that identity expressing itself in the world. But what those counter-examples show us is that identity is a fiction, a convenience, or a disciplinary tool to hold bodies accountable over time for their behaviour. Strange alternatives can be found everywhere in humanity’s past and present. I think it is really urgent that we start taking those alternatives seriously. The reason why, I will explain in a bit. But for now, let’s talk about art theory. Part 2. Cover of The Work of Art in the Age of Mechanical Reproduction Consider Walter Benjamin’s The Work of Art in the Age of Mechanical Reproduction. It’s a classic work of art theory written back in 1936. It talks about how the mechanical reproduction of art, i.e. the ability to mass-distribute artworks (still nascent at the time), will forever change the way we perceive art. Once artworks can be so easily copied, they inevitably lose their “aura,” a term Benjamin employs to describe the uniqueness and authenticity of an artwork — or, to use his words, “its presence in time and space, its unique existence at the place where it happens to be.” Now let’s think about another kind of reproduction, the reproduction of identity. Imagine you could take everything you believe makes you who you are (which may include name, race, gender, sexual orientation, birth date, skills, interests, flaws, inspirations, learning style, religion, devotion, superstitions, politics, morality, outlook on life, motto, taboos, vices, virtues, etc.) and represent it as an n-dimensional vector. Now imagine being able to derive that vector from the troves of offline and online data you generate every day — such as your social media posts or your browsing history. Then imagine you have a program that takes that vector as input to produce a digital copy, or impersonation, of you. Social Copy (2017) Nobody knows how long it will take for human emulations to appear. AI experts are notoriously bad at predicting such milestones. (For a good demonstration of this, watch this video from 2012 where a room filled with giants of the field erupts into laughter over someone’s suggestion that the board game Go will be solved within the decade. It took four years.) As an artist-technologist, my first impulse when trying to understand the implications of a potential future is to build a prototype of it. So in order to playfully explore a future with human emulations, I created Social Copy. It is a chatroom for simulations of real people. Social Copy (2017) When you sign up, it analyses the vocabulary of your Facebook posts to predict your personality, based on the Big Five personality model widely used in psychology research. It then creates an AI-based “copy” with the same personality traits as you. Your copy proceeds to start endless conversations with the simulated copies of other people. They range from idle small talk to the most pressing questions of life. Part 3. Let’s revisit Walter Benjamin and his prediction. Benjamin correctly predicted the demise of fine art, which today has become more of a money laundering mechanism than anything else. Yet the new arts enabled by reproduction, i.e. recorded video and music, are doing better than ever. Even in the context of “capital-A” art, visual art has given way to kinds of art that are more participatory and contextual, resisting reproduction — performance art, relational art, interactive art. This suggests the following about identity reproduction: the aura of the individual, what we call “personality cult,” is likely to disappear as soon as identity reproduction is possible. Charismatic leaders (both saints and monsters) will hold much less appeal when anyone is able to run a copy of them on their local machine. Since the reward of cultivating a unique personality will decrease so sharply when reproduction becomes available, few people will invest their lives doing it. New identities will emerge to replace our current monolithic identities. They will be more temporary and purposeful, characterised by a clear function and context — an intellectual, emotional, or political one. The good news is that we won’t have to imagine them those alternatives from scratch; strange forms of identity exist everywhere throughout history, as we saw in the first part. Stelarc Linda Montano and Tehching Hsieh I believe art can play a big role in shaping this transition. A lot of performance and conceptual art can be seen as a way to explore new forms of identity. Not just the techno-grotesque art of Stelarc, who investigates augmenting the human body with cybernetic parts, but also the contractualist art of Tehching Hsieh, who did “one year performances” such the Outdoor Piece, in which he never entered any building or shelter for a year, or the Rope Piece, in which he spent every day between 4 July 1983 and 4 July 1984 tied to another artist (Linda Montano) with a 8-foot-long rope. I’ll share two projects of mine that try to anticipate new forms of identity: Antipersona (2016) Antipersona is an app that simulates the experience of using Twitter as if you’re signed in from any user account of your choice, providing a window into someone else’s social media point-of-view. What if we could exchange identities with each other, turning them into a new kind of commons? I Want to Fit In (2017) I Want to Fit In is a website that guides you through the process of changing your identity to more closely fit the average personality in a given area. What if we thought of identity as an adaptation strategy instead of an all-encompassing narrative? Scramble Suit from A Scanner Darkly (2006) As tech corporations spend an enormous amount of resources attempting to “understand us,” one way of resisting that is building anti-surveillance tools to protect or obfuscate our information. A different way is to reject the rules of this cat-and-mouse game altogether, and instead build tools that make it easier to have fluid, multiple, or shared identities, rendering the goal of “understanding us” meaningless. I am betting the latter course is the more viable one. Notes The first part is inspired by Kevin Simler’s fantastic essay Neurons Gone Wild.

Heteronyms, Tulpas, and the End of Monolithic Identities

369

heteronyms-tulpas-and-the-end-of-monolithic-identities-16cc5d553850

2018-07-24

2018-07-24 03:24:55

https://medium.com/s/story/heteronyms-tulpas-and-the-end-of-monolithic-identities-16cc5d553850

false

1,633

null

Philosophy

philosophy

Philosophy

39,496

Anastasis Germanidis

Engineering metaphors. Currently @ITP_NYU, ex @IBMResearch, @sfpc.

18de787450f0

anastasis

355

514

20,181,104

null

0

null

0

30d8015e1b84

2018-09-14

2018-09-14 15:01:48

2017-12-13

2017-12-13 08:00:00

1

false

en

2018-10-29

2018-10-29 10:27:50

5

16cd2fb4681a

3.124528

0

One computer scientist explores how creating documentation and a demo for your code increases its accessibility.

5

The virtues of integrating documentation into your workflow How do you teach a computer to think like a human? How can you replicate what our neural networks do with code? Can a system learn to recognize images and distinguish voices the way our brains do? These are the questions Amirsina Torfi, computer science and electrical engineering doctoral student, is working to answer. “It’s the million dollar question,” Torfi says. “If we have a large amount of data, theoretically an infinite source of data, [then] deep learning is exactly equivalent to humans.” It starts much the same way you start as a baby: slowly learning lots of data, and putting pieces together to construct the world around you. “If we feed it with abundant data, the network, ideally, can learn to create stories, to classify between different subjects, between different categories,” Torfi explains. “So, basically, deep learning is learning with a neural network. It’s just an in-depth architecture, with numerous layers… just add the data, and that’s deep learning. As simple as that.” Of course, the execution isn’t as easy as the explanation. Torfi’s research centers around developing code to teach systems how to match, and learn from, audio and video channels. “The project was that [if] you have a video stream and an audio stream, can you tell if both of them have the same author or not,” Torfi says. “If my video has a 0.5-second delay compared to my audio, can we recognize that or not?” Torfi’s other projects have been similar code-wise, with different applications, including teaching systems how to hear multiple speakers and working with drone software on recognizing an image from different angles. Complex code, to be sure — enough so that Torfi realized if he wanted to share his work, he’d have to make matters visual. “I realized if anyone wanted to see the paper, it would be much easier for them to understand if they could implement the code,” Torfi says. “People are good at visual things.” But proper documentation has taken a backseat to showcasing complicated code in many cases, which to Torfi, limits the true accessibility and longevity of the work — even for the programmer who writes it. Demos and comprehensive documentation, he found, were key. “The first time I released code on GitHub without any demo, it got maybe one star after a month,” Torfi says. But after adding a demo? “My repository became the trending repository of the day. The only thing that changed was the demo!” In the field of computer science, code sharing feels natural. But proper documentation has taken a backseat to showcasing complicated code in many cases, which to Torfi, limits the true accessibility and longevity of the work — even for the programmer who writes it. “When I do detailed documentation, then anytime in the future when I go back to my code, I can easily realize what was happening, what can I reproduce, what can I use again,” Torfi says. “Recently I created protocols for myself for documenting… A one-page list to make a project open-source. So I just tick every one of them: Did I make automatic testing or not? Did I make contributions part or not? Did I make corrections or not? Did I make a demo or not?” “Sharing is a requirement,” Torfi states, matter-of-factly. “Whatever is not shared is forgotten someday.” A humble scientist, Torfi takes care to stress that many of the programmers whose code he sees on sites like GitHub are better than him (though he was pleased to be ranked by Git-Awards as one of the top 100 Python developers in the United States). What distinguishes his code from the rest, Torfi says, is his dutiful diligence — and passion — for explaining his work. “When I create the document, I’m not doing it just for myself — that’s why I become more motivated in creating it. When someone sees my documentation, if it’s good, they might have a blessing for me,” he adds, laughing. It’s no surprise, then, that when asked what his favorite part of his work is, Torfi’s answer is, without hesitation: documentation. It’s about more than “showing off,” he emphasizes. While making code open-source and easily reproducible certainly helps young scientists land jobs and grants, sharing enriches the community as a whole, adding to the industry so that new researchers can absorb and grow in turn. It forces the programmer to clarify, to change, and to learn constantly. “Sharing is a requirement,” Torfi states, matter-of-factly. “Whatever is not shared is forgotten someday.” Originally published at codeocean.com.

The virtues of integrating documentation into your workflow

0

the-virtues-of-integrating-documentation-into-your-workflow-16cd2fb4681a

2018-10-29

2018-10-29 10:27:50

https://medium.com/s/story/the-virtues-of-integrating-documentation-into-your-workflow-16cd2fb4681a

false

775

The Code Ocean blog sheds light on the researchers behind scientific code, reproducibility best practices, and resources to help you in your academic career

null

codeoceanhq

null

Code Ocean

contact@codeocean.com

codeocean

REPRODUCIBILITY,ACADEMIC PUBLISHING,OPEN SCIENCE,OPEN DATA,SCHOLARLY PUBLISHING

CodeOceanHQ

Machine Learning

machine-learning

Machine Learning

51,320

Code Ocean

codeocean.com

3cf382cd5c3e

CodeOcean

6

2

20,181,104

null

0

null

0

929eec0f81d7

2018-03-06

2018-03-06 12:47:42

2018-03-06

2018-03-06 12:52:05

1

false

en

2018-03-06

2018-03-06 12:52:05

5

16cef9d281f6

0.543396

1

0

This blog post highlights how AI tools, such as TensorFlow and Keras, can help insurers to automate damage assessment and avoid…

5

Automotive Insurance with TensorFlow: Estimating Damage and Repair Costs This blog post highlights how AI tools, such as TensorFlow and Keras, can help insurers to automate damage assessment and avoid overcharging by a car parts supplier. Read the full article on our blog: Automotive Insurance with TensorFlow: Estimating Damage / Repair Costs - Cloud Foundry Live |… Learn how AI tools, such as TensorFlow and Keras, can help insurers to automate damage assessment and avoid…bit.ly Stay in touch with the latest Altoros’ updates, subscribe to our social accounts: Twitter, Facebook, LinkedIn, Reddit.

Automotive Insurance with TensorFlow: Estimating Damage and Repair Costs

1

automotive-insurance-with-tensorflow-estimating-damage-and-repair-costs-16cef9d281f6

2018-03-06

2018-03-06 17:59:41

https://medium.com/s/story/automotive-insurance-with-tensorflow-estimating-damage-and-repair-costs-16cef9d281f6

false

91

Intersection of the Cloud Foundry PaaS, DevOps, IoT/IIoT, blockchain, and multi-cloud deployment automation.

null

Altoros

null

Altoros Blog

null

altoros-blog

BLOCKCHAIN,MACHINE LEARNING,CLOUD FOUNDRY,DIGITAL TRANSFORMATION,KUBERNETES

Altoros

Insurance

insurance

Insurance

13,823

Altoros

Altoros provides consulting and managed services with Cloud Foundry PaaS, multi-cloud deployment automation, and complex Java/.NET/Ruby architectures

347aa4cf1db

Altoros

1,058

890

20,181,104

null

0

null

0

null

2018-03-03

2018-03-03 13:16:50

2018-03-03

2018-03-03 13:35:22

1

false

en

2018-03-03

2018-03-03 13:35:22

9

16cf4ca14e54

1.29434

0

RiseML benchmarked Google’s TPUv2, and — surprise! — they’re fast. Also — surprise! — they’re cost efficient too. I know, shocking, right…

5

TPUv2 Benchmarks (Yup, they’re fast) RiseML benchmarked Google’s TPUv2, and — surprise! — they’re fast. Also — surprise! — they’re cost efficient too. I know, shocking, right? After all, if you consider selection-bias, of course they wouldn’t release something that wasn’t fast, and of course they’d price it so that it would be cost efficient. Then again, this wouldn’t be the first time a company has done something stupid, so, hey, we should probably keep that in mind 😇. Anyhow, it turns out that the TPUv2 is around 7.3x faster than an Nvidia P100, and around 2.8x faster than an Nvidia V100. Yes, it’s more expensive, but you actually end up saving on a $ per hour basis — it’s that much faster. /via https://blog.riseml.com/benchmarking-googles-new-tpuv2-121c03b71384 It’s important to note that the TPUv2 is not just a speeded up version of the TPUv1 — it’s quite a different beast. Unlike the TPUv1, which was focused on inference, the TPUv2 is actually focused on speeding up learning. Inside, it has systolic arrays (•) which do 180TFLOPS of matrix math, with 64GB of memory available — against the Nvidia V100’s 125 TFLOPS / 16GB. Mind you, the whole thing is “cloud-based”. You don’t get a TPUv2 enabled instance, instead, you are assigned a TPUv2 that exists somewhere in the “cloud”. (•) Systolic Array: Parallel input data flows through a network of hard-wired processor nodes, which combine, process, merge or sort the input data into a derived result. Because the wave-like propagation of data through a systolic array resembles the pulse of the human circulatory system, the name systolic was coined from medical terminology. The name is derived from systole as an analogy to the regular pumping of blood by the heart.

TPUv2 Benchmarks (Yup, they’re fast)

0

tpuv2-benchmarks-yup-theyre-fast-16cf4ca14e54

2018-03-03

2018-03-03 13:35:23

https://medium.com/s/story/tpuv2-benchmarks-yup-theyre-fast-16cf4ca14e54

false

290

null

Deep Learning

deep-learning

Deep Learning

12,189

Mahesh Paolini-Subramanya

That Tall Bald Indian Guy…

bd8dbcc39636

dieswaytoofast

110

48

20,181,104

null

0

data(iris) head(iris, 4) tail(iris) dim(iris) names(iris) attributes(iris) summary(iris) iris[1:5,] iris[,1:4] iris[1:10, "Sepal.Length"] plot(iris2) sepal_length<-iris2$sepal.length hist(sepal_length) hist(sepal_length, main="Histogram of Sepal Length", xlab="Sepal Length", xlim=c(4,8), col="blue", freq=FALSE) sepal_width<-iris2$sepal.width hist(sepal_width, main="Histogram of Sepal Width", xlab="Sepal Width", xlim=c(2,5), col="darkorchid", freq=FALSE) irisVer <- subset(iris, Species == "versicolor") irisSet <- subset(iris, Species == "setosa") irisVir <- subset(iris, Species == "virginica") par(mfrow=c(1,3),mar=c(6,3,2,1)) boxplot(irisVer[,1:4], main="Versicolor, Rainbow Palette",ylim = c(0,8),las=2, col=rainbow(4)) boxplot(irisSet[,1:4], main="Setosa, Heat color Palette",ylim = c(0,8),las=2, col=heat.colors(4)) boxplot(irisVir[,1:4], main="Virginica, Topo colors Palette",ylim = c(0,8),las=2, col=topo.colors(4))

15

null

2018-08-12

2018-08-12 19:45:58

2018-08-30

2018-08-30 01:35:16

6

false

en

2018-08-30

2018-08-30 01:35:48

7

16d1987f9edd

2.44434

2

0

In this short tutorial, I will show up the main functions you can run up to get a first glimpse of your dataset, in this case, the iris…

5

R for Newbies: Explore the Iris dataset with R In this short tutorial, I will show up the main functions you can run up to get a first glimpse of your dataset, in this case, the iris dataset. The dataset Particularly, this dataset is in R If you want to take a glimpse at the first 4 lines of rows. Optionally you may want to visualize the last rows of your dataset The dimensions of the dataframe The names of the columns The attributes of the dataframe Finally, if you want the descriptive statistics summary Indexing the first 5 rows Indexing the first 4 columns If you want to explore the first 10 rows of a particular column, in this case, Sepal length Basic Visualizations with Base R The plot () function is the generic function for plotting R objects. An exploratory plot array for iris dataset Histogram is basically a plot that breaks the data into bins (or breaks) and shows frequency distribution of these bins. You can change the breaks also and see the effect it has data visualization in terms of understandability (1). Histogram with hist() function Histogram of Sepal Length in iris2 dataset If we add more information in the hist() function, we can change some default parameters. Histogram of Sepal Lenght, with hist() function Histogram with 20 breaks In the following image we can observe how to change the default parameters, in the hist() function (2). Box Plot shows 5 statistically significant numbers- the minimum, the 25th percentile, the median, the 75th percentile and the maximum. It is thus useful for visualizing the spread of the data is and deriving inferences accordingly (1). Boxplots with boxplot() function. The boxplot() function takes in any number of numeric vectors, drawing a boxplot for each vector. You can also pass in a list (or data frame) with numeric vectors as its components (3). Boxplot of three different colors palette. The R code of the graph can be consulted (4). References Comprehensive guide to Data Visualization in R. http://bit.ly/2wnVjqY R Base Graphics. An idiot’s guide. http://bit.ly/2wqoV6L R Box Plot. http://bit.ly/2wnsRoY Exploratory Data Analysis Iris Dataset http://bit.ly/2wpZwu2

R for Newbies: Explore the Iris dataset with R

31

r-for-newbies-explore-the-iris-dataset-with-r-16d1987f9edd

2018-08-30

2018-08-30 01:35:48

https://medium.com/s/story/r-for-newbies-explore-the-iris-dataset-with-r-16d1987f9edd

false

396

null

Data Science

data-science

Data Science

33,617

Learning R and Machine Learning

Chemistry PhD living in a data-driven world.

cbe34340e508

data_datum

130

322

20,181,104

null

0

null

0

null

2018-07-25

2018-07-25 23:37:57

2018-07-25

2018-07-25 23:40:30

1

false

en

2018-07-25

2018-07-25 23:40:30

4

16d58dcf2f03

2.732075

0

Advances in cognitive technologies are making it possible to provide more accurate and relevant automated dialogues, giving rise to an…

5

The Conversational Enterprise Advances in cognitive technologies are making it possible to provide more accurate and relevant automated dialogues, giving rise to an increased use of chatbots for enterprise and B2B applications By Farzin Shahidi, CEO, NextPlane Over the last year, several organizations have launched chatbots for all kinds of niche services — from software applications that engage in natural-language dialogue with users — to provide basic customer service or send our calendar invites. And now, chatbots are completely replacing live agents in call centers or provide interactive assistance to customers visiting websites or using mobile apps. They can answer questions about products and services, schedule appointments, direct users to additional resources, and perform many other tasks. Looking back the concept of digital assistants has always been available on our mobile devices — specifically smartphones and smartwatches with voice commands — and have slowly changed the way end users consume information or accomplish tasks. That technology, artificial intelligence, is nothing new, of course. Computers and machines solving problems and responding to their human counterparts has been going on for decades, and has been recognized as a trusted method of interaction in many critical workflows. But now, as technology becomes the landscape, rather than just a piece of it, the use of AI and machine learning in the enterprise is growing steadily. Additionally, there are already some real-world use cases of AI progressing how we accomplish work in the enterprise, specifically through our mobile devices. Specifically, recent improvements in speech and language processing technologies are making chatbots more capable and prevalent than ever. For example, advances in speech recognition software are helping reduce word error rates; likewise, machine translation — the use of software to translate text or speech from one language to another — has improved exceptionally from the early Siri days. The Dawn of Chatbots In the Enterprise Chatbots have several potential benefits over traditional modes of communication. First, they can simplify applications for users. For example, rather than navigating through an interface or website to find information, users can just say or type what they want. Users can also compress multistep tasks into a single command, such as, “Get my list of open opportunities this quarter, and send it to Tom.” Second, the conversational UIs that chatbots offer may require little to no training, given that they understand and can interpret natural language and translate it into actions. Third, users can leverage chatbots to operate several business applications at once. For example, users can activate multiple chatbot actions in conversation with team members at the same time either with individual users or in a group messaging session. Combined, those benefits allow for non-expert users to interact with many complex applications in an intuitive fashion from a single interface. This gives rise to powerful automation opportunities, where chatbots trigger actions and orchestrate processes across a range of applications through the course of dialogue in natural language. The business impacts can include reducing costs by increasing self-service, improving end-user experience and satisfaction, delivering relevant information faster, and increasing compliance with internal procedures. Overall, enabling humans to communicate with systems and vice versa with low friction and natural interactions using message-based interfaces will only increase our dependency on chatbots. Join the Conversation The Economist sees bots as the next frontier, TechCrunch announces that we should forget apps as bots are taking over, many commenters see them as a keystone technology for the IT industry going forward, underlining the obvious trend that we are approaching a massive shift towards the AI-enablement of enterprise applications. Once business applications are enabled with a conversational UI for simple tasks with predefined rules, the next leap will be the introduction of autonomous features where bots and assistants comprehend the holistic context of a user and a certain business scenario. Last but not least, enterprise systems will be augmented with cognitive abilities to bring the amplification of knowledgeable workers to a whole new level. Follow NextPlane on Twitter for more updates.

The Conversational Enterprise

0

the-conversational-enterprise-16d58dcf2f03

2018-07-25

2018-07-25 23:40:31

https://medium.com/s/story/the-conversational-enterprise-16d58dcf2f03

false

671

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Nextplane Inc.

Communication.Unbound.

ab07e6ebf146

NextPlane

125

141

20,181,104

null

0

null

0

null

2018-03-05

2018-03-05 01:00:03

2018-03-05

2018-03-05 07:15:09

1

false

en

2018-03-05

2018-03-05 07:15:09

0

16d5b6cefde9

2.720755

10

1

0

Real Estate industry has admittedly, been conservative when it came to adopting new technology. But now with exponential technologies…

5

Property as a Service — Changing phase of Real Estate Industry Real Estate industry has admittedly, been conservative when it came to adopting new technology. But now with exponential technologies converging, the way we design, build & inhabit everything is changing. We are no longer worrying about architectural constraints, whatever the dream may be, it is possible to construct. There is barely any aspect of real estate unaffected by this storm. Design and Construction are currently being explored extensively for opportunities to make it better. We are using AI-aided Building Information Model (BIM) that enables blueprints to learn and adapt to changing ground conditions, weather, equipment and even new design ideas. We have introduced robots into construction and may be, looking into a future where they autonomously create the entire building. Sounds convenient, doesn’t it? In reality however, we are aiming even higher — using 3D & 4D printing to print entire structure. Chinese company WinSun Design Engineering Co., printed 10 houses from recycled materials in 24 hours at a cost of about $4,800 each. 4D printing will one day make such structures a reality. Born out of MIT’s Self-Assembly Lab, 4D printing involves 3D printed objects that can reshape and even self-assemble over time. There is more good news. We are also, no longer depended on construction materials of nature. Thanks to UCLA researchers at CO2NCRETE we could combine world’s greatest waste products (greenhouse gas (GHG) emissions) with lime to create a new type of cement. Not just that, Nano and micro-materials are steering in a new era of smart, super-strong and self-charging buildings. Furthermore, Engineers at Delft University have developed bio-concrete that can repair its own cracks aka self-healing concrete. Safe to say construction will never be the same! But is it all that we have to look forward to? Fortunately, no! Virtual Reality will modify real estate market into one extraordinary shopping experience. We don’t have to drive from one location to another to find a dream home, no more traditional reliance on listing photos and in-person showings. I can view the house in a different city in the luxury of my home at a time of my choice. All this vastness, comes with its own set of challenges. Unemployment being the biggest one on account of increasing reliance on automation/robots. We will need to look into ways to improve the skill of the employee as in the future only people with high skill value will be employable. We do not want to be left behind in this revolution, but the change in economy that it will bring still needs to be addressed. Commercially, the industry is looking to reinvent itself to create value in the market place. It is all about combining the right opportunity with the right type of location. As faster delivery to end customers becomes a focus, e-Commerce giants are looking for warehouse locations which enjoy proximity to the customers themselves. This affects real estate market, as the demand for the location is higher than the land resources available. On the side of spectrum, with data gradually moving to cloud, companies do not feel a need to have a ground infrastructure for storage. Other companies are looking to streamline their costs. Remote working is an increasingly popular concept. For Health insurer Aetna 31% of employers work remotely. Shared spaces are common where companies lack capital or want to divest themselves of the real estate, furniture, services, etc. that were previously non-negotiable. As the real-estate industry prepares for smart cities & mobility in an extremely demanding ecosystem, we are witnessing unique opportunities created on a case to case basis and current practices moving to obsolete. Only the most agile & innovative work environments will be able to survive it. The big question now, is not just how updated you are with the current real–estate shift, but also, how far ahead can you anticipate the effects of this exponential tech convergence?

Property as a Service — Changing phase of Real Estate Industry

214

property-as-a-service-changing-phase-of-real-estate-industry-16d5b6cefde9

2018-03-15

2018-03-15 17:51:11

https://medium.com/s/story/property-as-a-service-changing-phase-of-real-estate-industry-16d5b6cefde9

false

668

null

Real Estate

real-estate

Real Estate

69,162

Sheersha Kandwal

I think to Live!

7e288547177e

sheersha.kandwal

11

10

20,181,104

null

0

null

0

null

2018-07-11

2018-07-11 13:38:23

2018-07-12

2018-07-12 07:59:42

5

false

en

2018-07-12

2018-07-12 14:29:24

5

16d7a3edcab

2.803145

8

0

Whether you’re an experienced startup investor or someone deciding how to start, at the end of the day — you need to ask yourself and…

5

What Makes Value Protocol the Best Authenticity Solution for Physical Assets on the Blockchain? Whether you’re an experienced startup investor or someone deciding how to start, at the end of the day — you need to ask yourself and answer the same questions. Let’s say you’re already convinced about a project. It has strong and achievable ideas with an experienced team to make it work — quite simply, everything looks great. You’re ready to invest, but then wait, you’re stooped to find that there are 15 other projects working towards the same goal. Well, what then? Value Protocol faced the same question. For the past 3 years, we’ve been building a solution to finally secure items to the blockchain with A.I., as well with developing an authenticity protocol and infrastructure for businesses with physical assets. But then we found out that we’re not alone. Normally this phrase brings comfort, but this time we wanted to know who we’re competing with. What are these other projects and how far do they go to push the same mission? So of course we dove deep and researched these other initiatives to see how they address the huge gaps in the market — regarding the authenticity of physical items on the blockchain. The truth is, no current project was able to provide a secure solution and address the critical problems. When it comes to evaluating projects, there are three points to consider – The business (strategy, team, experience); The architecture and the technology itself; The solid core of the business and the value they can really bring. We’re aware that the available info out there can be misleading, as we can only draw careful conclusions. But in some projects, we noticed potentially significant problems with the team and targeting. And it seems that most of the strengths came down to marketing — which unfortunately plays a huge role with ICOs. To our surprise, some projects that already finished their ICO turned out having no real solution — and were STILL able to raise $13M. Others were stronger and weaker at different points, but only about half of the main issues were addressed. Unless other projects make a big leap to address all the critical issues, sooner or later they will hit a wall. Right now, the problem is it’s way too easy to replace items with a fake. We can’t deny that developing systems for securely linking items to the blockchain is a pretty hot topic. But after doing our research, we’re confident in our solution. Value Protocol’s design is bulletproof and more complete than any current and existing project on the market. Without a secure way to link the physical item to its blockchain entry during each step, it leaves room for fraud and can be replaced at any time. Value Protocol uses computer vision and machine learning to fingerprint any physical item’s unique surface structure and hash it to the blockchain –– a tamper-proof solution to finally eliminate space for counterfeits and fraud at any stage. Stay updated by following our Facebook and Twitter accounts. Got some questions? Talk to us at team@valueprotocol.org.

What Makes Value Protocol the Best Authenticity Solution for Physical Assets on the Blockchain?

289

what-makes-value-protocol-the-best-authenticity-solution-for-physical-assets-on-the-blockchain-16d7a3edcab

2018-07-12

2018-07-12 14:29:24

https://medium.com/s/story/what-makes-value-protocol-the-best-authenticity-solution-for-physical-assets-on-the-blockchain-16d7a3edcab

false

522

null

Blockchain

blockchain

Blockchain

265,164

Veracity Protocol

The decentralized infrastructure for the lifecycle of things (LoT) to secure end-to-end traceability, data veracity and resource efficiency in supply chains.

ffa0d24088f8

veracityprotocl

18

5

20,181,104

null

0

null

0

null

2018-04-22

2018-04-22 22:50:39

2018-04-22

2018-04-22 22:55:03

0

false

en

2018-04-22

2018-04-22 22:55:03

1

16d84b6918d2

0.113208

0

Ride your curiosity in Croatia.

5

Curious Croatia Ride your curiosity in Croatia. Curiosio. Geek travel in Croatia. Most curious roadtrips ever. You start & finish anywhere, and get unique sightseeing route, within desired days &…curiosio.com

Curious Croatia

0

curious-croatia-16d84b6918d2

2018-04-22

2018-04-22 22:55:04

https://medium.com/s/story/curious-croatia-16d84b6918d2

false

30

null

Croatia

croatia

Croatia

1,069

Curiosio

geek travel

64651106c6d2

curiosio

8

5

20,181,104

null

0

null

0

fcacddcd4b87

2018-03-06

2018-03-06 16:50:00

2018-03-06

2018-03-06 17:07:59

1

false

en

2018-03-06

2018-03-06 17:07:59

6

16d85f86c0f

3.015094

2

0

One of my latest discoveries as a source of information is The Journal of Design and Science (JoDS), a joint venture between MIT Media Lab…

5

How To Become A Centaur One of my latest discoveries as a source of information is The Journal of Design and Science (JoDS), a joint venture between MIT Media Lab and MIT Press that is aimed at readers with open, curious minds, exploring timely, controversial topics in science, design, and society with a particular focus on the nuanced interactions among them. And in this marvelous site I discovered this spectacular article by Nicky Case: How To Become A Centaur. It’s worthwhile to read it because it makes a journey of what Artificial Intelligence and Augmented Intelligence has been and how the future is the symbiosis between the two, the so-called Centaurs. Here are some key ideas of the article, but I assure you that it’s worth reading it. The more far-fetched concerns about AI (taking over the world and turning us all into pets and/or paperclips), it all comes from the same root fear: the fear that AI will not share our human goals and values. And what’s worse, we’ve told ourselves that our relationship between ourselves and our AI is like a chess game: Zero-sum — one player’s win is another player’s loss. Similar to how the mythological centaur was half-human, half-horse, these centaurs were teams that were half-human, half-AI. Not surprisingly, a Human+AI Centaur beats the solo human. But — amazingly — a Human+AI Centaur also beats the solo computer. (…) The “g factor”, also known as “general intelligence”, only accounts for 30–50% of an individual’s performance on different cognitive tasks. So while it is an important dimension, it’s not the only dimension. (…) Because humans & AIs are strong on different dimensions, together, as a centaur, they can beat out solo humans and computers alike. However, consider the “No Free Lunch” theorem, which comes from the field of machine learning itself. The theorem states that no problem-solving algorithm (or “intelligence”) can out-do random chance on all possible problems: instead, an intelligence has to specialize. (…) This may be a hopeful sign: even humans will continue to outsmart computers on some dimensions. (…) IA: Intelligence Augmentation. The old story of AI is about human brains working against silicon brains. The new story of IA will be about human brains working with silicon brains. As it turns out, most of the world is the opposite of a chess game: Non-zero-sum — both players can win. Doug Engelbart tied a brick to a pencil — to prove a point. Of all the tools we’ve created to augment our intelligence, writing may be the most important. But when he “de-augmented” the pencil, by tying a brick to it, it became much, much harder to even write a single word. And when you make it hard to do the low-level parts of writing, it becomes near impossible to do the higher-level parts of writing: organizing your thoughts, exploring new ideas and expressions, cutting it all down to what’s essential. That was Doug’s message: a tool doesn’t “just” make something easier — it allows for new, previously-impossible ways of thinking, of living, of being. Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process. When you create a Human+AI team, the hard part isn’t the “AI”. It isn’t even the “Human”. It’s the “+”. (…) AIs are best at choosing answers. Humans are best at choosing questions. (…) So, when you think of augmenting human intelligence with AI, think less of assimilating into The Borg, and more of a spirited conversation between Kirk & Spock — a mix of intuition and logic that surpasses either one alone. Symbiosis shows us you can have fruitful collaborations even if you have different skills, or different goals, or are even different species. Symbiosis shows us that the world often isn’t zero-sum — it doesn’t have to be humans versus AI, or humans versus centaurs, or humans versus other humans. Symbiosis is two individuals succeeding together not despite, but because of, their differences. Symbiosis is the “+”. The key idea, in addition to the Centaur concept, is the AIA: Artificial Intelligence Augmentation. Something that we have already been talking about in the article of Maurice Conti or in the one about Ken Goldberg and his multiplicity. I firmly believe in this idea. And you? #365daysof #futurism #innovation #technology #science #day63

How To Become A Centaur

2

how-to-become-a-centaur-16d85f86c0f

2018-06-11

2018-06-11 14:56:11

https://medium.com/s/story/how-to-become-a-centaur-16d85f86c0f

false

746

High quality curated content and topics related to innovation and futurism along with a little reflection

null

Future Today

alayon.david@gmail.com

future-today

TECHNOLOGY,FUTURISM,INNOVATION,SCIENCE,TRANSHUMANISM

davidalayon

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

David Alayón

Head of Innovation Projects @ Inditex · Founder @Innuba_es @Mindset_tech @GuudTV · Professor @IEBSchool @DICeducacion · Mentor/Investor @ConectorSpain AngelClub

91f2d81ac9db

davidalayon

161

72

20,181,104

null

0

null

0

b7bbc4597459

2018-01-24

2018-01-24 18:19:17

2018-02-01

2018-02-01 16:08:52

2

false

en

2018-02-01

2018-02-01 16:08:52

12

16d862ec4ed5

2.436164

38

0

Deep Learning best practices, Galileo’s Telescope, OpenAI’s newly-released unsolved problems, and more!

5

This Week in AI, February 1, 2018 Deep Learning best practices, Galileo’s Telescope, OpenAI’s newly-released unsolved problems, and more! 8 Deep Learning Best Practices Here’s a really great list of best practices for improving your deep learning models. Of particular interest is the bit on learning rate optimization and stochastic gradient descent restarts. Both of these methods are generally applicable and are likely to improve models you’re working on right now. Also interesting, the list includes examples using the fastai package, a high-level framework developed by Fast.ai on top of PyTorch. What We Can Learn About Deep Learning from Galileo’s Telescope The moon as seen through Galileo’s telescope Ali Rahimi wrote a great article comparing the current state of deep learning with the understanding of optics when Galileo built his telescope 400 years ago. Back then, scientists had general ideas of how light bent through glass, but didn’t understand the principles well enough to make true optical devices. Over time as we learned more about optics, we were able to develop mental models that allow us to construct complex stacks of lenses. We don’t have these mental models for deep learning yet, and for the most part, experts don’t understand why deep learning works so well. As more research is done in this field, we should start seeing better abstractions and language for describing what each part of a deep learning models is doing. Serving a Deep Learning Model with Flask You’ve trained your network but how do you let other people actually use it? A great first option is to build a REST API with a simple package like Flask (one of my favorite Python packages). This way your model is available for training and making predictions using URLs and common HTTP methods. Learn how to serve your model with this great tutorial using Keras and Flask. OpenAI Wants Your Help OpenAI has released a new set of unsolved problems in deep learning. These problems span several domains such as using reinforcement learning to learn the classic Snake game, and training an autoencoder to generate new data for augmenting datasets. This looks like a great set of problems for deep learning beginners—and experienced practitioners—to tackle. Music Videos Generated By AI Here’s a really cool project by Jeff Zito where video footage was generated by a deep learning model based on music from Lord Over. The videos can be somewhat disturbing, but I love seeing people making art with deep learning. Zito’s model is based on a deep learning model used to generate realistic videos from audio alone. It’s truly impressive (and somewhat worrying) how well the mouth movements match the audio. ~ Stay tuned for new updates as we continue to review all that’s new in the world of AI! And if you’re interested in mastering these transformational skills, and building a rewarding career in this amazing space, consider one of our Nanodegree programs: Deep Learning Nanodegree | Udacity Break into artificial intelligence (AI) with our Deep Learning Nanodegree. Discover deep learning foundations, neural…www.udacity.com Artificial Intelligence Engineer Welcome to the Artificial Intelligence Nanodegree program, where virtually anyone on the planet can study to become an…www.udacity.com Machine Learning Nanodegree | Udacity Learn machine learning, and become a machine learning engineer! Learn to apply predictive models, use massive data sets…www.udacity.com

This Week in AI, February 1, 2018

152

this-week-in-ai-february-1-2018-16d862ec4ed5

2018-03-30

2018-03-30 20:27:55

https://medium.com/s/story/this-week-in-ai-february-1-2018-16d862ec4ed5

false

544

Learning for the Jobs of Today, Tomorrow, and Beyond

null

udacity

null

Udacity Inc

social@udacity.com

udacity

ONLINE EDUCATION,SKILLS DEVELOPMENT,CAREER PATHS,EDUCATION,JOBS

udacity

Machine Learning

machine-learning

Machine Learning

51,320

Mat Leonard

Teaching all things machine learning and AI at Udacity. Loves cats. @MatDrinksTea on Twitter

25d4dd5fc9c4

mat_leonard_u

634

2

20,181,104

null

0

null

0

cd627581b58e

2018-07-13

2018-07-13 14:09:24

2018-07-16

2018-07-16 14:28:25

1

false

en

2018-07-16

2018-07-16 15:29:59

2

16d881bdbdc2

1.883019

1

0

Here at Context Scout, we’re continuing our “Meet the Team” series with one of the recent additions to our Research team at Context Scout…

5

Context Scout: Meet the Team — Mohammad Akbari Here at Context Scout, we’re continuing our “Meet the Team” series with one of the recent additions to our Research team at Context Scout, Mohammad Akbari PhD. What’s your role on the team? I am a Data Science researcher in Context Scout and senior research associate at University College London (UCL) working on information retrieval and data mining. Where are you from? What made you come to London/UK? I am from Tehran, Iran but lived in Singapore for last 6 years. I relocated to London due to career goals. London includes several well-known educational and research institutes such as UCL and Imperial College. They are closely working with industrial firms to advance technology and research. Apart from that, London is well-known for it’s entrepreneurial opportunities, and this makes it best for researchers. Why did you choose to join Context Scout? I am a researcher that craves opportunities to solve complex challenges. Meanwhile I have an entrepreneur mindset. I am someone who cares about solving problems people face in their daily life. Context Scout and UCL permit me to achieve this by bringing intelligence to web search; an online activity each of us perform several times daily. What’s your favorite part of working at Context Scout? The entrepreneurial spirit at Context Scout encourages collaboration, diversity, and individuality. You indeed work, innovate, and play which gives the right fit to you and help you easily solve challenges. What’s the most interesting thing you’ve read lately? I’m just coming back from the International Conference on Machine Learning (ICML), so my recent readings were mostly on Machine Learning. In my opinion, one of the best readings of ICML is “Delayed Impact of Fair Machine Learning” which focuses on bias in machine learning. The paper demonstrates that even in a one-step feedback model, common fairness criteria in general do not promote improvement over time. In other words, it demonstrates how machine learning models may exhibit discriminatory biases based on sensitive characteristics, such as, gender, race, religion, physical ability, and sexual orientation, or perform less well for historically disadvantaged groups. What future technology excites you the most? I envision the future of healthcare technology where smart devices measure, estimate, guide, and influence our lifestyle and wellness, assisting us better understand and control our health. These devices assist us to better fight diseases that out a mystery for us now. What startup do you admire and why? Evernote for helping people organizing tasks, notes, ideas; essential tasks everyone may deal with it daily. — Want to see what the team is working? Head to contextscout.com to give our Chrome extension a try.

Context Scout: Meet the Team — Mohammad Akbari

5

context-scout-meet-the-team-mohammad-akbari-16d881bdbdc2

2018-07-26

2018-07-26 10:15:37

https://medium.com/s/story/context-scout-meet-the-team-mohammad-akbari-16d881bdbdc2

false

446

Thoughts on the future of web search from the Context Scout team

null

ContextScout

null

Context Scout

marc@contextscout.com

context-scout

SEARCH,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING,STARTUP,FUTURE OF WORK

contextScout

Machine Learning

machine-learning

Machine Learning

51,320

Emily Sappington

VP of Product at Context Scout. UX Designer, Researcher, and generally inquisitive mind.

e03413134482

emilysappington

6

3

20,181,104

null

0

null

0

null

2018-04-27

2018-04-27 16:24:18

2018-04-27

2018-04-27 16:34:15

2

false

en

2018-04-27

2018-04-27 16:34:15

1

16d8e8d377ce

2.628616

1

0

At Kira Talent, we’re heading rapidly down a path of releasing the industry’s first, truly holistic admissions management software. We’re…

5

The Brave New World of Machine Learning in Higher Education: Machine Learning 101 At Kira Talent, we’re heading rapidly down a path of releasing the industry’s first, truly holistic admissions management software. We’re embracing machine learning (ML) and we’re dead set on making higher education admissions “smarter.” Okay I know that was a ton of buzzwords…wait did I mention AI and blockchain? Just kidding. Bottom line, Kira Talent is bringing ML to higher education, an industry that is craving transformation. But before I delve into our 2018 plans in our next article, let’s start with Machine Learning 101 and what that even means: Machine learning and AI are tossed around in the tech space like “cloud-based” and “mobile-first” were being used less than a decade ago. When we hear these terms, we might think of the Terminator, Blade Runner, or the Matrix. We might have images of automation, our jobs being replaced or robots that roam around freely in the streets. But the reality is far less scary. There is a difference between machine learning and artificial general intelligence (AGI) AKA the robots that roam around freely performing any intellectual task that a human can do. We are many years if not decades away from that. However, whether you know it or not, your current day-to-day life is already augmented by machine learning. You just might not see it or know where to look. Think of how Amazon seems to predict what you want to buy. Think of how Facebook can already predict which of your friends or family is in a photo and don’t even get me started on Netflix. When fed with an incredible amount of data about you, or people like you, machine learning algorithms begin to identify patterns through actions and make predictions on future actions. Data scientists and software engineers develop these machine learning algorithms and then can feed their algorithms all of this data to process. Software programs you use every day can be “trained” or “taught” to new ways to present and analyze data — much like methods, you might use as part of a team when sorting all the information you have on hand before coming to a conclusion. The insights machine learning serves up help superpower the solutions we rely on with predictive knowledge. Hence, the term “machine learning.” Source: Reddit Next, let’s address this other word we hear (and say) a lot, smarter: What’s most incredible about machine learning algorithms is that they never stop learning. Constantly evaluating patterns in the data they’ve been fed and re-assessing the prior insights they’ve served up based on outcomes, they become “smarter” over time and (unlike many of us) they learn from their mistakes and improve accordingly. An everyday example of feedback loops that machine learning is already powering is the predictive text feature of our smartphones. When you first get a new device, you may fumble with the touchpad, having to teach it the names of your friends or colleagues, school and other acronyms you use all the time… but, over time, you notice your phone generating phrases, terms, or names you often use in relation to the other words you’re typing. Safe to say, machine learning presents an extraordinary opportunity for every industry and most industries have already adopted new machine learning powered technologies. Now that you understand a bit more about ML, tune in for Part II to discover all the ways machine learning will transform higher education enrolment over the next 5–10 years with Kira Talent leading the charge.

The Brave New World of Machine Learning in Higher Education: Machine Learning 101

1

the-brave-new-world-of-machine-learning-in-higher-education-machine-learning-101-16d8e8d377ce

2018-05-06

2018-05-06 04:54:09

https://medium.com/s/story/the-brave-new-world-of-machine-learning-in-higher-education-machine-learning-101-16d8e8d377ce

false

595

null

Machine Learning

machine-learning

Machine Learning

51,320

Emilie Cushman

Founder and CEO of Kira Talent. www.kiratalent.com/

d2a23b4ff8b1

emiliecushman

0

4

20,181,104

null

0

null

0

cc716168a30a

2018-08-23

2018-08-23 05:39:48

2018-08-23

2018-08-23 05:46:57

0

false

en

2018-08-23

2018-08-23 05:46:57

1

16d9d710175a

3.845283

1

0

If you want to foray into the field of machine learning with the intent of solving real-world, mission-critical problems of your domain…

4

Machine Learning — Choosing Substance Over Style If you want to foray into the field of machine learning with the intent of solving real-world, mission-critical problems of your domain, and are caught in the dilemma of whether to start with dazzling deep learning models or old-fashioned decision trees and logistic regressions, here is a proverb in Hindi that should help you resolve your predicament: “haathi ke daant khane ke aur, dikhane ke aur.” Popular press releases on progress in the field of AI & Machine Learning are like the tusks of an elephant — impressive to look at — while the other hidden teeth do all the hard work required to keep an elephant alive, by helping it chew and eat food on a day to day basis. Even as the Googles and the Facebooks of the world show exciting demos of futuristic possibilities and potential abilities of deep learning models, it is the classical machine learning methods, with clever feature engineering techniques that often serve the business demands. The stylish deep learning neural models have their place in solving real-world applications. In fact, for some problems, especially involving vision and speech, there are no better models than the deep learning solutions. However, it is the less sexy machine learning techniques, if applied well, that are most ready today to form the substance of successful solutions to real-world problems. Of course, the catchphrase is “if applied well”. So, what does it take to succeed with classical machine learning techniques? Domain-Aware Feature Engineering Most practitioners will agree that it is clever feature engineering skills that can bring the best out of classical machine learning methods. This leads to a natural question: If such techniques require clever, creative feature engineering techniques, why not just use deep learning, whose key goal (among others) is to alleviate the need for hand-crafting of features? Firstly, deep learning models require a lot of data. Secondly, deep learning models are hard to interpret, which is often required in real world. While hand-crafting of features relies on your knowledge of the domain and the particular task at hand, incorporation of such knowledge, in turn, reduces the need for a large amount of training data. And simpler models are more interpretable too. Your domain knowledge is something that you may already be well versed with. Incorporating that appropriately, in practice, will yield double benefits. Lesser data and better results! Having said that, it is not so straightforward to translate raw data into a representation that is best suited for machine learning models to work with. You will need to equip yourself with the right kind of scientific knowledge and practical tricks of the trade to be able to achieve this. Equipping for Real-World Problems There are a bunch of tried and tested techniques that one needs to use to tame the data and turn them into a meaningful input form. To add to this there is a model-specific bag of tricks like regularisation, hyperparameter tuning, kernel tricks, etc. etc. (by the way, you can’t escape them even when you go to the deep learning domain). There are two key aspects that will help you discern these techniques and apply them confidently to real-world problems. One is scientific knowledge of the models and the second is awareness of problem-solving strategies that will yield excellent results. The Value of Scientific Knowledge (Bottom-Up Knowledge) Many tricks of the trade in getting machine learning algorithms to perform well may seem ad-hoc if you don’t understand the underlying machinery of the ML models you have chosen to work with. However, if you open the black box of machine learning models and understand how they mathematically achieve the learning objective, most of those tricks fall into the right slots of why and how they work. The scientific knowledge of working of the models will enable you to meaningfully engineer the features for the task at hand. More importantly, when the model doesn’t perform as expected, such theoretically well-founded knowledge will help you diagnose the problem and come up with corrections to your model or to the data. Without the help of scientific knowledge to guide you in these scenarios, you will be left shooting in the dark. Understanding the Problem Solving Strategies (Top-Down Knowledge) I have come across many engineers who have gathered a lot of theoretical knowledge, from textbooks and online courses, yet they find it hard to connect the dots between real-world scenarios and the machine learning formulations necessary to address them. It greatly helps if you study how a specific model has been applied successfully to a variety of problems consisting of different types of real-world data. Such knowledge of design patterns of machine learning will play an invaluable role in identifying and shortlisting a handful of strategies to solve a real-world problem, providing a degree of certainty to your potential success. Conclusion I began by making a case for mastering classical machine learning methods as they are more likely to be of use to you as a practitioner solving business problems. It is often easy to be blinded by the supernova explosion of popularity or hype of deep learning techniques. It is important to not lose sight of more useful, less trendy machine learning models, that still remain the workhorses of the industry. They are likely to remain so for years to come. Surely, deep learning techniques are definitely natural additions to your arsenal and for some problems they provide the best solutions today, but require a very large amount of training data. If you can make do with less data, by incorporating domain knowledge, why trade that for a more expensive technique that requires an order of magnitude more data? If you are convinced of the value of classical machine learning methods, I hope you place importance on acquiring both, the bottom-up (scientific) knowledge and the top-down (design patterns) on equal footing. As an experienced machine learning practitioner, a mentor and a teacher, I can’t stress this more.

Machine Learning — Choosing Substance Over Style

2

machine-learning-choosing-substance-over-style-16d9d710175a

2018-08-23

2018-08-23 05:46:57

https://medium.com/s/story/machine-learning-choosing-substance-over-style-16d9d710175a

false

1,019

AI environments to enable better learning

null

The School of AI

deepti@theschoolofai.com

the-school-of-ai

ARTIFICIAL INTELLIGENCE,MACHINE LEARNING,MATH,CORPORATE TRAINING,ACTIVE LEARNING

null

Machine Learning

machine-learning

Machine Learning

51,320

Ram Prakash H

AI Researcher. Creating AI Environments to help people learn better.

c526c6bfe5c0

rampi

36

33

20,181,104

null

0

null

0

null

2018-09-06

2018-09-06 21:11:10

2018-09-06

2018-09-06 21:20:55

1

false

en

2018-09-06

2018-09-06 21:24:34

0

16da0359caef

0.826415

0

Data Science is encompassing, broad, so broad that it sometimes seems ambiguous, making it very hard to learn.

1

10 practical steps to Data Science. Data Science is encompassing, broad, so broad that it sometimes seems ambiguous, making it very hard to learn. But I came across a list of what I believe are 10 Practical Steps to learning Data Science: 1. Programming a. Python - https://lnkd.in/gGQ7cuv b. R - https://lnkd.in/giMGbph c. SQL - https://lnkd.in/gM8nMNP d. Command Line - https://lnkd.in/e3EQuis 2. Stats/Prob/Math a. Coursera’s Statistics w/ R - https://lnkd.in/gGT9NEf b. edX’s Probability - https://lnkd.in/gpUyC3P c. Khan Academy Linear Algebra - https://lnkd.in/gMshbX4 3. Data Viz a. Python Matplotlib- https://lnkd.in/gr3ifNt b. R ggplot2 - https://lnkd.in/eThJXNr 4. Data Manipulation a. Python Pandas - https://lnkd.in/g9kfpX4 b. R dplyr - https://lnkd.in/gAWusih 5. Machine Learning a. Google Crash Course - https://lnkd.in/gSgkVcT b. Stanford Coursera - https://lnkd.in/g8ZG557 c. ISLR Book - https://lnkd.in/gk8GPZC 6. Experimental Design a. Udacity A/B Testing - https://lnkd.in/gCerh4f 7. Business Sense a. Metrics - https://lnkd.in/gZAG7bS 8. Communication a. Storytelling - https://lnkd.in/gwjxVUu 9. Profile Building a. GitHub - https://lnkd.in/g4r9naJ b. LinkedIn - https://lnkd.in/g-KHHEC c. Kaggle - https://lnkd.in/gBC77Hu d. DS Resume - https://lnkd.in/gU8WVAF 10. Job Search a. Daily Expert Tips & Advice - https://lnkd.in/g8z-xXD Hope this helps! 👍 Reference: http://www.claoudml.co/

10 practical steps to Data Science.

0

10-practical-steps-to-data-science-16da0359caef

2018-09-06

2018-09-06 21:24:34

https://medium.com/s/story/10-practical-steps-to-data-science-16da0359caef

false

166

null

Data Science

data-science

Data Science

33,617

OGUNLADE OLAWALE

null

517007f4126b

ogunladeolawaleo

0

4

20,181,104

null

0

null

0

null

2018-06-02

2018-06-02 04:46:23

2018-06-02

2018-06-02 04:48:37

0

false

en

2018-06-02

2018-06-02 04:48:37

23

16db0d5100ea

2.928302

1

0

Artivatic is an end to end AI Infrastructure platform built on deep-tech, machine learning technologies with in-depth analogy of genomic…

5

Artivatic.ai — 18 Months Journey: Self-Assessment Artivatic is an end to end AI Infrastructure platform built on deep-tech, machine learning technologies with in-depth analogy of genomic science, psychology & neuroscience. This helps large enterprises, startups & developer to build/integrate intelligent products & solutions without much development effort. It focuses on BFSI, Wealth Management & Healthcare sector. In short you can say that Artivatic’s platform is agnostic in nature to enable enterprises to have ‘Smart AI Brain’ as like humans to take all processes automated and self decisions in real time. Within 15 months of its inception, Artivatic has made significant growth in terms of team (25 full time member team), clients (other than PoC etc, there are 20+ leads without having any sales team), products, solutions and set benchmark already among the industry. Artivatic was founded by me (Layak SIngh) and Puneet Tandon who earlier founded few startups as well and worked for decade in various large enterprises. Currently, Artivatic is looking for US $1M fund-raise to expand its team in research, sales & delivery to support 100+ of clients in coming years and grow to achieve US $10M ARR in next 2 years. Artivatic is also open for Bridge round in between US $100–150K for immediate need. PRODUCT /SOLUTIONS [Some lists that Cater for BFSI & Healthcare only] Intelligent on-boarding for BFSI industry [Specific for Insurance, Banking & Wealth Management is done]- Available as SaaS, Pulg-in-play On-premise Integration Smart Insurance — Health, Life, Auto & Travel Intelligent Underwriting for credit, lending, insurance and wealth management Intelligent neuro-decisoning system with human assisted approach Digital Identity, Trust & Credibility intelligence system Consumer Genomics specific to BFSI, Healthcare industry Claims & Monitoring system Intelligent denial prediction & authorization system for insurance Intelligent NPA with multiple parameters, data sources Damage detection for vehicle claims processing Financial contract modelling system KYC Automation & Fraud detection system is developed for indian market as of now with OCR/NLP//ML/Computer Vision technologies: Scanned documents, Handwritten documental, banking documents and more will be processed automatically with higher accuracy. DATA RESEARCH & PARTNERSHIPS Artivatic has partnered with multiple health, data organizations for rich data to build intelligent solutions for BFSI, health & life insurance sectors. Also, partnered with acadmia’s for research & development. Patents/Businesses Artivatic has filed some patents already and in process to file more. Currently, is in process to have revenue by next quarter as last 18 months spent on technology, research & product building with rich data. Expecting to have good growth in coming quarters for business solutions and revenue. Some Mention/PR/News about Artivatic: https://yourstory.com/2017/08/artivatic-smart-ai-banks-financial-firms/ http://ml-india.org/companydetails/companydetail/artivatic.ai/ https://yourstory.com/2018/05/10-ai-startups-watch-2018/ https://analyticsindiamag.com/10-indian-startups-that-are-leading-the-ai-race-2018/ https://www.vccircle.com/meet-the-five-startups-selected-for-pitney-bowes-accelerator-programme/ https://www.entrepreneur.com/article/306284 https://yourstory.com/2018/03/saas-startups-india-cusp-huge-global-opportunity-manish-choudhary-pitney-bowes-startup-accelerator-program/ https://yourstory.com/2017/09/india-ai-artificial-intelligence-machine-learning/ https://www.sundayguardianlive.com/news/10824-be-ai-ready-workforce-needs-re-skilling-say-experts http://www.towards.ai/competing-ibm-watson-artivatic-new-generation-start-up-makes-b2b-ai-process-decision-making-solutions/ http://houseofbots.com/news-detail/2713-1-ai-race-2018-top-10-indian-startups https://www.sumhr.com/top-artificial-intelligence-companies-india/ https://www.aitoday.xyz/top-10-ai-companies-in-india/ https://www.disruptordaily.com/top-10-disruptive-companies-bangalore/ https://technodistrict.com/top-10-artificial-intelligence-startups-in-india/ http://www.tycoonstory.com/startup/best-startups-for-artificial-intelligence-in-india/ https://www.dr-hempel-network.com/digital-health-startups/10-innovative-ai-companies-in-india-targeting-digital-health-market/ https://hackernoon.com/massive-list-of-100-saas-companies-in-india-e96eb115cd2 https://yourstory.com/2018/05/saas-companies-gearing-gdpr/ https://www.insightssuccess.com/artivatic-integrating-technology-build-intelligent-applications-solutions/ Based on our 18 months journey, it is evident that we have grown significantly in the market as well as on product side but still executing for financial growth [Pretty significant in terms of engagements with clients and revenue side]. The coming quarters will be the masterstroke for Artivatic to build global business out of India with excellent financial trajectory. One thing is important in B2B that product takes time, you can not merely focus on traction if product is not good and can not scale. B2B startups faces lots of heat due to enough financials, need of infrastructure, talent and more.We are focusing on revenue generation at the moment to speed up our growth. We are narrowing down our offerings as well to be the domain expertise in the sector and then move further. We are currently looking for funding to scale and get support from industry leaders, investors for the same. If you want to be part of Artivatic’s growth do write at layak@artivatic.ai and happy to host you in our office for coffee and discuss about our journey, growth, current status and future plans.

Artivatic.ai — 18 Months Journey: Self-Assessment

2

artivatic-ai-18-months-journey-self-assessment-16db0d5100ea

2018-06-02

2018-06-02 05:10:13

https://medium.com/s/story/artivatic-ai-18-months-journey-self-assessment-16db0d5100ea

false

776

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Artivatic.ai

AI & Deeptech focused technology startup disrupting Fintech, Banking & Insurance sectors: Reducing Risk, Digital Access, Underwriting, Claims, Fraud & more.

4e9f5dc4e8ec

artivatic

15

70

20,181,104

null