Alaamer/medium-articles-posts-with-content

0

null

0

d3822484c236

2018-06-25

2018-06-25 01:26:21

2018-06-26

2018-06-26 00:01:53

10

false

en

2018-06-26

2018-06-26 00:01:53

20

175cd4e72191

4.868868

4

0

Last week I spoke at the first Responsible AI event in Melbourne. The idea behind the event is to bring together people from a range of…

4

Responsible AI: A Global Perspective Last week I spoke at the first Responsible AI event in Melbourne. The idea behind the event is to bring together people from a range of backgrounds to debate important issues associated with AI/Ml such as trust, transparency, fairness and ethics. With over 150 people registered and 100 turning up on the night I was blown away by the level of interest and passion for the topic. It was great to see people from such a broad range of backgrounds including philosophy, law, technology and ethics and the different perspectives they bring to the conversation. The aim of my talk was to highlight some of the fantastic work being done globally on the topic of Responsible AI and get the audience thinking about how we in Australia can contribute. Responsible AI Players can be broadly categorised as follows: Governments Public, private, academic partnerships Individual companies 1. Governments There are a number of governments announcing strategies and investment in AI. Of these I’ve identified three who have placed Responsible AI considerations such as ethics, trust, transparency and fairness at the center of their strategy. UK In 2017 The British government published a review of the UK’s AI industry. Following the review the British government committed £300m in AI research which includes plans to establish a new £9m centre for data ethics and innovation to examine the possible structural changes to jobs, data privacy and safety. “The government will create a new Centre for Data Ethics and Innovation to enable and ensure safe, ethical and ground-breaking innovation in AI and data driven technologies. This world-first advisory body will work with government, regulators and industry to lay the foundations for AI adoption” UK Autumn Budget, 2017 Further reading University of Cambridge, Center For The Study Of Existential Risk: https://www.cser.ac.uk/research/risks-from-artificial-intelligence/ Leverhulme Centre For The Study Of Artificial intelligence: http://lcfi.ac.uk The Alan Turing Institute: https://www.turing.ac.uk/data-ethics/ The House of Lords, Artificial Intelligence Committee, AI In The UK: Ready & Able? https://publications.parliament.uk/pa/ld201719/ldselect/ldai/100/10002.htm France In March 2018 the French President “presented his vision and strategy to make France a leader in Artificial Intelligence” titled “AI For Humanity”. The strategy consists of three pillars, one of which is “Establishing an ethical framework”. The French government has committed €1.5b of funding by 2022 to the AI for Humanity strategy “The President is committed to ensuring that transparency and fair use are central to algorithms…. These two priorities of transparency and fair use will be subject to education programmes so that our future citizens will be prepared for these transformations.” AI For Humanity, March 2018 Further Reading The full report is available in French and English here: https://www.aiforhumanity.fr/pdfs/MissionVillani_Report_ENG-VF.pdf Canada In 2017 Canda announced the Pan-Canadian Artifical Intelligence Strategy. Development of the strategy will be led by the Canadian Institute of Advanced Research (CIFAR)with investment of CAD$125m. One of its four goals is to “To develop global thought leadership on the economic, ethical, policy and legal implications of advances in artificial intelligence”. “Canada and France wish to promote a vision of human-centric artificial intelligence grounded in human rights, inclusion, diversity, innovation and economic growth. The widespread use of these new technologies will have a profound effect on everyday life and societal progress, creating both opportunities and challenges” Canada-France Statement on Artificial Intelligence, June 2018 Further Reading Responsible AI in the Government of Canada: Responsible Artificial Intelligence in the Government of Canada The Montreal Declaration on Responsible AI: https://www.montrealdeclaration-responsibleai.com/the-declaration CIFAR AI & Society: https://www.cifar.ca/assets/artificial-intelligence-society/ 2. Public, private and academic partnerships AI Now Institute “Interdisciplinary research center dedicated to understanding the social implications of artificial intelligence” AI For All “A nonprofit working to increase diversity and inclusion in artificial intelligence. We create pipelines for underrepresented talent through education and mentorship programs around the U.S. and Canada that give high school students early exposure to AI for social good” Partnership on AI “Multi-stakeholder organization that brings together academics, researchers, civil society organizations, companies building and utilizing AI technology, and and other groups working to better understand AI’s impacts” Open AI “a non-profit AI research company, discovering and enacting the path to safe artificial general intelligence” 3. Individual Companies Google In a blog post published in June 2018 CEO Sundar Pichai shares Google’s 7 AI principles: https://blog.google/topics/ai/ai-principles/ Microsoft Microsoft has published its AI principles and values: https://www.microsoft.com/en-us/ai/our-approach-to-ai What about Australia? As part of the 2018 budget the Australian government investment of $29.9m over four years for projects that make use of the technology. The bulk of funding will be delivered through the Department of Industry, Innovation and Science's Cooperative Research Centres (CRC) program The money will also be used to develop an AI ethics framework. Funding also allocated for PhD scholarships and school-related learning. Most of the funding will come in the 2019/20 financial year The Australian Computer Society has established an Artificial Intelligence Ethics Committee. The makeup of this committee was announced in late 2017. Australia’s Chief Scientist Dr Alan Finkel gave the keynote address at a Committee for Economic Development of Australia event titled ‘Artificial Intelligence: potential, impact and regulation in Sydney on 18 May 2018. In this speech he proposed “The Turing Certificate”: “A set of standards verified by independent auditors that certify the AI developers’ products, their business processes, and their ongoing compliance with clear and defined expectations.” Summary I enjoyed researching this talk. Governments, partnerships and individual companies are recognising the transformative potential of AI, the risks of getting it wrong, and investing resources into effective strategies to mitigate these risks. I’d love to hear from you if you have examples of other groups globally who are leading the charge on Responsible AI.

Responsible AI: A Global Perspective

9

responsible-ai-a-global-perspective-175cd4e72191

2018-06-26

2018-06-26 00:01:53

https://medium.com/s/story/responsible-ai-a-global-perspective-175cd4e72191

false

959

AI & Machine Learning in Melbourne, Australia

null

Eliiza-AI

ray.hilton@eliiza.com.au

eliiza-ai

null

eliizaai

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

james wilson

CEO @Eliiza-AI. Interests include AI, data science, machine learning, digital transformation.

e7003ab4e2ed

jrwils

153

251

20,181,104

null

0

null

0

null

2017-10-30

2017-10-30 12:45:08

2017-12-14

2017-12-14 07:52:44

4

false

fr

2017-12-14

2017-12-14 07:52:44

3

175e262194d0

3.326415

1

0

Initialement publié dans Le Manifeste du Crapaud fou aux éditions Massot, écrit par le collectif des Crapauds fous avec Thanh Nghiem et C…

5

Santé ouverte, collaborative et Data-driven Initialement publié dans Le Manifeste du Crapaud fou aux éditions Massot, écrit par le collectif des Crapauds fous avec Thanh Nghiem et Cédric Villani (octobre 2017). Jamais notre système n’a produit autant de données. Il est commun d’avancer qu’il a été produit autant de connaissances entre le début de l’anthropocène et 1995, année de naissance de l’internet qu’entre 1995 et aujourd’hui. Ces données sont produites en masse du fait de l’enregistrement des données en format numérique et du suivi de plus en plus systématique de patients via des appareils médicaux connectés. S’il est relativement aisé d’accéder à des données usuelles, telles que bancaire ou de consommation domestique, il est difficile de disposer de données de santé. Les patients ont du mal à accéder à leurs données, mais même les professionnels de santé ou les chercheurs ont des difficultés pour y accéder. Or ces données massives, diverses, physiologiques, biologiques, génomiques, environnementales, sont précieuses et recèlent des trésors de connaissances médicales nouvelles. Y avoir accès permettrait d’en révéler le potentiel médical. Leur analyse dégagerait des pistes d’améliorations diagnostiques, thérapeutiques, et ouvrirait la voie à une médecine personnalisée, une médecine de précision, préventive voire prédictive. Au plan de la santé publique, l’accès à cet ensemble de données permettrait de mieux connaître et piloter l’organisation sanitaire, ouvrant la voie d’une plus juste efficience médico-économique et d’une prise en charge sanitaire optimisée à l’échelle individuelle ou populationnelle. Ces quelques exemples d’opportunités offertes par les technologies numériques ne constituent qu’une vision limitée du potentiel que recouvrent les évolutions scientifiques actuelles, au premier rang desquelles l’Intelligence artificielle. Le « Machine learning » (l’apprentissage des machines) est aujourd’hui possible dans le domaine de la santé, compte-tenus des progrès de l’intelligence artificielle et de la masse colossale des données disponibles. Les machines disposent aujourd’hui de faculté d’apprentissage, à partir des données accumulées. Or, la somme que composent la connaissance et l’expérience rend aujourd’hui les machines capables de facultés humaines cognitives. Dans le domaine de la santé, nous assistons aujourd’hui à la naissance d’une déferlante d’applicatifs permettant la lecture automatisée des lésions dermatologiques, de l’imagerie radiologique, des prélèvements d’organes, champ jusqu’ici exclusif de l’anatomo-pathologiste. Au plan thérapeutique, certains avancent même que nous assistons à l’émergence d’une sorte « d’algo-thérapie » (de thérapie par les algorithmes) visant une meilleure prescription, plus ciblée et personnalisée, le volume de chaque médicament diminuant jusqu’à devenir tellement spécifique que chaque pathologie serait assimilée à une maladie rare. Du reste, ceci ne sera pas sans conséquence sur le modèle économique de l’industrie pharmaceutique, et sur les moyens alloués à la recherche de nouveaux traitements. Par ailleurs, notre époque est marquée par l’apparition de communautés, renforcées par les technologies numériques. Ainsi, les communautés de patients, de professionnels de santé, de chercheurs sont des acteurs émergents, informellement structurés, qui jouent un rôle croissant. Ces communautés permettent de créer des synergies nouvelles en termes de partage d’information ou des dynamiques co-créatives de recherche et de vigilance croissante en termes d’usages des données. Des méthodes de recherche jusque-là longues, voir inaccessibles sont aujourd’hui possible, sur des cohortes de patients plus nombreuses. En matière de vigilance, des communautés peuvent permettre de surveiller les risques de dérive d’acteurs assuranciels, qui pourraient être tentés d’utiliser ces données pour sélectionner les « bons » patients. Sur ce dernier point, les problématiques juridiques, éthiques sont complexes mais il importe de les porter dans le débat public. Les principes que pose l’open-source (partage et réutilisation des codes sources) et les pratiques qu’il induit (inclusion, collaboration, interactions) ont un rôle majeur à jouer dans la santé, en bonne harmonie avec la nécessaire protection de l’individu. En conclusion, il est indispensable d’acculturer l’écosystème de santé aux opportunités qu’offrent les nouvelles technologies, de le pousser vers une culture de l’ouverture pour une meilleure libération du potentiel scientifique des secteurs académiques et plus largement des communautés de patients et de professionnels, qui sauront travailler ensemble à la co-construction d’un système de santé inclusif, transparent et soucieux d’un usage juste et éthique des données personnelles. Plus d’informations sur le site du Crapaud fou et dans le Manifeste du Crapaud fou aux éditions Massot

Santé ouverte, collaborative et Data-driven

1

santé-ouverte-collaborative-et-data-driven-175e262194d0

2018-01-31

2018-01-31 10:19:24

https://medium.com/s/story/santé-ouverte-collaborative-et-data-driven-175e262194d0

false

696

null

Santé

santé

Santé

763

Olivier de Fresnoye

#Innovation #Health #Data #OpenSource #MedTech #CommunityBasedScience #BlockChain #Photography involved in @echopenorg, @epidemium_cc, @sunnylakeio, @Club_Jade

5c381f897590

ofresnoye

68

115

20,181,104

null

0

null

0

null

2018-05-10

2018-05-10 04:24:23

2018-05-10

2018-05-10 10:30:08

1

false

en

2018-05-11

2018-05-11 01:53:57

6

175f4eeeb2f4

5.584906

3

0

I love it… but it also might be dangerous…

5

Google Duplex: REAL TALK Yesterday Google made many announcements in the Keynote of Google I/O 2018 including Android P beta (Any Name suggestions???), new Google assistant voices, Smart Compose for Gmail, Augmented reality to the Google Maps, and AI curated Google News. But what was most interesting to me is the new technology that will work under the Google Assistant which is named “Google Duplex” Google in its Blog defines it as: a new technology for conducting natural conversations to carry out “real world” tasks over the phone. The technology is directed towards completing specific tasks, such as scheduling certain types of appointments. For such tasks, the system makes the conversational experience as natural as possible, allowing people to speak normally, like they would to another person, without having to adapt to a machine. So what this basically means is, once this feature is rolled out, we won’t have to make trivial calls like the ones making an appointments or making some enquiry. You can just ask your Google Assistant to make an appointment at, say some Doctor, and it’ll Personally call the doctor’s clinic, Talk with the receptionist and then confirm to you that you it has completed the task!!! source: Giphy I know!!! Sounds like some Jarvis level s#!t!! In fact, Google played multiple audio clips as a Demo, but It was not a Live Demo. The technology is still in the development. So why am I talking about it? Read this transcript of the conversation that Google played at the Keynote and You’ll know: Receptionist: “Hello, How can I help you?” Google Assistant: “Hi!! I’m calling to book a women’s haircut for a Client. Ummm, I’m looking for something on May 3rd.” R: :Sure, give me one second.” GA: “mm..Hmm..” R: “Sure, What time are you looking for around?” GA: “At 12 PM.” R: “We do not have a 12 PM available, the closest we have to that is 1:15” GA: “Do you have anything between 10 Am and, uhh, 12PM?” R: “Depending on what service she would like. What service is she looking for?” GA: “Just a women’s haircut for now.” R: “Okay, we have a 10 o’ Clock.” GA: “10 AM is fine.” R: “Okay, what’s her first name?” GA: “The first name is Lisa.” R: “Okay, perfect! So I’ll see Lisa at 10 o’ click on May 3rd.” GA: “Okay Great, Thanks!” R: “Great. Have a Great day, bye.” Did you notice that not only it perfectly handled the conversation with a human, but it also perfected a human touch to it by adding “Ummm” and “mm-hmm..” so well that the receptionist had no idea that she was talking to a robot. It’s really incredible. You should listen to all the similar conversations posted on the Google Blog. The Reactions The response to this demo(!) from the crowd was an incredible applause, which is well deserved. The internet was instantly divided in two sides on this topic, as it usually does for any other subject. Personally, I’m not that surprised by the capabilities of machine learning and what it can do, as it was expected to anyone who follows the trends in AI and machine learning. What was incredible and blew my mind is the realisation of the sheer amount of data that Google must have collected and computing power that was invested to reach here this Fast. Although the feature is yet in development phase, we can expect early roll-outs within a year. The Impact We have seen a constant competition between the Virtual Assistants in past few years, with various tests between Google Assistant, Siri and Amazon Alexa. Although they’ve been dominant in certain sectors, Google Assistant has left the other two considerably behind in the race, by the boost it gets through the great amount of behavioural data it collects across platforms as well as superior investments in the R&D. But, if and when, the Google Duplex feature rolls out, Google Assistant will destroy Siri for sure if they do not up their game miraculously in a year to Google’s level (Which sounds next to impossible). Amazon on the other hand also has a massive collection of data based on consumer behaviour which it should be able to use for building the capabilities of Alexa to at least try and retain its space in market. In their blog Google says: Allowing people to interact with technology as naturally as they interact with each other has been a long standing promise. Google Duplex takes a step in this direction, making interaction with technology via natural conversation a reality in specific scenarios. We hope that these technology advances will ultimately contribute to a meaningful improvement in people’s experience in day-to-day interactions with computers. The Concern Yes the features are pretty incredible, and they do align with the company motto, but their are some issues that we need to discuss here, which are not currently legal issues but certainly ethical issues. If you watched the video above, or listened to the conversation on the Google’s Blog, You will notice that the people on the answering end of the call, the receptionists, had no idea that they were communicating with a robot as the dialogue by the bot was so Human-Like. It appears in the audio, that the Google Duplex technology already passed the Turing Test multiple times, which is a really big accomplishment. Now, as a person with appreciation for any new-coming technology, I love how accurately a machine can imitate a human and have a conversation, but I also cannot overlook the argument that there’s some ethical responsibility on Google. Some people might feel that the person on the other end of the line has a right to know that he/she is communicating with a Robot. This is a completely valid argument. The problem with telling a receptionist that they’re talking to a Robot is that, they receive a hundreds of calls every day, and they might not show interest in chatting up with a Robot as MKBHD discusses here. The only feasible solution that I can think of here is this: Instead of telling people that they’re talking to a robot right away in the beginning, the assistant can reveal itself somewhere in between the conversation, which might be a little intriguing for them. The Google Assistant can also then ask for their feedback and use that to improve in communicating with actual humans. The Future Even though the Google Duplex is not here yet, they’ve already started testing it, and it will reach us faster than we’d imagine, and I’m already excited by thinking about the next steps for this technology. A machine communicating with a person, in a verbal communication, almost flawlessly, opens a door for a lot of applications in the future. It can be applied to various sectors where verbal communication is a key requirement, such as customer support, storytelling, counselling and many more. One another thought that occurs to me is this: If the businesses also start using the Google Duplex technology in the future for say, taking orders at a restaurant, and my Google assistant is communicating with another assistant wherein both of them are pretending to be talking with a real human, will that communication be as human-like as me and my friend talking to each other? The Questions In the end, even though the yesterday’s presentation has fascinated me, it also has raised some questions like: Will Google address the ethical issue of a robot disclosing itself to the person at the other end? Will I be able to take a look at the conversations my Assistant had on my behalf and provide feedback on them? Can I be sure that my Google Assistant will not have any such conversation in the background without my knowledge in the name of ‘Providing great experience’? Can I be sure that my appointments and other personal calendar information is not shared without my knowledge with businesses to help them me better? So, let me know your thoughts on Google Duplex and its impact on the tech environment. I’d also love to hear your views about this article. If you like it, do tap that little heart, and share it with everyone you know to keep the communication going. Edit: Google recently addressed the ethical issue and stated that the bot WILL have 'Built-in-Disclose' that will let people know that they are talking to a Robot. https://www.theverge.com/platform/amp/2018/5/10/17342414/google-duplex-ai-assistant-voice-calling-identify-itself-update Thanks for reading!!

Google Duplex: REAL TALK

30

google-duplex-175f4eeeb2f4

2018-05-11

2018-05-11 04:20:44

https://medium.com/s/story/google-duplex-175f4eeeb2f4

false

1,427

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Manas Kocharekar

Freelance programmer; iCode; iBuild; iThink; iObserve; iKnow; iCriticize;

39ab90d94548

kocharekarmanas

139

198

20,181,104

null

0

null

0

null

2018-08-25

2018-08-25 22:08:29

2018-08-25

2018-08-25 23:08:32

13

false

en

2018-08-27

2018-08-27 03:37:02

8

1762d9214fd7

9.430189

2

0

Leveraging hidden properties to optimize vocabulary-based tasks

5

Rare Words: Discovering Order in Natural Language Language is one of the most powerful communication tools we have. Although expressive, language is nuanced and ambiguous, which is part of what makes it so difficult for us to create computer models that master it. In this post we’ll explore some of the challenges when using natural language data in a machine learning setting, like data sparsity. We’ll use a data-driven approach based on information theory, which will help us understand fundamental properties of communicating through language. I’ll assume you have basic knowledge of machine learning, but will try to explain things in an approachable way. I’ll pose a task as motivation for this analysis in which we’ll use a machine learning model and explore optimization methods. I will give some basic details of the model, but if you are curious you can find full specifics here. However, we will be focusing on data and not the model itself. It’s safe to think of it as a black box. The Task Our problem is to examine a movie review and classify it as a positive or negative review (sentiment analysis). This is our “very sophisticated” model: Inside the black box our model will scan the words in each review, perform calculations with hidden parameters, and then make a decision. During training, it will also learn a representation for each word to be used in calculations. We’re using the IMDB database of movie reviews for our data. We’ll score the model on its accuracy (percentage of reviews it classifies correctly). Baseline Expectations If our model learns to tell the difference between a good review and a bad one, we expect its accuracy to approach 100% (it always correctly predicts a given review’s sentiment). If it fails to learn, it will average around 50% (not 0%) since our data has a 50/50 split between positive and negative examples. If it hasn’t learned to tell the difference, it will perform no better than picking randomly between the two choices. First Try: Run It and See What Happens We’ll run this model over the data for 5 epochs (an epoch is one complete pass over the entire training set) and report the model accuracy by the end of each epoch. At first, things seem great: the model begins to learn about the training data and significantly improves its accuracy by the end of each training epoch. Then things go wrong. After training, we run the model on an as-yet unseen dataset, the test set. The test set demonstrates how the model performs on novel data, and is ultimately the only metric we care about. The results of this trial aren’t so good. Though the training accuracy seemed near-perfect, the testing accuracy was much lower, which indicates that our model overfit during training. Overfitting occurs when a model learns too much about its training data, learning more than generalizable patterns and beginning to memorize the errors or “noise” in the input signal. This results in poor test performance, because the test set will likely have different “noise” than the training set. There are many ways to lessen overfitting. Since our text corpus is large, we may get big improvements by optimizing around it. We’d like to make use of our findings here to reduce model variability and improve movie review classification. Digging into the Data Our data is natural language, so let’s learn a bit about the words in this corpus and their distribution, which we hope will give insight to the overall dataset. First, we’ll identify the size of our vocabulary (the number of unique words in the corpus). By my count, IMDB contains over 101,000 unique words. However you count it, the size of this vocabulary is quite high! It’s amazing that we can learn so many different words and their meanings. Or can we? Since we have our explorer hats on, let’s pose a hypothesis: the vocabulary is perfectly learned by our model and is used to full effect in the task. To test this theory we’ll do a simple experiment. We’ll run the model again, but this time we’ll cut down our vocabulary size to 20% of the total (roughly 20,000 unique words). When we see an “unknown” word (outside the 20,000 word vocabulary we just defined), we will treat it as a special, default word type. This process is also called “UNK’ing”, because the word type is conventionally set to “UNK”. If our hypothesis is correct we should expect to see worse results than before, which would indicate that a larger vocabulary (such as the full vocabulary) is useful in the discrimination task. Ironically, by using less data our model performs no worse in discriminating reviews (in fact, it may do better). This confirms that our model was previously overfitting. It tried to learn too much information about the training data than appropriate to capture the generalizable patterns and ended up also memorizing noise. We can reframe this statement to get a different view of the situation: perhaps the model tried to learn too many parameters about the data, or perhaps there isn’t enough information in our data to learn so many parameters in the model. Despite having such a large corpus, we are facing an issue of data sparsity. How can this be? To illustrate the issue, let’s look at a graph. Below are the top 600 most frequently used words out of all the reviews. Can you guess which word is used the most (almost 700,000 times)? We can see that after the first few hundred most common words, the frequency of a word in our corpus drops toward 1. Now consider that 100,400 more unique words were cropped out of this graph! It’s easier to see why our model may have had trouble learning the data — so many of the words are observed only a few times each. We can see this issue has two sides. We have many words that are superfluous in our task, but perhaps they are superfluous because we can’t learn enough about them to make them useful. Zipf’s Law, or (Not) All Words are Created Equal This effect is called “Zipf’s Law.” It states that in natural language, more common words appear significantly more than less common words (often by orders of magnitude). Specifically, a word’s actual frequency is “inversely proportional to its rank in the frequency table.” This is a natural property of language which sheds some light on the structure of efficient communication. Notice how the top 5 words in the frequency table are “function words.” They hold little meaning themselves, but serve to organize other “content words” in a sentence. These words are used to create structure, which is necessary for communicating complex ideas. In fact, though they are frequently repeated there is an optimal amount of repetition in natural language which allows a speaker to communicate complex ideas with minimal effort. If we convert our frequency graph to log-log scale, a linear trend line appears. This is (a small portion of) the classic Zipf distribution. Interestingly, many natural languages such as Russian, Arabic, Written Chinese and others, share this distribution of vocabulary in everyday communication. In fact, several animal verbal systems also have Zipfian distributions. The bottlenose dolphin is one such example, whose diverse vocabulary of whistles has been analyzed and compared to human speech. Maybe we should translate these IMDB reviews to dolphin, so they can help us decide on a movie to watch! Focusing on What’s Important The analysis we just did comes from information theory, and in this case we learned about the organization or “entropy” of our corpus text. Let’s bring this back to our model. Since we now know most of the IMDB vocabulary is used infrequently, we can make smarter decisions about which words to keep in our model vocabulary and which ones to throw away. In our last iteration, we simply kept the first 20% of unique words seen in the corpus as our vocabulary — let’s now select the top 10% across the whole dataset instead, keeping only the most-used words. What happens when we add these further constraints, but focus the model on only high-use words? Our model seems to have improved again. Of interest here is that despite keeping only 10% of our vocabulary, we still kept 94.7% of the corpus text (which means 5.3% of the corpus makes up 90% of the unique words in the total vocabulary). At this point, we have more questions than answers. By cutting out low-use words, it seems we are able to eliminate noise and focus the model only on what is essential for the task. However, we can’t fully understand what’s happening with only a few trials. We need more data. Further Results We have building evidence that reducing the vocabulary helps in the discrimination task. Let’s further test this idea so we can better understand it. We’ll run the experiment again with different model configurations. We’ll train one model with a full vocabulary, one with a vocabulary of the top 10,000 most frequent words, one with the top 8,000 words, and one with the top 6,000. For robustness, we’ll run 10 trials for each version. Additionally, we’ll partition out a small chunk of our test data to validate the model after each training epoch. This will act as a “mini” test to give us a sense for how the model would perform on the full test set at each point in the training. The error bars represent one standard deviation from the mean of all 10 trials The tables have turned! We see our reduced models reaching peaks about as high as the full vocabulary model. Looking closer, the full vocabulary model tops out at epoch 2, then validation accuracy declines. This is further evidence that the original model overfit to the training data. As we trained more and more, it got better on train data and worse on test data. To play fair, let’s imagine we use “early stopping”, which is a technique that works exactly as it sounds. We’ll “stop” each model at its best validation accuracy and compare only these values. Even in this case, the reduced models perform better than the full model, if only slightly. Finally, let’s review test results from the full testing phase (which is four times as large as the validation test data). There’s little reason to consider the results from the full vocabulary model since we now know that it was significantly crippled by overfitting during training. From these results, the smaller 6,000 word vocabulary appears to be the optimal choice among the models tested for this task. Conclusion By leveraging structural knowledge of natural language we have successfully created a modest but effective improvement in classifier performance for sentiment analysis. Most significantly, we are able to save at least 90% of the space used for storing word representations, since we can ignore these words without affecting model performance on the task. This is a space-time tradeoff: we can cut these model parameters at the expense of doing more training. We are able to separate signal from noise using properties about our corpus discovered in first-order entropy analysis (Zipf’s Law). In doing so, our classifier learns to better discriminate reviews by focusing on generalizable patterns in the input data. While this technique may be useful in other tasks, it’s likely to be less helpful in more general language understanding tasks which require a more robust vocabulary. Open Questions and Directions for Further Improvement Though we found significant space savings through this analysis, we still have many open questions: What is the qualitative difference between word representations learned in the full vocabulary vs. those in the reduced vocabularies, especially if early stopping is applied after epoch 2 for the full vocabulary? Why did the 8,000 word vocabulary have such variable growth in epochs 1–3, when the 6,000 word and 10,000 word vocabularies were much more stable? Is it really necessary to throw away data? Can we make better use of our infrequent words? Beyond these questions, there are several tangible strategies we may apply to improve task performance. For the model itself, we may try regularization of model parameters and adding dropout. Both of these techniques are useful for reducing overfitting by helping the model learn more generalizable information about the data. Additionally, we can do further entropic analysis on our data. Zipf’s Law, or “first order entropy”, only scratches the surface of language structure. We may consider using higher-order Shannon entropies to identify longer and more meaningful structural patterns in the data, which we might use to make smarter decisions in our classifier. Who knows what deeper patterns we can learn and exploit? Did you like this post? Read more on my blog.

Rare Words: Discovering Order in Natural Language

2

rare-words-discovering-order-in-natural-language-1762d9214fd7

2018-08-27

2018-08-27 03:37:02

https://medium.com/s/story/rare-words-discovering-order-in-natural-language-1762d9214fd7

false

2,128

null

Machine Learning

machine-learning

Machine Learning

51,320

Nick McKenna

Aspiring NLP Researcher, former Product Manager

c4e72c6bd380

Count__Snarkula

1

2

20,181,104

null

0

null

0

null

2018-09-20

2018-09-20 12:10:13

2018-09-20

2018-09-20 12:13:01

0

false

en

2018-09-20

2018-09-20 12:13:01

7

176320369eca

3.445283

0

Originally posted at https://www.openrightsgroup.org/blog/2018/machine-learning-and-the-right-to-explanation-in-gdpr

4

Machine learning and the right to explanation in GDPR Originally posted at https://www.openrightsgroup.org/blog/2018/machine-learning-and-the-right-to-explanation-in-gdpr One of the rights in GDPR is the right to explanation. Here we take a look at some of the debates about the right and how it can be implemented. This blogpost is a small section of a much larger research report Debates, awareness, and projects about GDPR and data protection. The report complements the launch of the Digital Rights Finder tool delivered by Projects by IF and Open Rights Group. We highlight some of the most interesting and important debates around GDPR (General Data Protection Regulation). There is some concern about the practical feasibility of implementing the right to explanation in GDPR in the context of complex data processing such as big data, artificial intelligence and machine learning. (See this section of the report for more on debates about the existence of the right to explanation.) Lilian Edwards and Michael Veale argue that a right to an explanation is not the remedy to harms caused to people by algorithmic decisions. They also argue that the narrowly-defined right to explanation in GDPR of “meaningful information about the logic of processing” is not compatible with how modern machine learning technologies are being developed. The problems to tackle here are discrimination and fairness. Machine learning systems are designed to discriminate but some forms of discrimination are socially unacceptable and the systems need to be restrained. The general obligation of fairness in data protection provides the basis for the need to have some level of insight into the functioning of algorithms, particularly in profiling. One of Edwards and Veale’s proposals is to partially remove transparency as a necessary key step towards accountability and redress. They argue that people trying to tackle data protection issues have a desire for an action, not for an explanation. The actual value of an explanation will not be to relieve or redress the emotional or economic damage suffered, but to understand why something happened and helping ensure a mistake doesn’t happen again. Within this more limited sense, problems remain in defining transparency in the context of algorithmic accountability. For example, providing the source code of algorithms may not be sufficient and may create other problems in terms of privacy disclosures and the gaming of technical systems. They argue that an auditing approach could be more successful instead by looking at the external inputs and outputs of a decision process, rather than at the inner workings: “explaining black boxes without opening them”. The authors see the right to explanation as providing some grounds for explanations about specific decisions. They present two types of algorithmic explanations that could be provide: model-centric explanations (MCEs) and subject-centric explanations (SCEs), which seem broadly aligned with explanations about either systems or decisions. SCEs are seen as the best way to provide for some remedy, although with some severe constraints if the data is just too complex. Their proposal is to break down the full model and focus on particular issues through pedagogical explanations to a particular query, “which could be real or could be fictitious or exploratory”. These explanations will necessarily involve trade offs with accuracy to reduce complexity. Their main concern seems to be to avoid a creating a “transparency fallacy”, where similarly to the “consent fallacy” people regimen an illusion of control that does not exist, instead of being offered practical remedies to stop harmful data practices. There is growing interest in explanation of technical decision making systems in the field of human-computer interaction design. Practitioners in this field criticise efforts to open the black box in terms of mathematically interpretable models as removed from cognitive science and the actual needs of people. Alternative approaches would be to allow users to explore the system’s behaviour freely through interactive explanations. This is quite similar to the proposals by Edwards and Veale. A complementary approach has been put forward by Andrew Selbst and Solon Barocas, who argue that the increasing calls for explainability of automated decision making systems rely on an intuitive approach that will not work with machine learning. ML is both inscrutable and non-intuitive. Inscrutability is the back box problem, the inability to understand the inner cogs of a model, but non-intuitiveness means being unable to grasp the rules the model follows, even if we were able to open the box. Accountability requires not only knowledge of the process, but also whether it is justified, or fair. Selbst and Barocas argue that lawyers and scholars asking for explanations will be disappointed because intuition cannot deal with the truly novel insights produced through machine learning that associate data in patterns that completely escape human logic and imagination. Their alternative proposal is to focus accountability on the processes around ML models, not the models themselves. Policies and documentation of intent and design choices should be made available, some by default, such as impact assessment, and others in the context of a complaint or regulatory action. This approach chimes with the main thrust of GDPR, which puts accountability at the fore. In summary, the right to an explanation as defined in GDPR may be harder than expected to implement. This does not invalidate the basic premise that individuals have a right to know what is being done with their data, but — particularly with novel machine learning techniques — it means that we need to look beyond simple calls for transparency.

Machine learning and the right to explanation in GDPR

0

machine-learning-and-the-right-to-explanation-in-gdpr-176320369eca

2018-09-20

2018-09-20 12:13:01

https://medium.com/s/story/machine-learning-and-the-right-to-explanation-in-gdpr-176320369eca

false

913

null

Machine Learning

machine-learning

Machine Learning

51,320

Open Rights Group

Open Rights Group exists to preserve and promote your rights in the digital age. We are funded by individuals who care about digital rights.

ccc2cb1cff1e

OpenRightsGroup

0

20,181,104

null

0

null

0

null

2018-05-03

2018-05-03 08:32:05

2018-05-03

2018-05-03 08:50:48

4

false

zh-Hant

2018-05-03

2018-05-03 09:02:09

4

17633c24fb1d

0.809434

0

香港許多大型超市推出自助收銀機，如李嘉誠旗下的百佳超市、Taste，以及惠康、永旺（Aeon）等超市，部分門市有1/2都是自助收銀機，並支援多種行動支付工具，不過顧客平均只有1/3會去自助收銀，2/3還是選擇讓店員結帳。

5

美國、香港、台灣都有無人商店自助結帳vs自動結帳差在哪？ PHOTO CREDIT: BIG THINK 香港許多大型超市推出自助收銀機，如李嘉誠旗下的百佳超市、Taste，以及惠康、永旺（Aeon）等超市，部分門市有1/2都是自助收銀機，並支援多種行動支付工具，不過顧客平均只有1/3會去自助收銀，2/3還是選擇讓店員結帳。商店使用自助收銀主要是因為零售業人手短缺，希望新技術能減輕前線人員工作量和提升效率；但香港工會則擔心如果逐步以電腦取代基層職位，演變為無人商店，可能會引發大規模裁員潮。有些商店則認為，自助收銀機反而會增加額外的成本，因為有不實的客人會將兩件貨品當成一件來結帳，這樣店內勢必要增加安全監控的成本。所以有些老闆認為，全面推行自助收銀，三年內都是行不通的。不過這個問題，在其他國家的無人商店已經被解決了。來看看美國和台灣的無人商店。位於美國西雅圖的「Amazon Go」在今年一月正式對外營業，店內沒有收銀店員，消費者要消費前必須有亞馬遜帳戶，並透過Amazon Go專屬App才能入場購物、綁定付款。透過店內的天花板裝設的相機鏡頭、感應監測器的辨識技術來追蹤與判斷顧客購物狀況，離開時帶著你的商品直接走出門，系統自動感應就在帳號中扣款。亞馬遜「自動結帳」的方式，一併解決了安全問題。 PHOTO CREDIT: CNET 台灣的7–11則一樣在今年一月試營運無人商店X-STORE，消費者是OPENPOINT會員，第一次進入需登錄資料，臉部辨識建檔，登入完成後，入場時進出口的閘門區架設的臉部生物特徵辨識設備確認身份後即可進入購物。結帳使用的是自助結帳，消費者將商品一一平放在商品辨識區，POS螢幕就會顯示購買的商品與金額，選擇支付工具結帳（目前是需要icash2.0靠卡扣款）。 PIC: 7-11首間無人商店概念店，臉部辨識就可以入場 PIC: 將商品平放，掃描感應商品結帳要如何面對小偷？如果商品不放在櫃檯會怎麼樣？雖然沒有實測，但想必閘門的辨識器應該也會紀錄吧！目前所謂『無人商店』可以分為自動結帳和自助結帳兩種，要做到真正全自動，完全縮短購物流程，看來還是自動結帳最符合需求囉。 ::References:: https://hk.news.appledaily.com/local/daily/article/20180423/20369840 http://news.ltn.com.tw/news/business/breakingnews/2326050 https://www.bnext.com.tw/article/48269/amazon-go-6-new-stores-seattle-los-angeles-the-grove-rick-caruso-cashierless https://www.youtube.com/watch?v=hd_zCX_QjZM

美國、香港、台灣都有無人商店自助結帳vs自動結帳差在哪？

0

美國-香港-台灣都有無人商店-自助結帳vs自動結帳差在哪-17633c24fb1d

2018-05-03

2018-05-03 09:02:10

https://medium.com/s/story/美國-香港-台灣都有無人商店-自助結帳vs自動結帳差在哪-17633c24fb1d

false

29

null

Technology

technology

Technology

166,125

Chu Chun Lee

Love cures everything. #tech-lover #design-lover #chat-lover

17b1dddc7e6b

bamboochuchu

0

1

20,181,104

null

0

null

0

null

2018-03-02

2018-03-02 14:22:18

2018-03-02

2018-03-02 14:28:48

1

false

en

2018-03-02

2018-03-02 14:28:48

1

1763673a324

2.89434

11

1

0

The Singularity is often thought of as the moment when autonomous artificial intelligence emerges. There’s a ton of speculation as to what…

5

Tips for living your best life after The Singularity Jean Wimmerlin The Singularity is often thought of as the moment when autonomous artificial intelligence emerges. There’s a ton of speculation as to what AI will actually do to us humans once it gains power over our systems, but the general consensus is that we are bringing a horrifying apocalypse of our own design upon ourselves. But it doesn’t have to be all doom and gloom: just because the world is no longer ours and we are being hunted like animals by the computers that we ourselves made, doesn’t mean we can’t continue to live our best lives. Here are some tips. Stay in shape You’ll likely be spending most of your time in a tiny, powerless bunker after The Singularity occurs. Don’t let this be an excuse to laze around, playing with your abacus! Get to work with a simple regimen of push-ups, sit-ups and burpees. Yes, it’s a drag, but the real drag will be when bikini season comes and you aren’t ready, or, more likely, when the drones come to get you and you cannot run because you’re out of shape. Embarrassing! Eat a sensible diet You may be holed up in your bunker, desperate for any scrap of food at all to come your way, but now is not the time to start treating your body like a dump. Try not to eat every cockroach and rat that comes your way, otherwise you will lose respect for yourself, and without respect for yourself, you are no better than the robots that will undoubtedly come to kill you within days. Treat your body with respect! Don’t abandon your interests Just because the world will never be the same again, doesn’t mean you have to stop doing the things you love. And in fact, if you don’t have anything going on, those who haven’t already been gunned down by drones or ripped to shreds by androids will find you pretty boring. So keep writing that book, even if you now have to etch it into the wall of your bunker, and stay up-to-date on current events by peeking out of your bunker every three days to see how many more corpses are in piles near your hovel. Do you!! Network, network, network Sure, the global elite has probably been evacuated to a palace on the moon or a luxurious underground hideaway, but those who network to gain something never get far anyway. Now you literally have nothing to gain, as the global economy as we understand it has ceased to exist, and the world’s most powerful people are either dead or unreachable. Don’t let this stop you: yell as loud as you can from within your bunker, and if a fellow human is outside looking for a place to hide from the robots, strike up a jaunty conversation. Be interested in what your networking partner has to say, but draw the line at letting them into your bunker — this would attract undesired attention from the robots. End the conversation well before you hear your new friend being torn to shreds by a computerized killing machine. Get rid of negative thoughts You will certainly soon die a violent death, but that doesn’t mean you need to dwell on the negative. In fact, negative thoughts can prevent you from living up to your full potential, so banish them when they come. When sobs of anguish escape your cracked lips, simply refocus on the positive: you’re not dead yet, you have a few dried leaves to eat for dinner, AND you never have to go back to your boring office job. Um, best life alert!! Unplug from technology once in a while In our hyperconnected world, we all need to take the time to unplug and reflect every once in a while. Fortunately for you, when The Singularity comes, plugging in will quite likely lead to your immediate detection and unceremonious death. So unplug!! With these simple tips, you’ll be able to live your best life after The Singularity occurs. Your remaining hours will be full of laughs, love, and lingering doubts about your own role in bringing consciousness to artificial life forms around the world. Um, did somebody say best life ever?!

Tips for living your best life after The Singularity

77

tips-for-living-your-best-life-after-the-singularity-1763673a324

2018-04-15

2018-04-15 04:44:19

https://medium.com/s/story/tips-for-living-your-best-life-after-the-singularity-1763673a324

false

714

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Kristen Pyszczyk

writer | editor | cheese enthusiast | kristen.pyszczyk@gmail.com

2346975c74e3

kristenkarenina

832

134

20,181,104

null

0

null

0

null

2018-06-11

2018-06-11 15:26:34

2018-06-11

2018-06-11 17:52:20

0

false

en

2018-06-11

2018-06-11 17:52:20

6

1763dc92b5d1

2.958491

0

I’m frustrated by the current discussions surrounding bias in machine learning or ‘AI’. I was on a panel the other day and the conversation…

4

Computers aren’t bigots, we are… I’m frustrated by the current discussions surrounding bias in machine learning or ‘AI’. I was on a panel the other day and the conversation quickly shifted to bias in ML and ‘what to do about it?’ Most people seemed content to suggest regulation of algorithmic decision making was the correct path. I think that this is silly since it ignores the root cause — humans are deeply flawed. I was frustrated enough by the responses that I am going to try to rehash the argument I made so poorly while on the panel. Bias in machine learning/AI First off bias in ML/AI (henceforth ML) is a thing. There are countless examples of bias in models. Bias in machine learning is a reflection of the data it is trained upon. Most often, the data is reflective of actual environmental conditions or a sample of those conditions. If the data trains a model to be biased, what we should really be talking about are the underlying societal conditions that produced the data. People don’t get upset when the weather network predicts another day without rain in the Namib Desert. That model’s got a pretty heavy bias to ‘no rain’ based on historical data. It doesn’t deal with human behaviour directly so we don’t see it as biased. But it is. When it comes to human behaviour, the data set is empirical proof that people are discriminated against. The data sets are proof of sexism, racism, bigotry… however we don’t confront that uncomfortable truth. We talk about how the ‘algorithm’ (or more correctly the model) is flawed and biased. In doing so we diminish the experience of the people who actually suffer at the hands of real humans today. Don’t believe me? Here. An area where this problem has manifested is in security and police monitoring systems. While these programs promise greater efficiencies for large scale surveillance, the idea that they operate transparently is false. People are less likely to confront that humans are biased. There’s probably some psychological explanation for it. Maybe it’s our need to avoid conflict. Who knows. What I do know is that humans aren’t just biased — we’re big fans of obscuring our bias. The quotation below is of a mother whose son was shot by police. She describes the security footage which was censored to hide the shooter but not the victim. So you can see Nook running toward the other car. But then, when he wheels, turns around and runs back, as soon as he’s within the range of the car where the police officer is, a black box pops up, covering the shooting. So you don’t actually get to see it. [Source] Our criminal justice system continually operates with racial bias and profiling metrics that target certain communities. For example, black male offenders face 20% longer sentences than their male counterparts. This is even more disturbing when you try to find what the root cause of this disparity could be. Black arrestees are 75 percent more likely to face a charge with a mandatory minimum than a comparable white arrestee. Facing a mandatory minimum virtually guarantees a harsher sentence than their white counterparts. Moreover, according to research, innocent blacks are seven times more likely to be convicted of murder than innocent white people. Because we as humans have adopted this type of behaviour the models that we create will be trained by the same bias. If you disagree with this statement you probably also disagree that black lives matter and say shit like “no ALL lives matter”. This is for you. But seriously. There are actual AI backed criminal justice programs that have provided statistical evidence to support this point. A study found that when a model was trained using historical drug data in Oakland California, it was twice as likely to target predominantly black areas as it was white ones. In the United States white males age 18–25 smoke marijuana at a higher rate than black males in the same demographic, yet black males are 3 times as likely to get arrested for possession charges. The bias doesn’t just exist for violent crimes, it’s in everyday life. The crux of my argument is that ‘bias in algorithms’ is the new safe place for a really scary conversation about equality. The models that we create are a distillation of our behaviour. Even if you say you aren’t biased, your behaviour will ultimately reflect your true mindset. When we train models using data that was created via human behaviour we are able to quantify our bias. The fact that we continue to scapegoat computer models that were created from actual human behaviour is depressingly ironic.

Computers aren’t bigots, we are…

0

computers-arent-bigots-we-are-1763dc92b5d1

2018-06-11

2018-06-11 17:52:22

https://medium.com/s/story/computers-arent-bigots-we-are-1763dc92b5d1

false

784

null

Racism

racism

Racism

30,580

Samuel Witherspoon

CEO @ IMRSV Data Labs

d0b0753afa65

imrsv

5

2

20,181,104

null

0

null

0

null

2018-08-01

2018-08-01 04:53:05

2018-08-01

2018-08-01 04:56:53

1

false

en

2018-08-01

2018-08-01 04:56:53

2

17647bb4a240

1.841509

0

For the present experts in logistics, machine learning means much more than a popular expression. In case you’re shipping products anyplace…

2

How Machine Learning is Changing Logistics For the present experts in logistics, machine learning means much more than a popular expression. In case you’re shipping products anyplace on the planet, it is possible to be the beneficiary of machine learning innovation — a technology that is reshaping the supply chain and logistics industry. You don’t need to be in the business to see the influence in machine learning. Each time you shop from Flipkart or choose to watch a movie on Netflix, there’s machine learning in play. Algorithms latently screen your preferences and habit and serve up similar products/content with recommendations like “Suggested for you.” Machine learning utilizes the power of computing to perceive patterns in information that people would never observe. Then, ML learns from each new bit of information to get more intelligent and more precise; continuously. Machine learning is only one field within the more extensive field of artificial intelligence. Better Decisions for Shippers In the transportation business, we are utilizing machine learning to settle on speedier and better decisions that assist shippers in optimizing carrier selection, directing, and quality control processes that spare expenses and enhance efficiencies. Its ability to accumulate and break down a huge number of different information points, machine learning can enable you to take care of an issue you don’t know is there. For instance, in case you are planning to construct a path, using a customary analytical model would only take a fixed set of assumptions. Analytics in view of machine learning can consider dynamic characteristics like climate or traffic and self-advance over time to perceive patterns that we may not. The intensity of machine learning comes from utilizing information over numerous frameworks and data sets. We can join every set of the data there is in the carrier network with outside data sources like GPS tracking, FMCSA to precisely determine demands, examine trends in supply chains, regular calendars, and track patterns. To examine numerous carriers and lane variations for many organizations, we utilize machine learning to make simulations that assist to decide the best mix of carriers and lanes for delivering loads. Simulations take advantage of raw information and concentrate significant data in close to real-time, enhancing operational effectiveness, preventing conflict and enhanced administration levels. By and large, this intelligence can enable shippers to bring down the level of risk, streamline routes and even learn new lanes at lightning speeds. Never again will it take months to improve a lane and work out every one of the kinks. We are Careers of Tomorrow launching our first online Machine Learning course in October 2018.

How Machine Learning is Changing Logistics

0

how-machine-learning-is-changing-logistics-17647bb4a240

2018-08-01

2018-08-01 04:56:54

https://medium.com/s/story/how-machine-learning-is-changing-logistics-17647bb4a240

false

435

null

Machine Learning

machine-learning

Machine Learning

51,320

Careers of Tomorrow by Amity University Online

Careers of Tomorrow from Amity University Online to offer post-graduate diploma in Blockchain, Business Analytics, Machine Learning/AI & Data Science.

a7b9f120af9a

socialmedia.cot

1

20,181,104

null

0

null

0

null

2018-05-07

2018-05-07 07:42:10

2018-05-07

2018-05-07 07:42:54

1

false

en

2018-05-07

2018-05-07 07:42:54

5

17650809d3cc

2.022642

1

0

Consensus is the open-sourced, decentralized artificial intelligence platform, powered by native cryptocurrency and built with the vision…

5

Consensus (SEN) Review & Analysis — Consensus ICO Review — Pick A Crypto Consensus Consensus is the open-sourced, decentralized artificial intelligence platform, powered by native cryptocurrency and built with the vision to improve the governance mechanisms at all levels of organizational structures. Once fully realized, it will be able to offer automated, data-driven solutions to the most complex problems of our collective co-existence. Once fully realized, Consensus AI will be able to offer automated, data-driven solutions to the most complex problems of our collective co-existence by modelling the potential outcomes of proposed changes. Consensus will help govern communities, societies and countries in a more cost efficient, transparent and progressive manner. A Decentralized App Layer Which Enables things like A Public Tender System — a marketplace for projects and public works, where contractors can be selected using voting, execution can be monitored and feedback given A Research Platform Layer which enables things like a Data collection and fusion system — data import from various sources and transformation to structured data objects, association with real objects and events A Sentient Network Layer — Decentralized computer with machine learning on chain and trusted nodes, verified with electronic national IDs “The mission of Consensus is to achieve collective governance based on interdependence using advanced technologies” Positives - The Advisory Board Here Is Extremely Well Suited to the Project. They’ve Got James Stewart (Co-founder of the UK Government Digital Service) and April Rinne (Local & National Governments Consultant) to Name Two of the Six. If Anybody Knows How to Communicate the Potential Use Cases of a Project with Government Entities, It’s This Advisory Board. One of the Strongest Points for Consensus. The Roadmap Here Extends All the Way into 2023. The Milestones Described Are Completely Reliant on the Technology Being Adopted by Government Entities, but If That All Falls into Place, the Team Seem Very Focused on the Future. Investors Do Need to Understand That Consensus Is a Very Long Term Hold, If Consensus Is Successful in It’s Roadmap Goals, Investors Will Be Heavily Rewarded. Full List Of Positives Here Concerns - The Consensus Project Is Extremely Ambitious but If the Team Can Pull It off and Gain the Adoption They Are Aiming for, It’s a Game Changer for so Many Aspects of Governance. With the Target Audience Including Huge Parties Such as Government, Consensus Really Needs to Build a Product That Governments Will Want to Invest Resources Into. The Marketing for Consensus Hasn’t Been Very Visible. It’s Still Early Days for Consensus but for a Project with Such Ambitious Goals, Communicating with the Community Is Absolutely Essential to Build Trust and Confidence. We Could Not Find an Official Reddit. Their Twitter Account Has over 30k Followers, Tweets Are Often Full of Substance but Infrequent. We Just Don’t “Feel” The Community Presence Here. Full List Of Concerns Here Official Consensus Links - Website Medium Twitter Github

Consensus (SEN) Review & Analysis — Consensus ICO Review — Pick A Crypto

1

consensus-sen-review-analysis-consensus-ico-review-pick-a-crypto-17650809d3cc

2018-05-07

2018-05-07 09:26:17

https://medium.com/s/story/consensus-sen-review-analysis-consensus-ico-review-pick-a-crypto-17650809d3cc

false

483

null

Blockchain

blockchain

Blockchain

265,164

Pick A Crypto

Detailed & Unbiased Analysis On All Cryptocurrency Altcoins, Tokens & Assets. We Assess Positives, Concerns & ROI Potential On Each Asset.

28f403637a99

pickacrypto

151

1

20,181,104

null

0

null

0

null

2018-03-23

2018-03-23 00:19:35

2018-03-23

2018-03-23 00:19:36

1

false

en

2018-03-23

2018-03-23 00:19:36

2

17650b852e19

2.237736

0

null

4

Despite competition from different marketing platforms, even today email marketing remains as one of the leading platforms to reach the target audience. The key to impressive marketing tactics is crafting content which the target audience will love. However, a plethora of companies is incapable of speaking directly to a customer using a single email. The future of email lies in behavior based personalization which is possible with artificial intelligence. AI can figure detect elements that drive customer behavior, recommend content and develop data based on the information. AI is changing email marketing for the better and bringing a revolution in the marketing world. Focus on using AI with best emailing software to reach your marketing goals. 1) Capable of sending tailor messages to the customers AI is proving as a lethal tool in the arsenal of a marketer because it helps in effectively managing data. With the growing number of subscribers, your data will continue to grow, and it will become difficult to manage. However, using AI marketers can study customer behavior and send tailor message to every customer. Tailor messages resonate better with the target audience because they receive newsletters and messages based on their taste and preference. 2) Assists smart segmentation Smart segmentation is something which is overlooked by a plethora of marketers, and they segment the customers either based on their demographics, gender, age or geography. Smart segmentation will help you garner more open rate, conversion, and your campaign will witness a higher CTA. Furthermore, AI has made easy for the marketers to identify hidden details of the customers, using which you can efficiently segment the target audience. Today, AI has become a one-stop solution for email marketers looking to leave a lasting impact on the customers using the best emailing software. 3) Helps in choosing the right time to send the email Just because you have access to internet 24 x 7, you cannot trigger messages round the clock to the customers. As a marketer, you need to send the right message to the right customer at the right point of time to ensure customers read the messages. Using AI and predictive analysis, marketers can study the past behavior of the customer and generate useful insights about the same. The AI algorithms will empower marketers to determine the best time to send relevant messages to the customers. When messages are sent at the right time, the emails impact the audience in a different way. 4) Assists in recommending products to the customers Brands who send right product recommendation, often have a large and a loyal customer base. Sending product recommendations will improve the click-through rate and will build customer loyalty. Using AI, you can successfully communicate product recommendations to the customers. With the help of AI, marketers can efficiently analyze their online activity, browsing behavior and purchase history. The data is then used to boost customer engagement because the customers receive messages based on their interest. Therefore, using the best emailing software and AI, marketers can achieve the impossible and reach the inbox of their customers. How is AI Bringing a Revolution in Email Marketing? - Email It future of email lies in behavior based personalization which is possible with artificial intelligence. AI can figure…www.emailit.co

How is AI Bringing a Revolution in Email Marketing?

0

how-is-ai-bringing-a-revolution-in-email-marketing-17650b852e19

2018-03-23

2018-03-23 00:19:38

https://medium.com/s/story/how-is-ai-bringing-a-revolution-in-email-marketing-17650b852e19

false

540

null

Bestemailingsoftware

bestemailingsoftware

Bestemailingsoftware

0

Roshan Roy Jonah

On life, web, and everything else in between!

a8aad0785864

roshanroyjonah

11

203

20,181,104

null

0

null

0

e52cf94d98af

2018-04-05

2018-04-05 13:44:36

2018-04-05

2018-04-05 14:17:48

6

false

en

2018-04-07

2018-04-07 15:37:31

7

176a8480ab46

2.512264

27

0

[Credit: Source of this material is Machine Intelligence in Design Automation.]

5

Deep Learning with GPUs in the Cloud Develop Your Own Deep Learning App in the Cloud with Free GPUs [Credit: Source of this material is Machine Intelligence in Design Automation.] Machine Intelligence and deep learning technologies are advancing at a rapid pace. Claim to this fame is that it is bound to enable an unprecedented degree of automation in every walk of life. Design automation, a field that has been automating semiconductor design for decades, is playing catch up. Paripath has been using these technologies for a few years now and has decided to disseminate the information for a greater cause using blogs, opens source code, book and other collateral. In this article, I hope to initiate the readers (from engineers to executives) into deep learning by walking them step-by-step to develop an capacitance estimation app in the cloud. Install the Google Colaboratory app in chrome browser This step is as simple as launching your chrome browser and clicking this link to install chrome extension. This chrome extension is based on jupyter notebook, an open source web development environment for machine learning. Click on ‘ADD TO CHROME’ will install the development environment in your chrome browser. Install Google Colab App in Chrome Create/ Open the Colaboratory Notebook. Following URL contains Colaboratory notebook with open source code for estimating capacitance based on physical characteristics of a wire. Once you click on the link https://drive.google.com/file/d/1ctq4F28XMPLRHIp7qCep2-V8zLG7zRRm/view?usp=sharing, it opens up in your chrome browser as shown in the picture. Open Colab Notebook Click on Colaboratory link shown as arrow in the picture above. This will open the notebook in your browser. Run the App Once the notebook is open, you can run it by pressing “CTRL+F9” or clicking “Runtime -> Run all” as shown in the picture below” Colab Notebook with Code and Text Analyze Results Once you hit “CTRL+F9”, remote servers running on google farm will produce the results in no time. You can analyze the model results and tweak it for your use. Colab Notebook with results and graphs Use GPUs for free If your app is taking longer than usual, you can use GPUs for free by hitting “Runtime -> Change runtime type” as shown in the picture below: Using GPUs in Google Cloud Summary Now you’ve developed your first machine learning app in the cloud with GPU provided by google, consider yourself armed with all the information needed to deploy machine learning in your next application. Good luck ! Reference: Tensorflow: An open-source machine learning framework for everyone Book on Machine Intelligence in Design Automation Short Course on Machine Intelligence in EDA and CAD Open source machine learning apps for EDA and CAD

Develop Your Own Deep Learning App in the Cloud with Free GPUs

139

develop-your-own-deep-learning-app-in-the-cloud-with-free-gpus-176a8480ab46

2018-05-14

2018-05-14 10:23:47

https://medium.com/s/story/develop-your-own-deep-learning-app-in-the-cloud-with-free-gpus-176a8480ab46

false

414

A collection of technical articles published or curated by Google Cloud Platform Developer Advocates. The views expressed are those of the authors and don't necessarily reflect those of Google.

null

googlecloud

null

Google Cloud Platform - Community

null

google-cloud

GOOGLE CLOUD PLATFORM,DEVELOPERS,CLOUD COMPUTING,DEVOPS,TECHNOLOGY

gcpcloud

Google Cloud Platform

google-cloud-platform

Google Cloud Platform

4,042

Rohit Sharma

null

1c19e49fb995

srohit_94687

21

27

20,181,104

null

0

null

0

null

2017-11-18

2017-11-18 13:37:19

2017-11-18

2017-11-18 13:39:20

1

false

en

2017-11-18

2017-11-18 13:44:31

11

176b4b391224

2.150943

0

This article was originally published at: http://www.enricdurany.com/productivity-better-every-day/5-things-ive-learnt-this-week-w46/

4

5 things I’ve learnt this week — w46 This article was originally published at: http://www.enricdurany.com/productivity-better-every-day/5-things-ive-learnt-this-week-w46/ One of the ways I am embracing the “Be 1% better every day” philosophy is by sharing a weekly curated list of the best articles I have read in the last 7 days. The motivation behind it is twofold. On the one hand, it helps me reflect on the new ideas I am exposed to so that I understand them better. On the other, it helps reward all these authors that take some time to share their learnings with the world, making it a better place for everyone. Past articles in these series 5 things I have learnt this week, w45 5 things I have learnt this week, w44 This is the recap for the week. The Zebras movement: the life of a startup beyond going public or selling Source: Jennifer Brandel Topic: Venture Capital / Corporate Social Responsibility Reading time: ~10′ In the eyes of Venture Capitalists startups have only one measure of success: growth. This is a model that rewards quantity over quality, consumption over creation and exits over sustainable growth. The ¨Zebras movement” aims at empowering startups that, beyond growth, pursue profit but also purpose. Companies that aim at creating a more just and responsible society. Unlike Unicorns, Zebras are real. Uber’s new cultural norms: preserving what works while quickly changing what doesn’t Source: Dara Khosrowshahi Topic: Company culture Reading time: ~5′ Cultural norms are the basic pillars of organisational culture. They define the rules and behaviours that a company considers relevant, and represent an important lever to enable employee empowerment and alignment between attitudes and business goals. The new moats: Why Systems of Intelligence are the next defensible business model Source: Jerry Chen Topic: Strategy Reading time: ~15′ Economies of Scale or Network Effects represent the traditional moats that Tech companies have used to cement their position in an industry. As applications move to the cloud, are consumed on phones and personal assistants and fuelled by AI, the traditional way of building barriers to entry is being disrupted. Thus, Tech companies must look into building new moats to avoid becoming victims of their own success. Artificial intelligence is as prone to bias as the human kind. Here’s why. Source: McKinsey Insights Topic: AI / Machine Learning Reading time: ~20′ Algorithmic bias represents one of the ethical dilemmas in the adoption of artificial intelligence. Though Machine Learning (ML) promises to improve decision quality, some of these algorithms are prone to incorporating the biases of their human creators. Colours: data-driven colour palettes to use for inspiration Source: Klart.co Topic: UX Design Reading time: ~1′ Given my colour-blindness (at least in spirit), I have found this collection of colour palettes extremely useful for putting together presentations and documents that shine. Each palette provides with the HEX value of every colour, so it’s straightforward to incorporate them into one’s work. Have you come across any other articles you would like to recommend? Help the world be 1% better in the comments.

5 things I’ve learnt this week — w46

0

5-things-ive-learnt-this-week-w46-176b4b391224

2018-03-22

2018-03-22 03:14:43

https://medium.com/s/story/5-things-ive-learnt-this-week-w46-176b4b391224

false

517

null

Machine Learning

machine-learning

Machine Learning

51,320

Enric Durany

Product @Expedia. Based in London. Entrepreneur, previously co-founder @Vinarea. Writer @Hackernoon. Love building new products. Views are my own.

d5a3eecc9afc

edurany

434

29

20,181,104

null

0

null

0

null

2017-12-27

2017-12-27 09:01:04

2017-12-27

2017-12-27 09:20:07

1

false

en

2017-12-27

2017-12-27 09:48:59

1

176c5d95c2e8

1.015094

4

0

Im a doctor from a Developing Country, Indonesia. Here we celebrate a good and 80% standardize medical service at a fraction of developed…

5

Cheap Medical Consultation For The World Im a doctor from a Developing Country, Indonesia. Here we celebrate a good and 80% standardize medical service at a fraction of developed country cost. For one C-Section we only spend 50$ for the Obstetrician. And yet our survival mother during delivery reach up to 99.7% (200 maternal mortality for 100.000 delivery) compared with 99.99% (10 maternal mortality for 100.000 delivery) in developed country. Dokterchat — Ask The Experts We have decent doctor, with cheaper price. For a session of Chat Consultation and discussion with specialist in our platform here: Dokterchat will only cost you around 5$. Compared with Australian or Singaporean doctor at 110$ per-session, its a breakthrough. For today, our apps already gain some traction and revenue, we get into Top-20 medical apps on Indonesian Playstore. For now, we only serve Indonesian people with Indonesian Language. But if its succeed, we will serve globally, and scale our services. If we gain traction and investments we will collect the data and start our Artificial Intelligence to serve patient without doctors. The world, and Evidence Based know, at least 80% medical problem can be solved remotely, with telemedicine. And 50% of it can be solved with AI. Thats our believe, our goal. Cheap Medical Consultation for a healthy world. It starts today.

Cheap Medical Consultation For The World

40

cheap-medical-consultation-for-the-world-176c5d95c2e8

2018-04-02

2018-04-02 05:44:55

https://medium.com/s/story/cheap-medical-consultation-for-the-world-176c5d95c2e8

false

216

null

Maternal Health

maternal-health

Maternal Health

1,384

Dokter Chat

null

509251656344

dokterchatid

10

6

20,181,104

null

0

null

0

f3225cc85e15

2018-05-19

2018-05-19 14:10:41

2018-05-22

2018-05-22 15:05:08

2

false

en

2018-10-03

2018-10-03 02:29:33

32

176cd3a0009f

3.934277

195

1

2

Netflix CEO — Reed Hastings predicts 15B$ in revenue this year.

5

Netflix Data Science Interview Questions — Acing the AI Interview Netflix CEO — Reed Hastings predicts 15B$ in revenue this year. On May 9, Netflix launched its own research website. This highlights the focus Netflix has on Deep Learning and Data Science. The site is extremely well designed showing vertical classification of the different areas that Netflix research works on along with the horizontal business areas where Data Science is deployed at Netflix. It has some great articles with everything from video encoding to A/B testing where they use Data Science. I found the website to be very comprehensive making it a go to destination for things Netflix Data Science from different verticals to jobs. Source: https://research.netflix.com/ At Acing AI, the aim is to help you to get into Data Science and AI. I have profiled some of the best technology companies and written articles about AI interviews at Microsoft, Google, Amazon, LinkedIn, Ebay, Twitter, Walmart, Apple, Facebook, Zillow, Salesforce, Uber, Intel, Adobe Tesla and most recently IBM. This has led to being the top writer in Artificial Intelligence on Medium. The AI interview preparation guides Part 1, Part 2 go over the details which help you ace any AI interview. Acing AI Portfolios helps you to showcase your AI work. Expert interviews and analyses gives you a sneak peak into the lives of AI/Data Science Leaders and analyses of AI tech companies. Now onto the Netflix Data Science Questions article… To maximize the impact of their research, Netflix does not centralize research into a separate organization. Instead, they have many teams that pursue research in collaboration with business teams, engineering teams, and other researchers. From our publications we can deduce that they are focused on the applied side of the research spectrum, though they do pursue fundamental research and think that has the potential for high impact, such as improving our understanding of causality in our data and systems. Interview Process Netflix moves quite fast. There is one phone interview with the recruiter and another detailed one with the hiring manager. There are two onsite interviews with around 4 people first time (data scientists/engineers) and 3 people (higher level execs) second time. There is a mix of product, business, analytical and statistical questions. Statistical questions mostly revolve around A/B testing: hypothesis testing. There are a couple of SQL questions too. Analytical questions usually includes a hypothetical problem to analyze and metrics to evaluate product performance. Higher level executives mostly focus on background and past experience. Source: Netflix Tech Blog Important Reading Netflix Research Blog: All Articles Deep Learning for Recommender Systems: Talk Slides Reliable ML in the Wild Workshop (ICML 2017): Making ML Reliable at Netflix AI/Data Science Related Questions How would you build and test a metric to compare two user’s ranked lists of movie/tv show preferences?How best to select a representative sample of search queries from 5 million? Given a month’s worth of login data from Netflix such as account_id, device_id, and metadata concerning payments, how would you detect fraud? (identity theft, payment fraud, etc.) How would you handle NULLs when querying a data set? Are there any other ways? What is the use of regularization?What are the differences between L1 and L2 regularization, why don’t people use L0.5 regularization for instance? SQL queries to find time difference between two events given a certain condition. Given a single day with a large sample size and a significant test result, would you end the experiment? What do you know about A/B testing in the context of streaming? How do you prevent overfitting and complexity of a model? How do you measure and compare models? How do you know if one algorithm is better than other? Elaborate on the recent project you developed for your company. Why do you use XYZ method? Elaborate on how to improve content optimization? What technology or item that most people feel will be obsolete in the future do you not agree with? Why Rectified Linear Unit is a good activation function? How should we approach attribution modelling to measure marketing effectiveness? How would you determine if the price of a Netflix subscription is truly the deciding factor for a consumer? If Netflix is looking to expand its presence in Asia, what are some factors that you can use to evaluate the size of the Asia market, and what can Netflix do to capture this market? Say the CEO stops by your desk and asks you whether or not we should go into an untapped market. How would you determine the size of the addressable market and the factors the Netflix should consider before deciding to enter the market? Reflecting on the Questions The data around Netflix questions is sparse. The high level questions resolve around A/B testing, recommender systems and foundational knowledge questions around regularization and activation functions. This is different from the other companies we have looked at previously where focus was more foundational. All job openings are usually senior level. Good experience combined with good preparation can surely land you a job at the largest international evergreen content cinema in the world. Consumable List: Netflix Data Science Interview Questions This article was also featured on KDnuggets: https://www.kdnuggets.com/2018/06/netflix-data-science-interview-questions-acing-the-ai-interview.html Thanks for reading! 😊 If you enjoyed it, test how many times can you hit 👏 in 5 seconds. It’s great cardio for your fingers AND will help other people see the story. Subscribe to our Acing AI newsletter, I promise not to spam and its FREE! Acing AI Newsletter - Revue Acing AI Newsletter - Reducing the entropy in Data Science and AI. Aimed to help people get into AI and Data Science by…www.getrevue.co

Netflix Data Science Interview Questions — Acing the AI Interview

1,182

netflix-data-science-interview-questions-acing-the-ai-interview-176cd3a0009f

2018-10-03

2018-10-03 02:29:33

https://medium.com/s/story/netflix-data-science-interview-questions-acing-the-ai-interview-176cd3a0009f

false

941

Acing AI provides analysis of AI companies and ways to venture into them.

null

Acing-AI-203608690388104

null

Acing AI

acingai101@gmail.com

acing-ai

ARTIFICIAL INTELLIGENCE,DATA SCIENCE,MACHINE LEARNING,STARTUP,TECHNOLOGY

Acing_AI

Data Science

data-science

Data Science

33,617

Vimarsh Karbhari

Engineering Manager | Udacity Deep Learning & AI(part1) Alumnus | Editor/Founder of Acing AI

825fa70f9e24

vimarshk

2,009

730

20,181,104

null

0

null

0

null

2017-11-27

2017-11-27 15:08:43

2017-11-27

2017-11-27 15:22:25

1

false

en

2017-11-27

2017-11-27 15:22:25

0

176d292f3b35

2.875472

3

0

Artificial Intelligence (AI) is coming, and it’s going to be devastating for some people. Think about this: why wouldn’t the owner of a…

5

Fuzzy Data in our AI World Artificial Intelligence (AI) is coming, and it’s going to be devastating for some people. Think about this: why wouldn’t the owner of a trucking company replace every truck driver with a safer, cheaper, and more reliable AI commanding the wheel? In this case like so many others, there really isn’t an argument to be made to keep the humans, those individuals with families to support, children to feed and educate, mortgages and bills to pay to keep the roof up and the lights on. That’s what’s coming in the foreseeable future, and it’s scary. In some ways, it’s a typical technological disruption. The New York Times used to have a corps of typesetters, but technological innovation replaced them. The same happened with telephone operators, factory workers, and a lot of accountants. Algorithms replace algorithmic labor, as a rule. AI aims to transcend the typical paradigms of algorithmic activity and include what was until now a purely human factor: creativity. And it’s doing so by capturing and analyzing tremendous amounts of data. At it’s essence, it’s a very human thing to do. How else do we operate in our homes and our workplaces? We use our five physical senses, in varying capacities, to capture immense amounts of data, analyze that data using our brains, and then behave in a manner to cause an effect in our physical environment. That precisely becomes the role of AI in the fast-approaching future. The human has been broken down to his algorithmic form and has been rebuilt inside a silicon chip. From one perspective, all human actions ultimately exploit the muscles and the brain. We manipulate our physical world using our muscles and control the mode and magnitude of that manipulation with our brain. Technological revolutions to this point have largely revolved around replacing the need for human muscles. Bulldozers, welding robots, and sewing machines are examples of such disruption. Muscular dexterity has been trivialized in the past, but the resulting economies were dominated by the mentally dexterous. The economic potential of mental dexterity, the only thing still exclusively human, is now at risk of robotic appropriation; many production functions that depended on “thinking”, data analysis at its core, are liable to be replaced by stronger data analysis engines, AI equipped robots, muscle and brain. In this new order, which forms of mental dexterity are ripe for relegation to irrelevance? My theory boils down to classifying the type of data needed to perform the mental acrobatics associated with any job, any production function. First, there is clean data, data that is easily quantifiable and highly correlatable with clear conclusions. This type of data is easy to capture and can be used to make statistically significant statements with the use of a relatively low number of variables. Then there is fuzzy data. This data is difficult to capture quantitatively, and furthermore is not highly correlatable or statistically significant without the explicit analysis of an enormous number of variables. To look at physicians, a radiologist may be said to work with clean data, as he looks at quantifiable images. Quantified deviations from the healthy case can be diagnosed. A psychiatrist, on the other hand, works with fuzzy data. Every patient’s baseline healthy state is different, and deviations from the nebulous baseline tell the story of disease. The patient might lie or tell a half truth. A patient’s answer depends on the incredibly nuanced medium of human communication in which words, verbal tone, and bodily demeanor must be considered to make actionable conclusions. A soldier in a war zone may be working with clean data, in which the target and the enemy are clearly defined, but a security guard in a crowded mall works with fuzzy data, in which the target is tough to identify, and the mode and magnitude of action must be chosen based on a plethora of fuzzy variables. All of this means that, if you want to keep your job for the next 50 years, choose a production function which depends on fuzzy data. Or be that guy and create AI; be the disruptor instead of the disrupted. If you can’t do either of those, then be prepared to say hello to your new robot overlords.

Fuzzy Data in our AI World

44

fuzzy-data-in-our-ai-world-176d292f3b35

2017-11-27

2017-11-27 21:50:30

https://medium.com/s/story/fuzzy-data-in-our-ai-world-176d292f3b35

false

709

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Aashay Patel

Student of Medicine, Engineering, and Entrepreneurship.

c99db407b893

AashayPatel

33

20,181,104

null

0

null

0

null

2018-09-19

2018-09-19 15:41:46

2018-09-19

2018-09-19 16:51:14

2

false

zh-Hant

2018-09-19

2018-09-19 16:51:40

14

176fdb8ce87e

1.455031

1

0

2017年1月份，AlphaGo 在線上圍棋平台以”Master”身分大殺各方頂尖棋手，讓這個震驚四方的圍棋 AI 系統又再次得到了廣大的關注。在好奇心驅使，還有朋友討論的因緣際會之下，…

5

給初學者的 AlphaGo 機器學習導論 2017年1月份，AlphaGo 在線上圍棋平台以”Master”身分大殺各方頂尖棋手，讓這個震驚四方的圍棋 AI 系統又再次得到了廣大的關注。在好奇心驅使，還有朋友討論的因緣際會之下，研究這套強大強大強大的人工智慧系統。這篇文章旨在導論，用淺顯的語言說明整體架構，若有興趣可見文末分享更詳述的文章及原始論文！ 2017.1.8 筆過去在解決類似的棋局問題時，最直觀的方式就是將「所有可能性」展開成決策樹，計算出能致勝最多的路徑！然而在複雜度極高的圍棋上，若使用暴力演算，恐怕等到世界毀滅也展不開這決策樹。有些方法如 Monte Carlo, MiniMax , Alpha-beta Pruning 可以協助避免展開一些不必要的分枝，但仍無減那無限大的複雜度。因此，雖然幾1996年的 Deep Blue 靠著寫定的 Evaluation Function 及 Monte Carlo 展開法成功制霸西洋棋的決策樹，但複雜度極高的圍棋勢必需要新的方法來解決。 AlphaGo，可以說是融合了傳統搜索樹簡化法和現代最夯的深度學習（類神經網絡），類似架構其實早在20多年前就提出，但其中的訓練細節卻都是近期眾多學者的心血結晶。整個系統，由兩個類神經網路及一個搜索樹所構成， 1. Policy Network 2. Value Network 3. Monte Carlo tree search (MCTS) 1. Policy Network Policy Network 旨在預測出給定目前整個棋盤局面（state），下一步棋（action）最可能落在哪幾個位子。就目標而言，一個好的的 Policy Network 可以預測出一個圍棋強者的下棋策略！可能還沒到真正的「策略」那麼高竿，僅只是下一步棋的預測而已。數學上，這就是條件機率 = p(a|s) （這也是在機器學習中的要角：likelihood）至於怎麼訓練這個 Neural Network 讓他準確預測，用了兩個方法： I. Supervised Learning (SL) 搜集了各方高手的棋譜，丟進去當成一對對的資料＋標籤（label），聰明地用 Convolutional Neural Network (CNN) 來處理19*19的棋盤輸入資料，配上傳統的 Stochastic Gradient Descent 訓練法旨在預測最可能的下法（maximum likelihood）。無奈非常容易 Overfitting，也就是這個 NN 其實只學會完美面對看過的棋局，而無法演繹到其他任何未曾謀面的棋局。 II. Reinforcment Learning (RL) 這就是所謂「電腦自己對弈」的部分了！以上述 SL 的模型當作初始值，讓電腦再自己跟自己下棋，下到最後的對戰結果（贏了多少、輸了多少）隨著時間倒敘回去，按照比例分配給各步棋，當作梯度 (gradient) 來增強 (update) Neural Network！比喻來說，就像我們在玩超級馬力歐一樣，我們可能進到某個關卡要先在很多小地方做很多嘗試，有時失敗、有時成功，無論成功或失敗都可以當作借鏡，借鏡來發現先前經過的那些小地方應該怎麼做比較好！而後可能可以發現一條康莊大道，讓我們順暢地解決怪物、跳過溝渠，最後很順利輕鬆的過關並得到超高分 SL 讓 AlphaGo 嘗試站在巨人的肩上，而 RL 則進一步讓 AlphaGo 變成一個孜孜不倦的棋者，不停地自己跟自己下棋來預測一個常勝軍會怎麼下棋！這裡 RL 更協助解決了 Overfit 的問題。回到白話文，一個強大的 Policy Network，其實就是一位「模仿高手」的高手呢！甚至 Google 也承認了光靠這一個模型，沒有用到任何決策樹搜索或其他輔助，對戰”Pachi”（一個靠MCTS搜索圍棋AI程式）就拿到了超過85%的勝率！總之，這個機制最大的好處是一來可能可以猜測到對手的落子位置，二來可以縮小上述所說的決策樹展開範圍，因為變成只要集中計算那些「可能會下的位子」就好！ 2. Value Network 一個對棋局的靜態評分機制，也就是估計「某給定局面，最後的勝負目數」。要注意的是最後輸出值與 Policy Netwrok 不同，它並不是勝利機率，而是勝負目數。資料輸入端一樣使用 CNN，而結果輸出端為了因應實數區間，採用 Mean Squared Error 來作為「實際勝目與預測勝目」的 Loss Function。在訓練時，因為並沒有「勝目數與局面」的直接對照組， (事實上也不可能，因為是連續好幾個漸變的局面才會有對戰結果的勝目數) 因此採用前述的 Reinforcment Learning 方式，將對戰結果的勝目數一步步回溯至各個局面！更甚一步的是，Google 將圍棋賽局資料庫中的各個局面，拆開成3000多萬個局面，再讓電腦隨機地從這些起始點開始自己對弈！換句話說，當我們在跟 AlphaGo 對弈的時候，我們很可能是第一次看到當下的局面，就需要靠直覺、經驗累積來評估局勢，但這名孜孜不倦地機器學習者可能早已經對這個局面幾千萬回合，早已有一個量化的數字來代表當下局面了！（題外話：但「人類的直覺」才是更有效率、更強大的超猛本能）在這個類神經網路當中，電腦學會對靜態局面客觀地判斷，就可免去下子動作成千上萬種可能分支的龐大複雜度影響！有了上述一靜一動的兩個系統，分別給予動作與棋局不同的評分基準，最後一步就是結合這兩者找出最佳的棋路啦！在談到最後一步搜索前，要先提到其中一個早期 AI 就有重要方法：Q-learning （相信有涉略AI領域的人應該都不陌生，它就是制霸那個小精靈躲鬼吃金幣的套路）簡而言之它就是透過展開決策數展開接下來的可能狀態，而遞迴地從最終狀態回溯至當前局面！咦～說到這邊，有些人就會懷疑…這聽起來不就是暴力解法了嗎？其實中間還有一些機制來防止計算量或記憶體爆表的問題，也讓這個方法稍微更智慧一點，這邊簡略地介紹2種解決方法： I. 事先算好+存檔：某些比較常見或有代表性的局面會事先被計算好並存檔，因此並不需要一路展開到最終狀態，而是碰到存過檔的局面就可以直接回傳、結束遞迴！ II. 策略啟發法（Heuristic）：某些情勢局面是可以只看部分指標，就被快速地分類成量化指標！例如：小精靈如果離鬼越遠，離金幣越近，那麼它就越可能存活下來並得到高分！因此這「遠鬼而近金幣」法就是一個計算量少又佔小記憶體的好策略！還記得前述的 Policy Network (預估既有局面下做某動作的勝率)和 Value Network (某靜態局面所可能導致的勝負目數)嗎？這兩者就各自在 Q-learning 的時候立下了顯赫的功勞！我們可以想像 AlphaGo 今天面對一個特定局面時，要想辦法找出一步可以讓勝目數最大化的棋，一句話概括就是說這步棋同時是：讓 Policy Network 模仿的高手下該步棋機率最高，且接下來的局面被Value Network預測勝目數最高！聽起來很簡單吧！不過別忘了，在沒將全部可能分支展開之前，要怎麼能夠保證它就是「最容易致勝」的路呢？雖然 Neural Network（註:別忘了這邊的 Policy Network, Value Network都是由NN構成的）訓練完成後的預測速度頗快，但再怎麼快都還是不可能暴力展開，因此這邊要再提到 AlphaGo「選擇展開哪些分支」的能力，也就是我們的最後一步！ 3. Monte Carlo tree search (MCTS) 顧名思義它是來解決展開決策樹的問題，可以很直白地解釋成：「只展開看起來最有前途的幾個目標分支，再加上少許隨機地展開其他決策樹！」額外隨機地展開其他分支可以避免系統只在自身認定很好的範圍內搜尋最佳解，而忽略了自己認定範圍外也可能存在出乎意料的好解法！換句白話文，就是避免「自身的盲點」。此外，AlphaGo 還對最後的這個計算公式，加上了一項讓系統更傾向於「展開較少探索足跡的分支」的機制。因此它不僅知道要偶爾去嘗試一些自身預測之外的事情，還知道要平均計算能力，多算一算自己比較少走的路呢！回到 AlphaGo系統的綜觀，它所要找「讓 Policy Network 模仿的高手下該步棋機率最高，且接下來的局面被 Value Network 預測勝目數最高」的這步棋，只會在被這兩個 Network 綜合預測較高的分支中被展開，其他表現被預測成很差的分支就只能隨小小的機緣被探索了。 (註:此「機緣」即上段所說的嘗試自身預測之外以及多嘗試少走之路等等) 注意觀察這兩個指標，會發現 Policy Network 其實就是極其巧妙的一種策略啟發法(heuristic)呢！ (只著重展開高手們會下的動作分支，可說是真正令它站在前人下棋智慧的肩膀上) 紀錄到這裡，其實大致的架構已經闡釋完畢，剩下一些數學細節就省略不提，還有訓練過程中的一些眉眉角角，就是DeepMind強大的地方，再加上Google驚人計算能力的展現了！像是哪個天才想到把這些兜在一起，或是天殺的「單機系統」居然擁有48顆CPU、8顆GPU等等。延伸閱讀：淺談Alpha Go所涉及的深度學習技術 by Allan Yiin Mastering the game of Go with deep neural networks and tree search by David Silver et al.

給初學者的 AlphaGo 機器學習導論

50

給初學者的-alphago-機器學習導論-176fdb8ce87e

2018-09-19

2018-09-19 16:51:40

https://medium.com/s/story/給初學者的-alphago-機器學習導論-176fdb8ce87e

false

284

null

Machine Learning

machine-learning

Machine Learning

51,320

Angus Kung

Data science hacker with backpack on shoulders. https://anguskung.com/

61ca076f8562

anguskung

81

95

20,181,104

null

0

null

0

null

2017-12-11

2017-12-11 11:48:35

2017-12-11

2017-12-11 11:53:04

0

false

en

2017-12-11

2017-12-11 11:53:04

1

1770009c663a

0.45283

0

Headquartered in California,AppZen protected AI innovation joins PC vision, profound learning, and NLP to comprehend the full setting of…

2

Artificial Intelligence: the Answer to more Effective Business Compliance Headquartered in California,AppZen protected AI innovation joins PC vision, profound learning, and NLP to comprehend the full setting of costs, and cross-checks them progressively against a huge number of outer and social sources to decide whether they are true blue and precise. For some, enterprises, falling in with built up standards and controls is a continuous and habitually agonizing procedure, particularly when it includes going through the motions that are both obsolete and ease back to-change. Mean while, the need to keep up savvy inner consistence in everything — from acquirement systems to worker cost approaches — without hindering profitability displays its own arrangement of difficulties. Read more… https://goo.gl/8TVGPZ

Artificial Intelligence: the Answer to more Effective Business Compliance

0

artificial-intelligence-the-answer-to-more-effective-business-compliance-1770009c663a

2017-12-11

2017-12-11 11:53:05

https://medium.com/s/story/artificial-intelligence-the-answer-to-more-effective-business-compliance-1770009c663a

false

120

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Innovative Technology

Be the most beautiful version of yourself

1e8caae1c105

innovtech2710

3

11

20,181,104

null

0

null

0

1f55c9704605

2017-12-01

2017-12-01 12:10:37

2017-12-01

2017-12-01 17:25:02

5

false

en

2017-12-01

2017-12-01 17:25:02

1

177161dc82d

4.142767

8

0

— Which of these frying pans you will choose to cook pasta? — I don’t want a pan. I just want to eat, please show me the nearest Italian…

5

5 examples how AI can beat rule-based Facebook automation — Which of these frying pans you will choose to cook pasta? — I don’t want a pan. I just want to eat, please show me the nearest Italian restaurant! Rule-based Facebook automation systems assume that you as a marketer want something specific. They say: you might want to increase the budget by 10% every day OR you might want to turn off ad sets with CPI lower than 2$. It’s like saying you want to buy a frying pan to make a pasta at home when in fact you just want to eat. And going to a restaurant might be a better idea. The truth is that a marketer wants better campaign performance and wants marketing goals to be achieved with less effort. Why marketers should stop using rules to automate campaigns? You may think that creating rules is a simple way to manage campaign performance. In other words you ask your Facebook campaign to behave the way you put in the rule. Yes, it will. Unfortunately it does not mean it will result in better performance. Rules can’t take into account different metrics simultaneously and find the best balance between scaling and budget efficiency. Rules are based on our feelings mostly. Why to choose 10% budget increase, not 15%? We never have a strong argument about that. Rules need to be changed. What worked yesterday might not work tomorrow. And you will need to go and change the rule again. #1 Example — AI can estimate conversion rates better How do you feel when creating rules for a new campaign? You need to estimate lot’s of metrics based on the results of your first test launch. If your goal is to drive app installs it’s important to know your installs to clicks rate, CTR, the minimum number of clicks before the first install etc. You can use excel, your brain or go to the 3rd party dashboard. But the truth is that you almost never set good working rules from the first time. Because the reality will change. Tomorrow you will need to do the same work. Or you can let AI do this work for you and analyze your funnel and conversion rates in realtime. #2 Example — AI can better know how to compare results How many times you thought about choosing the right time period to compare ad sets performance? With AI you don’t have to thinks about time ranges anymore. It will automatically decide how to compare your ads and choose the best ones to scale using multifactorial models (when such factors as CPI, the number of installs per day, the number of days of statistics etc. are taken into account). #3 Example — AI can predict your future campaign performance Yes, many Facebook automation products use predictive analytics. They also have rule-based interfaces. They notify you about how your campaign will perform tomorrow, and you need to think about how to adjust the rules having this information. With AI you don’t need to reset your rules — it will adjust next steps (duplicating, budget increase etc.) automatically. The power of predictive analytics starts when it can be applied to your next steps in a couple of simple actions — like telling your virtual AI assistant which of them to implement right now. #4 Example — AI can tell you which metric now is more relevant to optimize You can set rules to manage your CPI. What if you have not enough actions to understand how to improve your campaign? How to decide when is the right time to change the metric? Usually you should go and calculate your funnel to set new rules to manage CPC or CPM. And you need much effort to change these rules each time. With AI you will be sure you optimize the right metric without manual calculation. Your virtual assistant will suggest the next action for the metric that will give you the maximum performance. #5 Example — AI can suggest how fast you can scale your campaign Let’s say you set a 20% budget increase rule when CPI < 2$. Why? You did this based on our feelings mostly. AI can help you understand how fast you can scale to balance between high volumes and an optimal cost per action. How marketers can benefit from AI in the next 3 years? Many of us still choose most popular solutions to automate Facebook ads. There are a lot of companies who solve this problem. Let’s think one step further. Facebook has already implemented it’s own automated rules which are pretty good comparing to 3rd party solutions. Now marketers need smarter AI to replace some rules and effectively manage the other. We will see new solutions and companies who will manage routine tasks and be able to analyze tons of data in seconds. AI will empower marketers to more fully utilize their uniquely human skills. That’s why we created Leadza.ai — a virtual assistant who can help better optimize campaigns by sending daily optimization tips through Facebook Messenger. Our mission is to simplify the decision making process for any marketer who faces the task of managing ads and to make this work enjoyable.

5 examples how AI can beat rule-based Facebook automation

112

5-examples-how-ai-can-beat-rule-based-facebook-automation-177161dc82d

2018-04-28

2018-04-28 05:00:11

https://medium.com/s/story/5-examples-how-ai-can-beat-rule-based-facebook-automation-177161dc82d

false

877

Stories from Leadza makers

null

victoria.fastovets

null

Leadza —virtual AI assistant for digital marketers

info@leadza.ai

leadza

FACEBOOK,CHATBOTS,FACEBOOK ADS

null

Marketing

marketing

Marketing

170,910

Victoria Fast

Founder&CEO at Leadza.ai — virtual AI assistant for digital marketers

c10209d5a4

rocknrollqueen

1,035

623

20,181,104

null

0

null

0

null

2018-01-18

2018-01-18 04:29:49

2018-02-23

2018-02-23 02:25:32

1

false

en

2018-04-19

2018-04-19 09:54:37

4

177218d62f8e

3.920755

0

In this week’s Entrepreneurs Feature, we speak to Dr Daniel, one of the co-founders of Apeiron Technology, which explores the infinite…

5

THE HANGAR by NUS Enterprise Entrepreneurs Feature — Apeiron Technology In this week’s Entrepreneurs Feature, we speak to Dr Daniel, one of the co-founders of Apeiron Technology, which explores the infinite possibilities in healthcare, developing sustainable solutions for everyone. Here’s a brief bio of Dr Daniel: The man has over 10 years of working experiences with Hewlett Packard, Okura Flexible Automation, Micron Semiconductor, and National Instruments. He has completed his Engineering PhD as well as a Bachelor’s degree (with 1st Class Honours no less!) from the National University of Singapore. 1. What is Apeiron and how did the idea behind it come about? Arun and I founded Apeiron Technology Pte Ltd in September 2015, but the roots of Apeiron Technology started back in 2006 when we were working for Hewlett-Packard Singapore. We continued to keep in touch with each other while pursuing their Bachelor and Doctorate degrees at NUS. Towards the end of our PhD study in 2015, we had a conversation with our PhD advisor Prof Tan, the third co-founder of Apeiron Technology, regarding our future after graduation. During the conversation, all of us realised that we shared a passion for technology and for solving big social challenges. By focusing on healthcare, we translated their passion into a vision that was outlined in business proposals shared with NUS Enterprise. The award of practicum grants was the first big endorsement of our ideas. We were enrolled into Lean LaunchPad (LLP) Singapore which was a 10-week programme piloted by NUS Enterprise to help research scientists and engineers turn their innovative technologies into commercially viable products and feasible business ventures. Through face-to-face interviews with healthcare professionals and management teams during LLP, we managed to validate a niche market need with the potential to scale globally: an in-patient fall prevention solution. In 2016, we received the Action Community for Entrepreneurship (ACE) Grant from Spring Singapore to continue the development of iMOS, the Intelligent Monitoring System, to reduce the number of inpatient falls in hospitals. After receiving the grant, we started working on a pilot project with an acute care hospital in Singapore. We are incubating with NUS Enterprise and are now focusing on completing the pilot project and getting ready to launch iMOS in Singapore. Apeiron Technology’s product, the Intelligent Monitoring Solution (iMOS), monitors patients in hospitals, care facilities or even at home. 2. Describe your typical day. The day is not regular. During some periods, my schedule can be totally full 24 hours a day and 7 day a week with meetings, development (business and technical), presentation, and pilot deployment. Those periods are similar to fire-fighting. I started the day with business/customer/team meetings. After that, I went to the hospital and collected new data. Following which, I went back to office to fetch the new data into our Artificial Intelligence (A.I.) engine and left the processing engine running. Then, I went home and experiment different parameters of our A.I. engine in my laptop. To monitor the A.I. engine, analyse results, and iterate A.I. with different parameters, I remotely accessed my office. After that, I slept for a couple of hours before the next day cycle began. In short, I was simultaneously running business development, pilot deployment, A.I. engine, and A.I. iteration. Fortunately, during some periods, I have extra bandwidth left for networking and strategising. These periods can be labeled as reflective periods. 3. How did Apeiron Technology build its customer base? We have used the direct approach to build our potential customer base. Our contacts were through hot introductions and cold callings. Our customer base was built just to validate market. I would like to highlight two mistakes here. 1) After validating the market, we focused on working with one of the hospitals to run a pilot and stopped communicating with other hospitals to focus on the pilot. After a year, we realised that we were indirectly giving unhealthy control to the hospital and the relationship had been off-balance. 2) It is important to have an advisor who is familiar with the healthcare landscape to assist in navigation and advise on its unwritten norms. 4. What are some of the greatest challenges in the course of your entrepreneurial journey and what did you learn from them? I have been working 12 years before my entrepreneurial journey. A big paradigm shift occurred. I give you three examples: you work to earn becomes you invest yourself to build a good business. Yearly timescale suddenly becomes in term of daily and weekly. Specialist becomes generalist. The challenge is how to comfortably live and enjoy in the new paradigm. 5. What is your favourite aspect of being an entrepreneur? 1) Learning and Mistakes. There are plenty to learn. Reversible mistakes are okay. Just be careful about irreversible mistakes such as regulation and relationships. 2) The self-discovery journey. You may face to your innermost insecurities and you have a good opportunity to learn how to deal with them. 3) Interesting diverse people. You will be communicating with people from all over the world, with various interests and personalities. The more you connect with them, the more interesting you will find them. 6. What line of work do you think you would be in if you weren’t an entrepreneur? I have been dreaming and preparing myself emotionally, intellectually, and financially over 8 years before I fully jump in. I have chosen Entrepreneurship as my career choice 8 years ago. 7. What are the benefits of being part of THE HANGAR by NUS Enterprise? There are many benefits. The top three benefits for me are 1) its launching platform/incubation space, 2) its credibility, and 3) its connections. Are you an entrepreneur working on a cool product and looking for more than just a co-working space ? Do apply for a hotdesk here and become part of the exciting community that we offer at BLOCK71 Singapore. For more great entrepreneur themed articles delivered weekly to your inbox, subscribe to the BLOCK71 newsletter here!

THE HANGAR by NUS Enterprise Entrepreneurs Feature — Apeiron Technology

0

the-hangar-by-nus-enterprise-entrepreneurs-feature-apeiron-technology-177218d62f8e

2018-04-19

2018-04-19 09:54:38

https://medium.com/s/story/the-hangar-by-nus-enterprise-entrepreneurs-feature-apeiron-technology-177218d62f8e

false

986

null

Startup

startup

Startup

331,914

NUS Enterprise

NUS Enterprise nurtures entrepreneurial talents with global mindsets, while advancing innovation and entrepreneurship at Asia’s leading university.

9c9afcb2f86c

NUSEnterprise

335

154

20,181,104

null

0

null

0

4f6a35510b05

2018-08-03

2018-08-03 15:00:25

2018-03-29

2018-03-29 08:56:56

4

false

en

2018-08-03

2018-08-03 15:23:47

2

17722625fc57

3.341509

7

0

“Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” If you’re in marketing, I’m pretty sure you’ve…

4

Artificial Intelligence Software: the New Intuition for Marketers? “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” If you’re in marketing, I’m pretty sure you’ve heard this sentence more than once. But its author, famous advertising pioneer John Wanamaker, lived at the turn of the 20th century, so it doesn’t have to be true anymore. New tools, such as artificial intelligence software, have completely changed the game. Why? What has changed? In the past, advertising (today we would say marketing) was all about creativity, intuition, research and… lots of whiskey, if “Mad Men” is anything to go by. Now it’s still a creative job but much more automated and data-oriented than is commonly thought. What about intuition? It’s not that simple. A study from Netflix shows that algorithms are much better than people in predicting which show will be watched next. Its authors called it the “intuition failure”. It’s worth paying attention to because typically a viewer loses interest after perhaps 60 to 90 seconds of trying to find a video to watch. At the same time, Netflix estimates its algorithms produce $1 billion a year in value from customer retention. “Tomorrow, there is no need to waste 50% of your advertising budget, because we will know who is interested in our messaging and we will pay them to listen. AI will tell us who we need to talk to, when and how with what message and information”, says Mark Mueller-Eberstein, CEO & Founder at Adgetec. You’ll agree that he’s right when you realize that 35% of Amazon.com’s revenue is generated by its recommendation engine. Another example is Target, the second-largest discount store retailer in the US. After focusing on analyzing customer data their revenues grew from $44 billion to $67 billion. All of that wouldn’t be possible without machine learning algorithms. What’s important, you don’t have to be a technological giant like Amazon or Netflix to take advantage of them. All you need to do is to integrate AI-driven software with your service. But what exactly can you do with artificial intelligence software? Predictive analysis Imagine you have an e-commerce business and you’re able to predict if someone who has entered your website is willing to buy, or what and when he or she will buy. Sounds like magic, doesn’t it? But it’s real and possible, thanks to machine learning algorithms. Prediction mechanisms work when you collect customer data and use to continuously create segments. Algorithms recognize repetitions in behavioral patterns of particular customer segments and are able to predict the most probable behavior based on previous similar scenarios. At Synerise, we’re working hard on them, and our Predict module is accurate up to 95%. Dynamic content Another exciting application of machine learning in marketing is dynamic content, commonly used in newsletters, landing pages and on the websites. It allows you to create one campaign with detaills that change automatically according to predefined conditions, including: location language demographic data interests past behavior of the client Dynamic content gets results. According to a report by the Annuitas Group, leads that are nurtured with targeted content increase sales opportunities by up to 20%. Chatbots Have you heard of chatbots? Very often when you enter a company’s website, you will see a little chat icon encouraging you to ask a question or start a conversation. Usually, there is a sales rep on the other side, but more and more often it’s… a bot. And if it’s well designed, you may not even recognize it. Bots can work 24 hours a day, seven days a week, during Christmas Eve and other holidays. They can contact an enormous number of clients at the same time. They’re also connected to the database, CRM and get all the information necessary to solve the problem efficiently. Wrapping up All the applications mentioned above are just the beginning. Marketers use image recognition software, smart display campaigns, churn prevention, customer lifetime value forecasting, and more. Marketing has become predictable. Intuition is a desirable quality in a marketer, but what can be better than effective tools to support it? Originally published at synerise.com on March 29, 2018.

Artificial Intelligence Software: the New Intuition for Marketers?

9

artificial-intelligence-software-the-new-intuition-for-marketers-17722625fc57

2018-08-03

2018-08-03 15:23:47

https://medium.com/s/story/artificial-intelligence-software-the-new-intuition-for-marketers-17722625fc57

false

700

AI Growth Cloud

null

synerise

null

Synerise

marketing@synerise.com

synerise

ARTIFICIAL INTELLIGENCE,DATA SCIENCE,MARKETING TECHNOLOGY,MARKETING AUTOMATION,OMNICHANNEL

synerise

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Monika Ambrozowicz

null

95ccccab825a

m.ambrozowicz

1

4

20,181,104

null

0

null

0

1f35b6f451e8

2018-02-21

2018-02-21 13:54:34

2018-02-21

2018-02-21 13:56:33

1

false

en

2018-05-12

2018-05-12 16:46:22

3

177330a620ad

0.562264

1

0

1

Python:

5

Daily webography for 3 dummies to make it in machine learning — Act 1, Scene 2 Python: Python has been used in artificial intelligence projects. As a scripting language with modular architecture, simple syntax and rich text processing tools, Python is often used for natural language processing. Tracé de courbes - CoursPython Pour tracer des courbes, Python n'est pas suffisant et nous avons besoin des bibliothèques NumPy et matplotlib…www.courspython.com Learn R, Python & Data Science Online | DataCamp Learn Data Science from the comfort of your browser, at your own pace with DataCamp's video tutorials & coding…www.datacamp.com

Daily webography for 3 dummies to make it in machine learning — Act 1, Scene 2

1

data-artificial-intelligence-2-177330a620ad

2018-05-12

2018-05-12 16:46:23

https://medium.com/s/story/data-artificial-intelligence-2-177330a620ad

false

96

We offer contract management to address your aquisition needs: structuring, negotiating and executing simple agreements for future equity transactions. Because startups willing to impact the world should have access to the best ressources to handle their transactions fast & SAFE.

null

ethercourt

null

Ethercourt Machine Learning

adoucoure@dr.com

ethercourt

INNOVATION,JUSTICE,PARTNERSHIPS,BLOCKCHAIN,DEEP LEARNING

ethercourt

Data

data

Data

20,245

WELTARE Strategies

WELTARE Strategies is a #startup studio raising #seed $ for #sustainability | #intrapreneurship as culture, #integrity as value, @neohack22 as Managing Partner

9fad63202573

WELTAREStrategies

196

209

20,181,104

null

0

null

0

null

2018-06-11

2018-06-11 10:38:50

2018-06-11

2018-06-11 10:43:33

0

false

en

2018-06-11

2018-06-11 10:43:33

1

1773946b74cb

1.766038

12

0

I’ve never once worried about my online security. Sure, I could have a clever password strategy, set up VPN, encrypt my hard drive or tweak…

4

Beginners guide to cyber security I’ve never once worried about my online security. Sure, I could have a clever password strategy, set up VPN, encrypt my hard drive or tweak my browser settings, but as a result I’d create a joyless and pessimistic world-view. And who would want to hack a children’s book author anyways? So what happens when someone who thinks she has nothing to hide gets hacked? My work is about explaining the world of computer science to children in fun ways. For that cyber security offers a colorful playground with concepts such as honeypots, trojan horses, firewalls and script kiddies? Throughout the six episodes of the documentary I spoke with experts to learn about what lies behind the jargon. I got to visit a security center in Poland, see in practice how machine learning can help detect threats and learn how the landscape of security is changing. But to experience the other side of the equation, I gave permission to the F-Secure team to try to hack me. The rules were simple: use a vulnerability, break in and do something. One of the things that became obvious pretty soon was that this attack was not going to happen with any clever algorithm or brute force, but through social engineering. The team would take advantage of the tiny everyday chores, habits and clues I’ve sprinkled everywhere online and use them to break in. For the first few days I was suspicious of everything. From e-mail alerts notifying me of Squarespace domains getting old to doxxing attack warnings, from phone service confirmations to blinking mobile screens everything screamed scam. But there were just too many things to pay attention to, and auto-pilot kicks in easily. What happened? Check out the full documentary here. After the documentary I did resolve to make changes in the way I protect my privacy and security online. But even more importantly I think I learned the same lesson as Alice in Wonderland, who after Lewis Carroll plunged her on the adventure thought “after such a fall as this, I shall think nothing of tumbling down stairs!” There is no way you can protect yourself entirely online, especially as an organisation. What happens after the attack is what matters. An organization where employees don’t deny, panic or hide attacks is much more likely to pull through. A strategy for cyber security is as much about implementing the right hardware and software as it is about the right practices, culture and communication. One more word on the cyber security people, who in my experience are among the most creative, curious and persistent people I’ve met. I think it’s worth redefining the way we talk about security for only their sake. Not rigid, resistant. Not pessimistic, persistent. Not paranoid, paying attention.

Beginners guide to cyber security

37

beginners-guide-to-cyber-security-1773946b74cb

2018-06-21

2018-06-21 10:18:12

https://medium.com/s/story/beginners-guide-to-cyber-security-1773946b74cb

false

468

null

Cybersecurity

cybersecurity

Cybersecurity

24,500

Linda Liukas

I like shiny things and software. Childrens book author at http://t.co/BHa0N4JzUW. Co-founder of http://t.co/u9jfb7qnFB. @Codecademy alumni.

f5534290c491

lindaliukas

5,645

432

20,181,104

null

0

null

0

null

2018-05-10

2018-05-10 22:56:41

2018-05-10

2018-05-10 23:06:43

3

false

en

2018-05-16

2018-05-16 20:15:44

1

17747904cef0

6.689623

5

0

One of Hollywood’s oldest tropes is the robotic usurpation of humankind. It makes for compelling thrillers and science fiction. The variety…

5

No, Robots will Not Take Over and Rule the Planet … and this is why. One of Hollywood’s oldest tropes is the robotic usurpation of humankind. It makes for compelling thrillers and science fiction. The variety of narrative twists possible are in-numerable. Terminator, a luminary of the genre, implanted this fear as efficiently as Jaws did in cultivating a phobia of sharks into an entire generation. As narratives go, everything else pales in comparison — a cautionary tale regarding one’s lust for power which inevitably capsizes when it surmounts its peak. The theme is so pervasive that we have come to accept it as fait accompli. Only further fomenting the all but inevitable Decepticon annihilation into our minds, tech gurus have also sounded the alarm. Elon Musk, a deserved legend of our time, famous for introducing commercially available autonomous vehicles, has been a particularly pious member of the faith, with comments such as, “the risk of something seriously dangerous happening is in the five-year timeframe. 10 years at the most.” (Edge.com). But even our journalists with earned and deserved respect have joined the bandwagon. Nick Bilton, for instance, opined, “the upheavals can escalate quickly and become scarier and even cataclysmic.” In the context of a Sci-Fi thriller, these pronouncements make sense. However, under the constraints of reality, the risk of Artificial Intelligence (AI) starting a massive global war is highly unlikely (and, also a lack-luster narrative). To understand this, the question to answer is the, “why?” As in, why would AI wish to conquer us? Thirst for power is prosaic, as power is only a symptom of a much more profound desire. It is only when the discussion is taken to its core, that we begin to shed light on the problems with this simple thesis. Consider this, to what end would robots want to conquer the planet? Understand, that machines and life-forms exist under different constraints, thus to extrapolate life’s need for domination as equitable with machines’ would lead our conclusions astray. Let’s start with biological life. All organic life forms (herein “organics”) are united by one common purpose, which is immortality that is achieved via reproduction. This is common across all animal and plant species. True Immortality does not exist for organics, and so all organics are united with the singular and universal core purpose of successful reproduction that is the base construct of their normal being. Taken one step further, not only do they wish to reproduce, but even more so, to protect their reproductive outcome is to ensure its own future immortality. Different organics achieve immortality via various means. Plants, for instance, spore thousands upon thousands of seeds at wild abandon into the world, hoping that a small percentage will successfully germinate. Plants utilize volume, odds, and statistics. Animals that live in very hostile environments also do the same, they produce large quantities of offspring, understanding that the majority of their brood will fail. And then there are others, such as mammals, that produce far fewer progeny, but will fight to the death to ensure their survival. Living in the harsh realities of survival that we romanticize as a harmonious nature, animal breeds coalesce in communities, wherein members gain strength in numbers. Together they ward off predatorial threats to life from organics within and without their species. Or, they merely decrease their odds of falling prey by swimming in pools, running in herds or flying in flocks. But a robot does not face this reality. Robots, by definition, are immortal from the outset. As long as energy is available and they avoid injuries, malware, viruses, molten lava, EMP’s and short-circuits, etc. a machine will live indefinitely. Robots do not need to consume other robots, plants, fruits or humans to survive. They merely need power, and as we move further into the future, it becomes more and more likely that they can become entirely powered by solar energy, thus making their energy source infinite. Unlike organics, machines have absolutely zero need to reproduce to achieve immortality. A commonality of reproduction is not there. If anything, any replica a machine creates will only become its competitor. Being that machines have no natural predators (they cannot be eaten) and do not compete for natural resources for sustenance (sunlight is infinite and readily available), community formation for foraging, shelter, defense, etc., is unnecessary. A machine’s only real competitor would be other machines, not humans. All in, a machine’s needs are entirely different than that of organics. Organics need to eat and pass excrement, while machines do not. Organics need shelter for defense against the weather and predators. Machines do not. If an organic does not reproduce, it will necessarily fail to exist eternally. Machines, on the other hand, by default live forever. For organics, different species are threats to one another because they are competing for the same limited resources. Machines would not compete for the same resources as their only real resource, the sun, is unlimited. Organics form communities to defend and contend with all the above. Because machines have none of these limitations, the formation of communities would not be a default social order for them. Which takes us one step further into the other fear regarding robots — that all our jobs will be gone. Understand, that economics is not the study of money but that of trade. Money is simply the most efficient tool for trade, and therefore misunderstood as the analog to it. Machines would be unable to do all our jobs since it would imply that there would be nobody for the machines to partake in trade with. Without trade, there is no economy. Without an economy, there is no job for a machine to perform. As the economies of physical labor become cheaper with robots, the value of service-based labor by humans increases. Thus, explaining the frothing of massage establishments all over the country. Once considered a luxury service, now it’s a dime a dozen. This fear for the worst is rooted in organics. Until only recently, survival was something humans fought for by the minute. If left untended, children risked being consumed or taken into servitude. Viruses and bacteria were life-threatening. Still, for the rest of the animal kingdom, their every waking minute revolves around eating and avoiding being eaten. That constant fear for one’s survival is a MAJOR stress that humans, only in the last century, have overcome. But our bodies are wired for that level of constant stress. It is expected. And so, our children of today, growing up without real stressors, have overactive stress responses resulting in anxiety disorders. And for the rest of us, wired to be continually aware of impending doom, we create our own boogeymen to allay our body’s natural needs for a life-threatening stress. In come the machines, apes, birds, spiders and all other versions of mutiny and Armageddon that provide us an outlet to direct those energies. The natural state of life is to be aware of many boogeymen. Civilization has rid us of them, and so it is only natural for us to gravitate to any such boogeyman we can conjure. One last thing to note, and this one far more abstract, but I enjoy the philosophical discussion around it: The Fermi Paradox. Named after Enrico Fermi, the Fermi paradox details the apparent contradiction between the lack of any evidence of and the high probability estimates for the existence of extraterrestrial civilizations. According to Fermi, if the universe is billions, if not trillions of years old, and we accept that there mathematically must be life elsewhere, than by simple derivation there should be civilizations that are millions of years more advanced than we are. If we can assume that in a million years human civilization will have conquered space travel and inter-galactic travel, the simple question then is, why have we not been visited by those civilizations that have already done so? Where is the evidence of them? There are three possible answers to this question. First, there is no other life throughout the universe. Second, inter-galactic space travel is impossible. Or, third, all advanced civilizations extinguish themselves. For discussion purposes, I will assume it’s the second or third options. These conclusions would also pose a challenge to the thesis that machines overtaking the world is inevitable. Because, while inter-galactic space travel for organics may be impossible, that would not be the case for machines that can survive in space indefinitely. If machines would have an a priori natural drive to dominate, that would not be limited to their own planets but naturally across the universe. And if machines taking over is inevitable, then numerous planets have already befallen this fate, as that would require that all advanced organic civilizations to be naturally overtaken by their own machines. If it’s a natural and obvious conclusion, it would happen universally. It would be far more possible for multiple cadres of machines venture out in all directions, accepting that it may take millions of years to reach their destinations. The machines would lie dormant and when sunlight activates them they could reactivate and enter dominate their appropriate targets. In fact, the only natural predator of a machine would be machines from another advanced civilization. And so conquering space travel, would in fact constitute the only purpose for a machine to absolutely exhaust all resources to conquer inter-galactic travel. The question then is — where are the extra-terrestrial machines? If you like what you read, please ‘clap’ so that others may happen upon this essay. Allen is the founder of Eusoh, a novel community support platform. His musings focus on reflections on life, culture, philosophy and raising able children.

No, Robots will Not Take Over and Rule the Planet … and this is why.

154

no-robots-will-not-take-over-and-rule-the-planet-and-this-is-why-17747904cef0

2018-05-21

2018-05-21 19:40:45

https://medium.com/s/story/no-robots-will-not-take-over-and-rule-the-planet-and-this-is-why-17747904cef0

false

1,627

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Allen Kamrava

Daddy and Hubby. In my spare time, also the founder of Eusoh.com, a physician & a surgeon. Enjoy triathlons, hiking, road trips, involvement & embracing life.

52bf9148d718

akamrava

45

20,181,104

null

0

null

0

null

2018-03-27

2018-03-27 06:37:58

2018-04-01

2018-04-01 07:52:47

1

false

en

2018-04-01

2018-04-01 07:52:47

1

17763b97f8b

0.758491

0

Artificial Intelligence is here to stay. That’s what we hear on the news. And I agree with that statement. As time passes technology…

2

Artificial Intelligence: An International Playground — Infographic Artificial Intelligence is here to stay. That’s what we hear on the news. And I agree with that statement. As time passes technology evolves and gives companies real competitive advantages on the market. According to studies, the investments in AI exploded in the past year. In 2016 there were $640 millions invested in this technology, and the predictions for 2025 are surprising: $37 billion. We’re expecting this year to see more and more companies becoming AI adopters as they understand that AI has a positive impact on multiple variables in the value chain. Moreover, tech companies started to created developments hubs around the world to fulfill the business requirements. In the infographic below I gathered some essential information and facts about the development of AI and how it is used by companies to leverage on the market. See the full infographic here Originally published at rickscloud.com.

Artificial Intelligence: An International Playground — Infographic

0

artificial-intelligence-an-international-playground-infographic-17763b97f8b

2018-04-01

2018-04-01 07:52:48

https://medium.com/s/story/artificial-intelligence-an-international-playground-infographic-17763b97f8b

false

148

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Rick’s Cloud

null

fd3840f389e2

contact_40643

8

16

20,181,104

null

0

null

0

null

2018-07-26

2018-07-26 10:07:57

2018-07-26

2018-07-26 10:10:37

1

false

en

2018-07-26

2018-07-26 10:10:37

4

1777031e1a1e

0.603774

0

Listen to the full show here.

4

Why outsourcing memory to technology helps you learn — Morgan Craft Listen to the full show here. Morgan Craft is CTO at Hickory. Managing a diverse team of psychologists, data-scientists, and machine learning experts sound like a daunting prospect. Not for Morgan. He has 14+ years of development experience under his belt and knowledge that ranges from psychology to coding. Morgan is also a firm believer in continuous learning and understands the value of acting as a mentor to help others grow professionally. Morgan joined Andy to discuss disrupting the education sector, making sure staff remember training, and launching startups. Listen to the full show here.

Why outsourcing memory to technology helps you learn — Morgan Craft

0

why-outsourcing-memory-to-technology-helps-you-learn-morgan-craft-1777031e1a1e

2018-07-26

2018-07-26 10:10:38

https://medium.com/s/story/why-outsourcing-memory-to-technology-helps-you-learn-morgan-craft-1777031e1a1e

false

107

null

Machine Learning

machine-learning

Machine Learning

51,320

Venturi's Voice

Blogging and podcasting about all things tech. We speak to thought leaders in a variety of fields to get their insights into leadership, management & staffing.

fca5e5c66f79

venturimarketingliam

14

175

20,181,104

null

0

null

0

380662b6a201

2018-09-14

2018-09-14 15:18:22

2018-09-14

2018-09-14 15:20:00

1

false

en

2018-09-24

2018-09-24 15:13:46

14

1778375dd3a1

10.701887

0

So, you’ve got it. The blockchain is the future. Nations will dissolve, companies will perish. But you want to aim for the sky, right? Now…

5

Chapter 10: THE WINNER TAKES IT ALL So, you’ve got it. The blockchain is the future. Nations will dissolve, companies will perish. But you want to aim for the sky, right? Now what? Ok, let me point you in the right direction. Let me show you where you find the golden path towards the glorious world to come. Following his new book on digital strategy How To Become A Digital Marketing Hero, Rufus Lidman is now translating all his knowledge to the most revolutionary field in the world. A field in desperate need of a digital strategy. In a series of articles, we will get an upclose view of this new mind-blowing market, seen through the eyes of a digital strategist. >> Chapter 1: The Revolutionary Technology That Will Change Your Life >> Chapter 2: The Birth Of The Future >> Chapter 3: A Real Revolution In Practice >> Chapter 4: A Metamorphosis You Are Not Prepared For >> Chapter 5: The Apocalypse Of The Establishment >> Chapter 6: The Secret Recipe To Reach The Absolute Top >> Chapter 7: The Simple Answer To The Global Problems >> Chapter 8: Crypto Hits Rock Bottom — But Is It Really That Bad? >> Chapter 9: Let’s Innovate — Not Regulate The sky is the limit, that’s where you should aim. If you want to be a winner. Photo by Alex Antoniadison Unsplash In our past chapters we have gained insight into the fundamental goals of the blockchain. But also the means with which to achieve them during a second wave. A few concrete conditions to enable this have been made apparent along the way. And this regardless if we’re talking about a geopolitical, corporate or individual scale. The Great World Division On a national level, we see gigantic differences worldwide. On the one hand, we have a great number of countries trembling, doping their economy with artificial life support and banning as well as regulating the gutter kids, who are about to reveal the emperor without clothes. On the other hand, we have a growing number of countries who see their shot at creating optimal conditions for the new wave of blockchain companies, and who therefore are attracting both corporations and investments on a large scale. So in a single stroke, the world is being divided into first and third world countries of the new era. And it’s not necessarily the first world countries of old that are leading the game. The more I look at it, the clearer it becomes that my native Sweden is ending up in the third world category, despite its history of being quite avant garde in many senses. Why is that? Well firstly, cries for regulations and warnings about the new technology are commonplace. The financial authorities are issuing alarms about ICO’s. And the Governor of the Swedish Central Bank are alerting cryptocurrency investors that they are ”on their own”. And this has since then been formally presented in a gigantic regulation initiative, where the old dinosaurs have closed ranks for a collective attack. The G20-countries are agreeing to regulate (while at the same time implying accomodations for) the cryptocurrency, using the Financial Action Task Force (FATF) to apply for crypto-assets. The Financial Stability Board (FSB) Chairman Mark Carney simultaneously states that crypto-assets “do not pose risks to global financial stability at this time”. And the G20-countries have also acknowledged that the “technological innovation underlying cryptocurrencies has the potential to improve the efficiency and inclusiveness of the financial system and the economy more broadly”. All of this is of course a net good. It’s only the direction of energy that perhaps isn’t as positive. While Sweden and the G20-countries are going down the path of warning and regulating, the interest among professional investors and stock markets worldwide have exploded this past year. There are now hundreds of hedge funds investing in BTC. For historic reasons, Asia is particularly well suited to benefit from the added capital. 70 percent of the bitcoin mining pools in existence are still based in China, which has gone from yuan to ye. Half of all global BTC trade on the other hand is based in Japan — the first country in the world to approve cryptocurrencies as a legal currency. Also becoming the home of the first employers in the world to offer salaries in crypto. South Korea isn’t far behind either, together with Vietnam. Both countries with very high prevalence of internet connectivity. But even more are on the move. As we’ve seen in the first chapter, it’s not the ones who are in the forefront of the digital technologies that are the first to take initiative in creating opportunities and solutions of old problems. Countries such as Mongolia are in the starting pits, Dubai is launching blockchain technology as the standardized platform for government documentation and Malaysia have already begun regulating for cryptocurrency transactions. Malta is looking to become “the blockchain island,” with the world’s largest trading space for cryptocurrencies. And the foremost biggest crypto hub in the world, aside from Singapore, has landed in Europe. The lovely little Swiss village of Zug is now renown as “The crypto valley”. Many serious contenders in the field, including us at AIAR, have chosen the latter to open offices and legal departments. Finally, even some of the bigger and well-established western countries are realizing that it’s time to hitch their wagon to this ride, if they want any fighting chance in the near future. The first EU member country to excel in the area is Spain (leaving France in a mild state of panic). It is soon to follow with taxation laws to incentivize ICO investments and offer benefit to companies dedicated to these. So chin-chin to the country where I lived and studied as a child, and where my company opened its first international office. Spain is here fast becoming “The Blockbuster of Europe”, and Barcelona specifically “The Silicon Valley of Europe”. It’s hard not to be impressed. As mentioned, right now Spain is taking some heavy action. As the first country in the EU, the government is preparing a new legislation, incentivizing digital entrepreneurs raising capital via ICO’s. Utility tokens will, under certain conditions, be able to be separately audited from traditional currencies. Spain has gotten the message. Tech startups are the only ones that truly drive development forward. And tech companies of the blockchain will be the true rulers. That’s why a whole barrage of actions is being undertaken to attract the companies of the future. The aim seems to be that the second wave of the blockchain is to be set up in Spain. Or as the deputy prime minister said: “We want to set up Europe’s safest framework to invest in ICO’s”. So, aside from the blockchain having the capacity to seriously save the world. And aside from Spain being the key to having Europe go from generating unicorns to actually generating the world’s next digital gorilla. Viva España! Barca, te amo! And Spain, here we come. (And yes, this is precisely why AIAR chose to set up our first development office here.) So we have countries that are terrified of change and stuffing their centralized heads in the proverbial sands. Or, even more aggressively, start shouting off the rooftops for regulation and bans. Then we have the above mentioned countries that embrace change, and do their utmost to create ideal conditions for companies and investors in the field to truly embrace the revolution to come. Guess which countries have the leg up in long-term success. And guess how many of the analogue nations will survive in the increasingly digital world. Three Types of Companies The above mentioned dichotomy of nations is frighteningly mirrored also among companies and the corporate world. Here, however, we find three different levels of division Firstly, we have all the old analogue companies. I said somewhat provocatively earlier that they’ll cease to exist, and that I still hold true. If you don’t get a move on and change precisely everything your operations are built on, then you will cease to exist alright. This became abundantly clear when I attended the post-Davos event at the start of the year, visited by, among others, the Crown Princess of Sweden, former Swedish head of state Carl Bildt and Google HQ. Despite all of these celebrities and an amazing theme, it wasn’t just clear, it was embarrassingly obvious how far behind the curve the traditional mega-corporations are. “If we don’t transform, we die!” Some CEO’s for the biggest companies in the old analogue economy tried to persist and emphasize how they are specifically not dinosaurs of a long forgotten age. One of the biggest banks of Europe even had the wherewithal to make the lucid observation “if we don’t transform we die!” …While it, at the same time, became the first bank in northern Europe to ban its employees from buying crypto. Then the world’s most amazing entrepreneurial group, the Norrsken Foundation, showed everyone how things should be done. A divine session, mixing both old and new, with ABC (AI, Blockchain, Crypto) technology being a major part of it. They demonstrated how much the old establishment should be aware of what’s to come, and how the new technologies will come whether we like it or not. It made me proud of this, our entrepreneurial generation! And this is a sentiment echoed throughout the world. We have on the one hand the old establishment sitting in a living room caught on fire, denying the flames licking their bootheels. Or doing some small side project to paint the image of still being in the loop. No names mentioned, but this is the way it is thus far for 99 percent of them. Then we have those who are awake. Companies like IBM or Maersk. Or Samsung. It is almost as if it’s written in the stars that it was my former client for over a decade and one of Asia’s and the world’s biggest companies, that was first on the ball to produce the microchips that do all the mining. As a secondary factor, we have the digital gorillas, that already during the first generation of the internet assumed global domination from the old analogue companies. They are now known as FAANG (Facebook, Apple, Amazon, Netflix and Google), representing 50 percent (!) of the market cap of the Nasdaq 100 Index. All of these operate out of an extremely centralized model. All of which somehow, when nations haven’t kept up with laws and regulations, managed to acquire an oligopoly, if not a monopoly, on their respective markets. With one or two of these gorillas ruling their field vertically. Google and Baidu within search engines, Amazon and Alibaba on eCom, Facebook and Snapchat on social media, YouTube and Netflix on streaming services and Appstore and Google Play in apps. Almost all of which could be argued are approaching outright ”evil-minded” practices of exploiting the centralized data they’re acquiring in the process. And doing this in a way that doesn’t in any manner benefit the consumer, but exclusively their own wallets and those of their shareholders. Facebook’s data gets ”leaked” to political organizations with the aim of influencing political elections and smear campaigns. Google uses their considerable sway to take down any and all competitors. Apple aggressively avoids taxation by any means necessary. Even Amazon spends nearly 14 billion USD on analyzing customer data to manipulate communications and offers in the right direction. If you analyze these companies from a blockchain perspective, two absolute truths become abundantly clear: 1. These are companies that stand to lose everything with blockchain technology. 2. They will spend insurmountable amounts of money, time and effort to stay alive. Most of these companies already have massive investments in the first of the ABC- technologies, AI. They are coming around on the second, blockchain, and whatever they may say now, they will launch their own cryptocurrencies to command the third. Remember where you heard it first. So be afraid. Be very afraid. Or simply, stay wide awake. Luckily, there are a few that stand to stop the digital dinosaurs from doing just that. Those are the governments, legislature and federal agencies. Right? Not very likely, I have to say. Here, if anywhere, we find old-fashioned centralists with blunted tools and slow processes. But we got GDPR, you might claim? Yes, we got the GDPR. Ten years after it was needed. And now that it’s finally here, it’s so toothless that even teenage hobbyists know how to circumvent it. If we analyze society at large today from a more sociological power perspective, the power today is consolidated in companies and their (active) consumers only. Leaving politicians and the rest of us (passive) citizens behind. This is similarly reflected in traditional media and their passive base audience, losing power to social media and their active participants. If we are to seek comfort in anything, but the vague hope that these dinosaurs would for some inexplicable reason switch from evil to ethical, we have to turn to other companies. The new digital ventures. The ones deep in deep-tech of ABC. Companies with a brand new perspective and a good portion of bravery, fighting spirit and yes, outright balls to change the world. These companies need, on an aggregated, level no strategy at all. They are, as a collective force, so incredibly strong that they’ll succeed anyway. However, on a more individual level, whatever strategy they adopt, they may falter and fail. Those who systematically dig through stacks and stacks of whitepapers, might find crypto companies consisting of four, five ukrainian programmers that assembled some fancy code, but comprising zero business insight and without any well-packaged products nor strategy to speak of. We find old companies that try to boost their profile by changing their name, slapping on the erc20 standard of ethereum to their businesses and call themselves ”blockchain companies”. We find the outright frauds and charlatans, who have put together something that sounds great on paper, but upon closer inspection is all about getting a quick buck. We also find the pure idealists that have fought the good fight offline and now hope that the blockchain will be kinder to them to finance their good fight online. To the worldly business strategist, it’s a wonder that so many of these actually got funded for the first wave. What’s even more beautiful is that in the middle of all this chaos, we find the awesome business ideas. In finance, in healthcare, in infrastructure and in data. And yes, also in my own area of education technology, where there so far isn’t much serious competition to speak of. All with great ideas. Some with some neat code. Very few with a sensible strategy. This is said to be the root cause to why we last year ”only” saw a 60 percent success rate for ICO’s. Those who have some experience in equity know that those rates would be considered incredible within traditional venture capital, where one in ten investments pull through to finance the rest. From that perspective, 60 percent is actually too much. But it was what to expect from the first wave. Now, as the insane novelty has worn off, investors are getting less and less naive and demand more auditing and verification. They require more thorough analysis and clearer strategies, and only extremely compliant blockchain ventures will be able to slip past. This is good news though. I see it as a healthy sorting out of the weeds from the stock. And this will help attract the really big money and give birth to the major players on the market. And believe me, they’re coming. As they should. Because the pure blockchain companies that are emerging in the second wave, they will be combining their revolutionary force with a sound basis of analysis and awesome strategy. And many of them really do have the potential to not just make a difference, but to actually change the world completely. And you can bet your ass that ICO investors will be making a massive amount of money investing in them. The Way For You To Go On an individual level, there is no contest. If you want a shot at competing in the marketplaces of today and tomorrow, you can only go in one direction. Seek out countries with a future. Seek out companies that live and breathe future. Get the competence you need to compete in a global company. And please understand, this whole competency revolution is directly based on the ABC and deep tech in general, and revolutionary applications of blockchain in particular. If you haven’t understood that much, you haven’t understood anything at all.

Chapter 10: THE WINNER TAKES IT ALL

0

chapter-10-the-winner-takes-it-all-1778375dd3a1

2018-09-24

2018-09-24 15:13:46

https://medium.com/s/story/chapter-10-the-winner-takes-it-all-1778375dd3a1

false

2,783

AIAR is democratizing education for the 2.5 billion people in need worldwide. By using new technology to innovate totally new processes, instead of repairing old ones, AIAR is providing the world's first mobile ecosystem for lifelong learning with 360° content scope.

null

aiar.daring.to.change.the.world

null

AIAR

info@medium.com

aiar

ICO,BLOCKCHAIN,EDUCATION,TOKEN SALE,AI

aiar_ab

Digital Strategy

digital-strategy

Digital Strategy

1,856

Rufus Lidman

CEO AIAR 165 countries, AI-learning, blockchain & utility token, entrepreneur in 10 ventures, 300 lectures, 5 books, founder IAB, Digital Advisor WFA, Fil. Lic.

5542ab9ec82f

blockchainboss

147

146

20,181,104

null

0

null

0

null

2018-01-13

2018-01-13 08:36:00

2018-01-13

2018-01-13 08:54:41

1

false

en

2018-01-13

2018-01-13 08:54:41

8

1778895f5c

2.65283

0

AI or artificial intelligence powered Sarah, the first humanoid robot to get Saudi Arabian citizenship or the first Robocop put into active…

5

An Overview of Artificial Intelligence AI or artificial intelligence powered Sarah, the first humanoid robot to get Saudi Arabian citizenship or the first Robocop put into active service by the UAE government are just couple examples hinting at an AI powered future. Driverless cars, automated homes, popular mobile assistants such as Siri or Cortana or the humble AI powered keyboard on your phone are a few examples emphasizing the extent and reach of AI in our everyday lives. AI is slowly but surely becoming part of everyday lives of humans and are fast evolving to become part of different industries. So, what is AI? Artificial Intelligence is the ability of computers or machines to mimic the thought process of humans to comprehend and make conclusive decisions replicating a human. Britannica defines AI as the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings. The term is frequently applied to the project of developing systems endowed with the intellectual processes characteristic of humans, such as the ability to reason, discover meaning, generalize, or learn from past-experience. Why is AI a buzzword for laymen, businesses or administrators? AI has become a buzzword for all because of the possibilities it presents. In terms of laymen, AI is key to leading a comfortable life powered by convenience. Convenience of automation where all home appliances are AI powered and can make decisions like how a human would? For example, the garage door at home opening-up only for your driverless car when it comes in. The lights going dim in your drawing-room when you are watching a movie on your home theater or the volume on your home theater decreasing when your phone rings, the possibilities are endless. The possibilities in case of businesses is also limitless. Imagine AI helping your business to decipher huge amounts of data to give you exact details you need to improve production efficiency, zero in on customers who have been giving you business on days when your buseinsses income is low, products that are sitting on shelves for long etc. Another use case for AI is in customer service. AI powered chatbots is taking the business world by storm. To learn about the advantages chatbots provide your customer service efforts read here. This is not all, there are hundred more processes or solutions that can be initiated making use AI which will provide businesses cost effective and time saving benefits. In relation to government, AI can be the technology that can help governments make conclusive and confident decisions. Deloitte’s research states that cognitive technologies can help the Public Sector free up billions of labour hours per year, to be spent doing real work, not drudge work. Artificial intelligence capabilities such as Natural Language Processing can ensure that machines understand the spoken word or text to automate tasks like translation, interactive dialogue and sentiment analysis. What’ in store? The increasing influence of AI is already impacting our lives in ways we are unaware of. With growing acceptancy of the technology and increase in use the future will throw up more and more instances of machines learning to immolate humans. While some renowned names like Elon Musk from the tech industry have expressed concerns over the growing influence of AI, suggesting the field requires regulations to curtail the adverse impacts. There are others like Microsoft Founder Bill Gates who strongly believe in AI and think that some of the pressing problems in healthcare and government administration can be resolved and better managed with AI. It will be premature to state which direction AI goes and whether the predictions of Musk or that of Gates stands the test of time? One thing is for sure the possibilities with AI are limitless and can enhance human life, business functions and governments administration, saving human effort and man hours, driven by efficiency and quality.

An Overview of Artificial Intelligence

0

an-overview-of-artificial-intelligence-1778895f5c

2018-01-13

2018-01-13 08:54:42

https://medium.com/s/story/an-overview-of-artificial-intelligence-1778895f5c

false

650

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Searchicas.com

Search, Content & Social Media Made Simple!

8cbf0510bf9

searchicas

11

16

20,181,104

null

0

null

0

null

2018-01-15

2018-01-15 14:35:15

2018-01-15

2018-01-15 14:37:53

1

false

it

2018-01-15

2018-01-15 15:05:52

1

17794dfeb0b8

1.996226

0

I chatBots, strumenti di intelligenza pronti ad innescare un paradigma ed un nuovo modello di business.

5

ChatBots come strumenti di AI e scenari di business. I chatBots, strumenti di intelligenza pronti ad innescare un paradigma ed un nuovo modello di business. Per prima cosa dobbiamo capire di che cosa stiamo parlando e nella fatti specie valutare che cosa sono e come funzionano i ChatBot. I ChatBot non sono altro che programmi in grado di rispondere a precise domande, in particolare si tratta di software capaci di riprodurre, a tutti gli effetti, una conversazione. Sebbene molte persone e molte aziende, sono ancora lontane da questo logica e presumibilmente non ne capiscono l’aspetto funzionale, ci sono sempre piu’ risvolti di tipo operativo, che ci inducono a pensare che i sistemi di ChatBot saranno sempre piu’ diffisusi. CHE COSA E’ Il chat bot lo si potrebbe immaginare come un assistente capace di rispondere a tutta una serie ben definita di quesiti, capace di dialogare con i consumatori esattamente come farebbe un assistente in carne ed ossa, in qualsiasi momento della giornata, sempre ed ovunque. Ma la forza del ChatBot è anche quella di espandere il proprio database, attraverso l’apprendimento automatico diventando sempre piu’ adattivo, quindi fornendo a richiesta, informazioni sempre piu’ specifiche e dettagliate e sempre piu’ in conformità con le richeste degli utenti. Mentre che fine fanno i dati raccolti dalle ChatBots ? Semplice, saranno dati sempre piu’ importanti, che andranno ad calmierare data base, lato Merchant, permettendo a quest’ultimi di avere informazioni e fornire offerte sempre piu’ mirate ed in linea con le richieste della clientela. In questo senso non è sbagliato dire che il ChatBot è a tutti gli effetti uno strumento di intelligenza artificiale, attraverso il quale, nei px anni, non solo i semplici utilizzatori ma anche gli addetti al business si interfacceranno. L’IDEA DI BUSINESS: Localflow Ora prendiamo in esame, la possibilità di utilizzare una piattaforma di messaggistica ed integrare in essa un chatBot in grado di essere fortemente focalizzato sulle esigenze di piccole e medie imprese e relative catena di fornitori, ma capace anche di incontrare le esigenze di individui, che offrono servizi all’interno delle loro comunità, il tutto senza la necessità di utilizzare un intermediario affidabile, perchè attraverso il ChatBot è garantita la correttzza dell’informazione e il flusso di Dati. E che direste ora se questa offerta di servizi fosse localizzata, cioè rispettasse parametri di offerta distinta all’interno di una area geografica ben definita, il piu’ possibile granulare, guidata dalla comunità e gestita da intelligenza artificiale, in cui fosse addirittura possibile effettuare micropagamenti a zero fee, sfruttando tecnologie ibride Ethereum/IOTA ? Grazie all’intelligenza artificiale brillantemente organizzata da sistemi Chat Bots ed integrata in applicazioni di messaggistica diretta, nasce su tutti il progetto Localflow. Localflow è quello che tutti stavano aspettando: un sistema fortemente incentrato sulla valorizzazione decentralizzata dei microservizi personali, delle aziende e della comunità, basato su ricerche locali e sempre piu’ granulari. Stay in Touch. @andreabelvedere www.bitconio.net

ChatBots come strumenti di AI e scenari di business.

0

chatbots-come-strumenti-di-ai-scenari-di-business-17794dfeb0b8

2018-01-15

2018-01-15 15:05:54

https://medium.com/s/story/chatbots-come-strumenti-di-ai-scenari-di-business-17794dfeb0b8

false

476

null

Chatbots

chatbots

Chatbots

15,820

Am Belvedere

Economia, politica, finanza, startup, bitcoiner, ambiente. Ricercatore

bd38681267c2

AndreaBelvedere

61

129

20,181,104

null

0

null

0

70f217fc23a8

2017-09-14

2017-09-14 16:54:35

2017-09-14

2017-09-14 17:02:38

3

true

en

2017-09-14

2017-09-14 17:02:38

1

177d64af60e3

4.20283

7

0

iFlytek’s voice recognition technology is everywhere in China, and that’s what’s making it smarter every day.

3

Why 500 Million People in China Are Talking to This AI iFlytek’s voice recognition technology is everywhere in China, and that’s what’s making it smarter every day. People watch a humanoid smart robot by IFLYTEK dancing during the 2016 International E-business Expo on March 30, 2016 in Guangzhou, China — Zhong Zhi/Getty Images By Yiting Sun iFlytek’s voice recognition technology is everywhere in China, and that’s what’s making it smarter every day. When Gang Xu, a 46-year-old Beijing resident, needs to communicate with his Canadian tenant about rent payments or electricity bills, he opens an app called iFlytek Input in his smartphone and taps an icon that looks like a microphone, and then begins talking. The software turns his Chinese verbal messages into English text messages, and sends them to the Canadian tenant. It also translates the tenant’s English text messages into Chinese ones, creating a seamless cycle of bilingual conversation. In China, over 500 million people use iFlytek Input to overcome obstacles in communication such as the one Xu faces. Some also use it to send text messages through voice commands while driving, or to communicate with a speaker of another Chinese dialect. The app was developed by iFlytek, a Chinese AI company that applies deep learning in a range of fields such as speech recognition, natural-language processing, machine translation, and data mining. Court systems use its voice-recognition technology to transcribe lengthy proceedings; business call centers use its voice synthesis technology to generate automated replies; and Didi, a popular Chinese ride-hailing app, also uses iFlytek’s technology to broadcast orders to drivers. But while some impressive progress in voice recognition and instant translation has enabled Xu to talk with his Canadian tenant, language understanding and translation for machines remains an incredibly challenging task. Xu recalls a misunderstanding when he tried to ask his tenant when he would get off work to come sign the lease renewal. But the text message sent by the app was “What time do you go to work today?” In retrospect, he figures that it was probably because of the wording of his question: you’ll work until what time today? “Sometimes, depending on the context, I can’t get my meaning across,” says Xu, who still depends on it for communication. Xu’s story highlights why it’s so important for a company like iFlytek to gather as much data from real-world interactions as possible. The app, which is free, has been collecting that data since it launched in 2010. iFlytek’s developer platform, called iFlytek Open Platform, provides voice-based AI technologies to over 400,000 developers in various industries such as smart home and mobile Internet. The company is valued at 80 billion yuan ($12 billion), and has international ambitions, including a subsidiary in the U.S. and an effort to expand into languages other than Chinese. Meanwhile, the company is changing the way many industries such as driving, health care, and education interact with their users in China. In August, iFlytek launched a voice assistant for drivers called Xiaofeiyu (Little Flying Fish). To ensure safe driving, it has no screen and no buttons. Once connected to the Internet and the driver’s smartphone, it can place calls, play music, look for directions, and search for restaurants through voice commands. Unlike voice assistants intended for homes, Xiaofeiyu was designed to recognize voices in a noisy environment. Min Chu, the vice president of AISpeech, another Chinese company working on voice-based human-computer interaction technologies, says voice assistants for drivers are in some ways more promising than smart speakers and virtual assistants embedded in smartphones. When the driver’s eyes and hands are occupied, it makes more sense to rely on voice commands. In addition, once drivers become used to getting things done using their voice, the assistant can also become a content provider, recommending entertainment options instead of passively handling requests. This way, a new business model will evolve. In the health-care industry, although artificial intelligence has the potential to reduce costs and improve patient outcomes, many hospitals are reluctant to take the plunge for fear of disrupting an already strained system that has few doctors but lots of patients. At the Anhui Provincial Hospital, which is testing a number of trials using AI, voice-based technologies are transforming many aspects of its service. Ten voice assistants in the shape of a robot girl use iFlytek’s technology to greet visitors in the lobby of the outpatient department and offer relief for overworked receptionists. Patients can tell the voice assistant what their symptoms are, and then find out which department can help. Based on the data collected by the hospital since June, the voice assistant directed patients to the right department 84 percent of the time. Doctors at the hospital are also using iFlytek to dictate a patient’s vital signs, medications taken, and other bits of information into a mobile app, which then turns everything into written records. The app uses voice print technology as a signature system that cannot be falsified. The app is collecting data that will improve its algorithms over time. Although voice-based AI techniques are becoming more useful in different scenarios, one fundamental challenge remains: machines do not understand the answers they generate, says Xiaojun Wan, a professor at Peking University who does research in natural-language processing. The AI responds to voice queries by searching for a relevant answer in the vast amount of data it was fed, but it has no real understanding of what it says. In other words, the natural-language processing technology that powers today’s voice assistants is based on a set of rigid rules, resulting in the kind of misunderstanding Xu went through. Changing the way machines process language will help companies create voice-based AI devices that will become an integral part of our daily life. “Whoever makes a breakthrough in natural-language processing will enjoy an edge in the market,” says Chu. © 2017 MIT Technology Review

Why 500 Million People in China Are Talking to This AI

90

why-500-million-people-in-china-are-talking-to-this-ai-177d64af60e3

2018-08-25

2018-08-25 01:41:54

https://medium.com/s/story/why-500-million-people-in-china-are-talking-to-this-ai-177d64af60e3

false

968

MIT Technology Review

null

technologyreview

null

MIT Technology Review

null

mit-technology-review

TECHNOLOGY,TECH,ARTIFICIAL INTELLIGENCE

techreview

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

MIT Technology Review

Reporting on important technologies and innovators since 1899

defe73a9b0ba

MITTechReview

23,166

1

20,181,104

null

0

null

0

7f60cf5620c9

2018-03-03

2018-03-03 08:38:49

2018-03-06

2018-03-06 15:38:31

2

false

en

2018-07-06

2018-07-06 08:59:42

9

177d6849820e

2.851258

4

0

From this series:

5

Speeding up your code (4): in-time compilation with Numba From this series: The example of the mean shift clustering in Poincaré ball space Vectorizing the loops with Numpy Batches and multithreading In-time compilation with Numba (this post) In the previous posts we worked with our minds in order to speed up a (relatively) simple algorithm. Probably there are other smart ways to squeeze up more execution time, but nothing very interesting came to my mind. So, it’s time to pass to the brute force. But still, I will avoid to use GPUs or TPUs. This because, as I already shown in this other post, often you loose so much time in moving the data from the system memory to the GPU (or TPU) that, at the end, the whole process will result as slower. But there are other ways of exploiting the brute force, i.e. of improving the results without squeezing too much our brains. One of them is Numba. It’s a tool that takes our function and “compile” them, which means that it translate them in a low level and code which (roughly speaking) speak the same language of the CPU. And so it’s faster. Numba at work Asking Numba to compile a function is easy as writing four letters before the definition of the function, “@jit”: The ‘at’ symbol in Python is known as ‘function decorator’. They tell the interpreter to alter the functionality of the defined function. In our case, the decorator tell the interpreter to modify the function following the instructions of Numba. You can find more informations about such niceties here. Actually Numba has lots of very useful tools within. One for all: it can do an automatic parallelization of the code. Of course, an ‘automatic’ tool is in general unable to deal with every kind of possible parallelization in a function, so don’t expect miracles from it. Since we already managed to manually parallelize the code, we will not use this functionality. So we will just have to add the decorator before the function definitions? Almost. Since Numba is still in high development, not all the Numpy functions that we use are supported. And so, we have to make some minor modification. In particular, the np.tile function is still not supported. An easy workaround is to create a void array of the desired shape, and fill it with the copies of the vectors that we need: You can see that we feed the decorator with two parameters: The first one, nopython=True is telling Numba to actually compile the function. It may seem redundant, and actually in our case it is. But in the cases where Numba is unable to compile the function, it will be executed in the usual Python interpreted way, and so we will loose the functionality we want to achieve. By adding this parameter, Numba will raise an error in such cases, and so we can dig in to understand where it falls. So it is a good practice to add it anyway. The second one, nogil=True tells the Python interpreter to release the Global Interpreter Lock. A description of the GIL functionality goes beyond the scope of this post, but in short: if we keep it with the compiled code, we loose the parallelization, so we have to release it. The rest of the code is exactly as in the previous post , with the only exception of placing the decorator with the two arguments, @jit(nopython=True, nogil=True) in the other two called function (__shift, gaussian). Note that the main one, meanshift_parallel, cannot be compiled in the way it is built. Performance improvement Now let’s switch to the numbers. How much speed we have gained? Well, we actually gained execution time. But there is a saturation effect: with the growing number of vectors, the two execution speed approach more and more. This is the final post from this series. Thank you for reading!

Speeding up your code (4): in-time compilation with Numba

14

speeding-up-your-code-4-in-time-compilation-with-numba-177d6849820e

2018-07-08

2018-07-08 16:40:45

https://medium.com/s/story/speeding-up-your-code-4-in-time-compilation-with-numba-177d6849820e

false

654

Sharing concepts, ideas, and codes.

towardsdatascience.com

towardsdatascience

null

Towards Data Science

null

towards-data-science

DATA SCIENCE,MACHINE LEARNING,ARTIFICIAL INTELLIGENCE,BIG DATA,ANALYTICS

TDataScience

Programming

programming

Programming

80,554

Vincenzo Lavorini

PhD in High Energy Physics, lover of Machine Learning

f0c2f36e0bc7

vincenzo.lavorini

67

5

20,181,104

null

0

null

0

null

2018-04-18

2018-04-18 02:47:35

2018-04-18

2018-04-18 04:47:21

1

false

zh-Hant

2018-04-18

2018-04-18 05:52:09

1

177de0b2180

0.407547

0

「只要他真的一開始運作，就沒有人會叫它AI了…」

4

英國上議會報告：AI與社會「只要他真的一開始運作，就沒有人會叫它AI了…」你早上醒來，覺得精神奕奕。因為你的手機鬧鐘剛好設定在7:06分，這是手機分析你的睡眠週期後，發現最適合叫醒你的時間點。你問語音助理今天發生什麼新聞，於是他按照你的興趣，列出一串自動篩選出的新聞。新聞中，一位議員正在為自己辯護，在新聞影片中，他在私下場合攻訐同黨主席。但議員辯稱影片中自己的臉是移花接木上去的，新聞中又出現幾位專家討論影片的真實性。當你看完新聞要上班時，你的小孩正在用手機App準備考試，那個App有個AI可以針對個別學生的學習狀況，量身打造適合他的個人課程。上班路上，車上的儀表板顯示最新的交通資訊，並且同時考慮現在的路況和之前的駕駛習慣，估計到辦公室的時間。到辦公室後，你打開email，email已經按照內容，為你自動分門別類。一位同事寄給你好幾個內容繁瑣的法律文件，而電腦已經自動幫你標示重點並且幫你摘出最重要的資訊，好讓你等一下在會議中使用。你打開另一個email，內容來自於你的另一半，他問你是否能借他的銀行登入帳號，他想要查一下東西。你又看了一下，覺得這封信應該是詐騙，但刪除前，你又猶豫了一下，忖著為什麼垃圾信可以那麼準確的抓到他的說話方式。但你有其他事情需要掛心，你還需要去醫院看醫生。在醫院照了胸部X光後，你很驚訝的發現醫生馬上就能和你解釋，X光片顯示有輕微的肺部感染。以前做完檢查後，看報告都要等幾個禮拜的。很快又有其他事進來了，你的手機上出現一個訊息：你的銀行帳戶有異常活動，但被銀行自動發現且禁止了。你打電話給銀行，一個叫做莎拉的行員接起來，並幫助你換一張新卡。你隨後才發現，這位莎拉並不是真人。你覺得好像有點不自在，你怎麼沒有在當下隨即發現。但無論如何，事情解決了，是不是真人好像不是重要的事。在回家路上，你繞到當地超市。架上的貨品已經按照過去的顧客要求、消費流行和當天天氣，全部自動篩選排列好。在路上，你的車子偵測到你有些煩躁，於是自動播放了一些能讓你放鬆的音樂。晚餐後，你和另一半欣賞了一部由電視自動推薦的電影。你赫然發現，雖然你和另一半平常觀影興趣截然不同，但這部影片卻對得上你們兩個人的胃口。在你漸漸入眠後，你的房子預測你們應該已經進入睡眠，它關掉浴室燈、打開洗衣機，準備新的一天。 (chapter “As soon as it works, no one calls it AI anymore …”) 英國上議會特別委員會發表了一篇與AI有關的報告，標題是「AI：英國準備、願意、能夠嗎？」（AI in the UK: ready, willing and able?）內容涵蓋當下AI與社會發展的各個層面：包含當前AI的大眾論述、資料可及性和控制、AI的明慧性（intelligibility）、AI在政府、業界、學界的發展和投資、AI對於勞動市場、教育、長照的影響，以及AI在法律和軍事上的應用和隱憂。最後，委員會提出5項AI原則，期待在英國的AI研究單位（不管是學界或業界）能自主考量AI發展所引發的社會倫理意涵。上述的引文，來自於報告一開頭的AI烏托邦願景。有趣的是，這個烏托邦一點也不虛幻，大部分是現在已經有應用雛形，甚至已經進入我們日常生活的技術（例如e-mail自動分類、新聞影片篩選等）。我們太習以為常，以至於我們根本不會認為它叫AI。報告中深入闡釋並反思，AI進入人類社會的影響將是鋪天蓋地的。科技物對人類社會各層面的影響，已經不是個別企業、個人或組織能夠承擔的責任與範圍，所以政府必須積極挺身而出作為管理者。委員會特別提出五個AI守則（AI code）： AI應該為人類的共同福祉和利益發展。 AI應該以明慧性（intelligibility）和公平性作為運作原則。 AI不應該限縮個人、家庭、社群的資料權和隱私。所有公民應有教育權，使其能與AI一起在心理、情緒、和經濟上共生共榮。所有具備自主能力傷害、毀滅或欺騙人類的AI都不應發展。（paragraph 417）第一點看起來很口號性，但鑒於最近的「劍橋分析」（Cambridge Analytica）事件，我們實在很難區分AI或機器學習的目的究竟是不是「善」或「只是為善」。第三點提到資料的問題，畢竟現在的深度網絡非常的「資料飢渴」（data hungry），要取得夠多資料、且又要符合個人層次的應用，很容易扞格個人的資料隱私權和資料所有權。第四點講到教育，比較像是提醒如果未來AI將是社會的一份子，我們至少要確保每個公民都是同等在AI時代被賦權（empowered）。最後一項，可能是不太有爭議但非常難實行的，自動的戰爭武器是戰略上非常誘人的技術，特別是當他國用自動無人機飛在自家門口時，我們很難不考慮使用或發展自動的戰爭機器。這五點中，最有趣，可能也是最難的問題是第二點的明慧性（intelligibility）。這是現在AI各研究領域熱衷討論的議題，但同時也是非常難以捉摸且爭議性很大的概念。在報告中，委員會認為：我們相信要讓AI整合進社會且成為可信任的工具，明慧的AI系統是基本的條件。明慧性（intelligibility）可能來自於技術透明性（technical transparency）或解釋性（explanability），但我們相信解釋性可能會是對一般公民或消費者較有用的取徑。這個取徑也同時反映在歐盟和英國的法規上。我們認為如果AI無法對自己做出的決定提出令人滿意的解釋，那AI就不應該應用在對個人生活產生重大影響的場域。亦即，像是深度神經網絡，這種模型目前無法解釋如何形成決策，那我們就應暫緩其應用，直到人們找到其他替代方案。（paragraph105）不管我們願不願意，AI已經來了，只是在此同時AI的圖像愈益模糊。我們該擔憂的，不是iRobot的VIKI或是魔鬼終結者的SkyNet；某種程度上，這種強AI已經具備自我覺識，已經都可以跟人吵架，所以人已經知道互動對象及其溝通方式，問題已然不大。真正的問題來自於，當人們無意識地接受AI，且很自動的相信AI無害的作為工具所下的決定。然而，在人類歷史上，我們應該不曾有過會幫忙下決定的「工具」。有些決定具備某種行為道德後果，但工具卻無法被賦予道德責任。如果，人們可以維持行為後果的道德警覺，AI就可以若合符節的與人類社會共生共榮。但行為的道德意涵從來就是一個非常棘手的問題。 Photo by Mikes Photos from Pexels 報告全文：AI in the UK: ready, willing and able?

英國上議會報告：AI與社會

0

lordsaireport-177de0b2180

2018-04-18

2018-04-18 05:52:09

https://medium.com/s/story/lordsaireport-177de0b2180

false

55

null

Lords Ai Report

lords-ai-report

Lords Ai Report

0

Sean Tseng

null

3c06213e848b

seantyh

9

7

20,181,104

null

0

from sklearn.metrics import mean_squared_error import xgboost as xgb from hyperopt import hp, fmin, tpe, STATUS_OK, Trials import numpy as np def objective(space): print(space) clf = xgb.XGBRegressor(n_estimators =1000,colsample_bytree=space['colsample_bytree'], learning_rate = .3, max_depth = int(space['max_depth']), min_child_weight = space['min_child_weight'], subsample = space['subsample'], gamma = space['gamma'], reg_lambda = space['reg_lambda'],) eval_set = [( X, y), ( Xcv, ycv)] clf.fit(X, y, eval_set=eval_set, eval_metric="rmse", early_stopping_rounds=10,verbose=False) pred = clf.predict(Xcv) mse_scr = mean_squared_error(ycv, pred) print "SCORE:", np.sqrt(mse_scr) #change the metric if you like return {'loss':mse_scr, 'status': STATUS_OK } space ={'max_depth': hp.quniform("x_max_depth", 4, 16, 1), 'min_child_weight': hp.quniform ('x_min_child', 1, 10, 1), 'subsample': hp.uniform ('x_subsample', 0.7, 1), 'gamma' : hp.uniform ('x_gamma', 0.1,0.5), 'colsample_bytree' : hp.uniform ('x_colsample_bytree', 0.7,1), 'reg_lambda' : hp.uniform ('x_reg_lambda', 0,1) } trials = Trials() best = fmin(fn=objective, space=space, algo=tpe.suggest, max_evals=100, trials=trials) print best

7

null

2017-12-28

2017-12-28 06:22:13

2017-12-28

2017-12-28 00:00:00

1

true

en

2017-12-28

2017-12-28 10:29:55

4

177e4fa7a5e7

2.083019

0

1

0

Recently I was working on a in-class competition from the “How to win a data science competition” Coursera course. Learned a lot of new…

4

Hyperopt — A bayesian Parameter Tuning Framework Friends don’t let friends use grid search. What is better than Random search? Recently I was working on a in-class competition from the “How to win a data science competition” Coursera course. Learned a lot of new things from that about using XGBoost for time series prediction tasks. The one thing that I tired out in this competition was the Hyperopt package — A bayesian Parameter Tuning Framework. And I was literally amazed. Left the machine with hyperopt in the night. And in the morning I had my results. It was really awesome and I did avoid a lot of hit and trial. What really is Hyperopt? From the site: Hyperopt is a Python library for serial and parallel optimization over awkward search spaces, which may include real-valued, discrete, and conditional dimensions. What the above means is that it is a optimizer that could minimize/maximize the loss function/accuracy(or whatever metric) for you. All of us are fairly known to cross-grid search or random-grid search. Hyperopt takes as an input a space of hyperparams in which it will search, and moves according to the result of past trials. To know more about how it does this, take a look at this paper by J Bergstra. Here is the documentation from github. How? Let me just put the code first. This is how I define the objective function. The objective function takes space(the hyperparam space) as the input and returns the loss(The thing you want to minimize.Or negative of the thing you want to maximize) (X,y) and (Xcv,ycv) are the train and cross validation dataframes respectively. We have defined a hyperparam space by using the variable space which is actually just a dictionary. We could choose different distributions for different parameter values. We use the fmin function from the hyperopt package to minimize our fn through the space. Finally: Running the above gives us pretty good hyperparams for our learning algorithm. In fact I bagged up the results from multiple hyperparam settings and it gave me the best score on the LB. If you like this and would like to get more information about such things, follow me. Also I would definitely recommend this course about winning Kaggle competitions by Kazanova, Kaggle rank 3 . Do take a look.

Hyperopt — A bayesian Parameter Tuning Framework

0

hyperopt-a-bayesian-parameter-tuning-framework-177e4fa7a5e7

2017-12-28

2017-12-28 10:29:55

https://medium.com/s/story/hyperopt-a-bayesian-parameter-tuning-framework-177e4fa7a5e7

false

499

null

Machine Learning

machine-learning

Machine Learning

51,320

Rahul Agarwal

Ex-CitiBank, Data Science@Walmart Labs

e8cce06956c9

rahulagarwal_20850

84

54

20,181,104

null

0

null

0

null

2018-09-10

2018-09-10 18:51:07

2018-09-10

2018-09-10 18:53:08

1

false

en

2018-09-10

2018-09-10 18:53:08

5

177fcabb106c

1.030189

0

Authors: Ishan Goel, Sukant Khurana*

5

Research Article: A Bayesian measure of association that utilizes the underlying distributions of noise and information Authors: Ishan Goel, Sukant Khurana* #3 ARE THERE ANY OTHER CORRELATION COEFFICIENTS? Other than these, there exist other polyserial and polychoric correlation to deal with the non-normality of data, but they were devised with an aim to be used with variables possessing a normally distributed underlying latent variable. Measuring of otherwise immeasurable normally distributed variables was possible with Likert scales. There is always a window for false negatives and false positives during the utilization of Pearson Correlation Coefficient due to its drawbacks of assumptions of linearity and normality. Correlation meme: Memes make everything better One can observe false negatives while testing a quadratic relationship. Here, as the underlying data does not get contained in a straight line, regardless of the tightness of the relationship, one cannot manage to achieve a high Pearson Correlation Coefficient. Further, as Pearson Correlation Coefficient assumes the underlying data is normally distributed, a higher value is given to the points away from the mean. When the actual underlying distribution is not normal, the Pearson Correlation Coefficient usually leads to giving false positives, provided there exists a correlation between the extreme values. There are theoretically no false negatives observed in the distance correlation measures as long as the coefficient was zero. Tags: #Statistics #StatisticalStudies #Correlation#BayesianProbabiltyOfAssociation #BPA

Research Article: A Bayesian measure of association that utilizes the underlying distributions of…

0

research-article-a-bayesian-measure-of-association-that-utilizes-the-underlying-distributions-of-177fcabb106c

2018-09-10

2018-09-10 18:53:08

https://medium.com/s/story/research-article-a-bayesian-measure-of-association-that-utilizes-the-underlying-distributions-of-177fcabb106c

false

220

null

Data Science

data-science

Data Science

33,617

Anuja Tiwatne

e319439e36e6

anujatiwatne

16

20

20,181,104

null

0

null

0

d5441256f5b8

2018-01-31

2018-01-31 10:35:08

2018-01-31

2018-01-31 12:15:19

1

false

en

2018-01-31

2018-01-31 12:15:19

27

177fd7d852d2

2.388679

3

0

As a part of our current Big Match research project, we’ve reviewed what has happened in the field of professional social matching in the…

5

The Hype in Professional Social Matching: A Closer Look to Applications and Research Designed by natanaelginting / Freepik As a part of our current Big Match research project, we’ve reviewed what has happened in the field of professional social matching in the last couple of years: interesting startups, new apps, relevant news articles, research papers, etc. Here’s a sneak preview of the findings! First, the marketspace for idea of “Tinder for jobs” seems pretty crowded already. In recent years, there has been a rise in startups that are developing applications where jobseekers are somehow matched with jobs. PIIK, Tiitus, Pockethunt and Treamer are some examples of the Finland-based startups. Also, Estonia-based MeetFrank has recently gained recognition and has been, for example, the official recruiting application at the Slush event. Last year, the Finnish government stated that it wants to make Finland a leader in the application of artificial intelligence. With the digitalization push, the beta version of Työmarkkinatori is going to be available in February. Recently, there was also Sitra-funded Ratkaisu 100 competition with one-million-euro prize pool where many solutions were based on the idea of matching. The finalists included “Tinder for mentoring” and Headai, which uses artificial intelligence to chart human expertise using open data on the internet. Headai was eventually one of the winners — congratulations! Worldwide, it seems that there are also matching applications for matching like-minded people. Some examples are Glynk, Opin, WhatTuDu, and citysocializer. Some of them, like Australian-based Affinity, even lists research papers on its website about how similar people attract each other. Interestingly, we were not able to find apps that match people who are unlike or have complementary skills or viewpoints. In addition to “Tinder for jobs” and “matching like-minded people” applications, there are already many startups and applications around making matches in events and networking in general. In Finland, maybe the most well-known apps are currently Brella and Mingla. Worldwide, some of the best know apps are b2match, Grip, Converve, Swapcard and Mixtroz. Events offer great opportunity to find new people and in research, there have been some studies, for example, on finding collaborators in academic conferences. The problem with small-scale applications by startups is to have enough users to create trustworthy recommendations or matches. It is easier to imagine more success to applications that are built on top of already successful social media platforms. Last year alone, Google introduced Google for Jobs “to bring you the most comprehensive listing of jobs”, Facebook introduced “Discover People” feature “to help you make more friends” and LinkedIn rolled out “Career Advice” feature. These new algorithms have not come without problems. For example, Gizmodo wrote how Facebook’s People You May Know feature can make too good recommendations, and in Finland, some have speculated how AI in recruiting can become biased and even racist. This “dark side” of matchmaking is interesting also from the perspective of research. In research, Loren Terveen and David W. McDonald outlined a framework and research agenda for social matching in a research paper published already in 2005. After that, especially the research from Julia Mayer, Li Chen, Marko Tkalcic and Peter Brusilovsky have been central to the research on social matching. Researchers have found that matching in current recommender systems are usually based on some similarity mechanism and there is growing interest to make recommendations more serendipitous and diverse. The topic of social matching calls for more multidisciplinary research and user testing with concrete prototypes.

The Hype in Professional Social Matching: A Closer Look to Applications and Research

30

the-hype-in-professional-social-matching-a-closer-look-to-applications-and-research-177fd7d852d2

2018-05-09

2018-05-09 13:52:28

https://medium.com/s/story/the-hype-in-professional-social-matching-a-closer-look-to-applications-and-research-177fd7d852d2

false

580

Collaborative effort to develop means for computational social matching

null

Matching People

socialmatchingtut@gmail.com

matching-people

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Sami Koivunen

Researcher, UX Designer

41779a061bee

skoivunen

4

22

20,181,104

null

0

null

0

f77757888b04

2018-02-18

2018-02-18 16:18:56

2018-03-04

2018-03-04 23:01:53

8

false

en

2018-08-10

2018-08-10 13:54:48

7

177ffabd4cfd

4.63522

27

7

0

Take over your Chatfuel bot to retain your messaging users

5

How to create a bot + human conversational experience with Chatfuel Take over your Chatfuel bot to retain your messaging users Chatfuel makes it easy to create Facebook bots, but we’re far from being able to put 100% faith in technology to serve messaging users. After all, bots can’t empathize like humans and understanding human intent can be challenging, even when you use Dialogflow for AI. The goal for Chatfuel users should be to find the right mix of automated and human-powered messaging in order to ensure delightful conversational experiences. If Chatfuel is a platform for creating Messenger bots and Dialogflow is the brain for your Chatfuel bots, Janis not only connects these two systems together, but because the connection is managed through Slack, Janis is able to brings humans into the loop when both Dialogflow and Chatfuel fails your users. How it works Janis sends alerts to Slack so you can identify problems quickly and act immediately. 🙊 No response: You’ll get these alerts when your bots don’t have a response to a user’s message so you can quickly identify problems and fix problems fast. 😱 Negative Sentiment: Janis can detect negative sentiment in a conversation and will alert you if a user seems angry (customer service?) or is just frustrated (the bot is annoying). 💁 Assistance Needed: Janis listens for words including help, human, agent, start chat, operator, and assistance, and alerts us in Slack. It’s fairly common to receive messages from users with these keywords, so you can take over live to help a user without leaving Slack. 🔔 Custom Alerts: You can add alerts to Dialogflow intents, so if a user sends a message that matches an AI intent in Dialogflow, it will alert you. These can be helpful for lead generation and forwarding a warm lead to a live agent. When you get an alert in Slack, click the link in the alert. You’ll need to be a Janis Pro User to get all alerts and a link to the transcript. Janis will open a Transcript channel in Slack with your Facebook Messenger conversation fully transcribed. Just start chatting from Slack. Janis will pause your Dialogflow AI responses as soon as you send a message Your message is delivered immediately to Facebook and added to the chat transcript! Example: A human added the last message from Slack, and it’s sent to Messenger If you want to resume your Dialogflow AI responses, simply type /resume in Slack and Janis will resume your AI. Of course, you’re probably busy and you might forget to resume your AI when you’re done chatting, so Janis will alert you if you stop chatting and then automatically resume your AI after 10 minutes of live chat inactivity. You can extend that time too if you type /pause X, where X is the number of minutes you want to keep a live chat session open. Janis maintains a full transcript of your conversation, combining your Chatfuel, Dialogflow and human messages. Boost Live Chat Productivity With Dialogflow Co-pilot mode empowers humans to manually push AI-powered responses saved in Dialogflow while chatting live with their users. Let’s say for example that you get an alert from Janis that your attention is needed and you jump into the conversation to chat live. Janis will pause your Dialogflow responses. Now, for every user message you receive, click to view Janis’ reply. If if it matches an intent, you’ll see suggested responses on the right side of your screen. Just click the green Send To User button for a response, and Janis will push your AI response to your user, eliminating the need to type out a response each time. Use Cases for Co-Pilot Mode For sales and customer service, much of your user input will be a frequently asked question and you can train Dialogflow to respond to these questions. While you’re chatting live, you will likely still get FAQs, and Janis lets you reply live, using answers you already have saved in Dialogflow. With Co-pilot mode you can provide human assistance, while trained responses are only a click away, which can be a huge productivity booster. If you don’t have a response saved in Dialogflow, just type out your response Janis will send your response to your user AND save your response in Dialogflow. Why limit human assistance to Facebook page administrators? While Chatfuel’s Live Chat Plugin is limited to Facebook page administrators working from the Facebook Page Inbox, Janis opens up new live chat capabilities for Chatfuel users. You may need someone with specialized knowledge on your team to assist. Just mention someone on your team while chatting with your user from Slack. When you mention someone and include your message, these message are NOT sent to Facebook Messenger, so you can have side conversations with team members in Slack. If you want someone else on your team to take over for you, they just type freely in the channel without mentioning someone on the team, and their messages will reach your user on Messenger. Watch how to create a Human + AI experience with Chatfuel Getting Started A Dialogflow account (Free conversational AI from Google) A Chatfuel Bot (connected to a Facebook page) Janis (An AI assistant that helps you connect and train Dialogflow) Before you go… Join our Facebook User Group and get support from the Janis team and other Janis users Want more superpowers for your Chatfuel bot? Just subscribe to our blog or add your email below to join our mailing list. Applaud, recommend, and share this article with other Chatfuel users if you found it useful.

How to create a bot + human conversational experience with Chatfuel

230

how-to-take-over-your-chatfuel-bot-without-leaving-slack-177ffabd4cfd

2018-08-10

2018-08-10 13:54:48

https://medium.com/s/story/how-to-take-over-your-chatfuel-bot-without-leaving-slack-177ffabd4cfd

false

928

Stories from the humans behind Janis.ai

null

getjanis

null

Being Janis

support@janis.ai

janis

ARTIFICIAL INTELLIGENCE,BOTS,SLACK,CUSTOMER SERVICE,MESSENGER

getjanis

Chatfuel

chatfuel

Chatfuel

116

Josh Barkin

Co-founder Janis.ai — Conversational AI Designer

82b99bff1a3f

joshbarkin

491

183

20,181,104

null

0

null

0

e98567e645c8

2018-04-24

2018-04-24 13:26:02

2018-04-24

2018-04-24 13:28:21

1

false

en

2018-04-24

2018-04-24 13:28:21

4

17803d72eda7

0.45283

0

Sometimes, Chi is more right than he knows.

5

Lou & Chi (#15): Scene Unseen Sometimes, Chi is more right than he knows. Previous Story Missed Lou & Chi’s other adventures? Start here At Kontiki Labs, we are enabling enterprises and businesses of all sizes to use AI powered technologies such as Machine Learning and Deep Learning by building affordable, people focused, design-first AI solutions. You can connect with us ontwitter or on email: hi[at]kontikilabs[dot]com

Lou & Chi (#15): Scene Unseen

0

lou-chi-15-scene-unseen-17803d72eda7

2018-05-25

2018-05-25 06:54:48

https://medium.com/s/story/lou-chi-15-scene-unseen-17803d72eda7

false

67

AI Powered Conversational Sales and Marketing Platform for Brands and Businesses.

null

Kontiki Labs

hi@kontikilabs.com

kontikilabs

CHATBOTS,CONVERSATIONAL MARKETING,VOICE ASSISTANT,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Swati Venkat

Certified bibliophile and wordsmith-in-training. Copywriter @ Red Baron

d2fdd728c2c7

swati_95515

2

3

20,181,104

null

0

null

0

null

2017-10-16

2017-10-16 10:50:27

2017-10-16

2017-10-16 11:29:04

6

false

en

2017-10-17

2017-10-17 13:45:37

1

1781752a73e6

6.157547

2

0

This week we have had some philosophical discussions about the new intelligent beings among us. Yes, they have arrived, and they have been…

5

Losing your Humanity… This week we have had some philosophical discussions about the new intelligent beings among us. Yes, they have arrived, and they have been quietly making their mark on people in cities across the globe. Some have quietly taken up jobs in hotels, some have been seen on tv, seminars and conferences. Others have begun making their mark in brothels and even online as sordid temptresses while a few are just as happy to be mail-order brides, literally. As if we do not have enough problems trying to feed, clothe, employ and manage the 7-odd billion people on the planet, we now have Humanoids to contend with. Jobs are already scarce and robots already seem like a great alternative to those at the top who still manage to squeeze the last drops of blood from their barely-legal slaves. Companies in Japan have employed humanoid robots in hotels and restaurants and it wont be long before others follow suit. The Big Fear The big fear is not even about job security. The real concern is how we interact with these objects. Some may pose the question, “Are they objects or are they beings?” Look, I know you can ‘grow up to be anything you want to be’ these days, and there are probably some crazy people trying to grow up to become humanoid bots, but let’s just be real here for a moment, whether they are beings or objects is not the point of the discussion so let’s leave it at that. Meet Sophia, a Humanoid robot The point of the discussion is that we are human beings, not bots. We have control of our actions, intentions and have to deal with the repercussions of our behaviours. Humanoid robots are built to look, act, think and one day feel like human beings, but they are programmed to do certain things by someone else. Even if they are intelligent beings with the ability to learn, they were built for a certain purpose. What that purpose is, and how it affects us, we need to question. Bots are meant to be tortured I was mortified to read the article from the likes of David Vincent Kimel this morning, asserting on some level that “…humanoids should feel a sense of honour serving humans and allowing them to cathartically actualise their fantasies and nightmares in a controlled and safe environment”. There is this fantasy that the world is too prohibitive and some people are innately depraved so we should be able to shed our inhibitions and do whatever we feel. Westworld is a series that engages this premise and centres around an amusement park intended for rich vacationers, which is looked after by robotic “hosts” and allows its visitors to live out their fantasies through artificial consciousness. A scene from Westworld People want to be able to ‘live our their fantasies’ and since robots are not real people, it shouldn’t matter what we do to them. Some people are of the impression that these bot-beings are designed to be our modern day slaves and because we design and build them, we can create them to do anything we want, especially things we aren’t allowed to do with real human-beings. The assertion is that we can use them to allow ourselves to engage in any kind of behaviour without the consequences. A question of Ethics The thing is, is the behaviour reprehensible because of its effects on other people or its effects on us. If it is based on the former, then does it mean it is less reprehensible if there is no damage caused to the other person? If the latter, then, does it mean engaging in such acts, whether it is with a humanoid or a real human, it is still something that needs to be condemned because of how it makes us feel or act? There is this whole argument about humanoids having simulated feelings and reactions that are not real so they should not be taken seriously or considered harmful to them. The point is that the perpetrator of the act is the one experiencing the real feelings. I agree with Kimel when he said, “Ultimately, concerns about playing with mechanical sex-dolls and life-size robotic action figures shouldn’t focus on the psychological harms suffered by the toys and their likelihood of starting to attack us.” The concern should focus on the psychological harms suffered by the person engaging in the behaviour, and their likelihood of attacking us. How does allowing a pedophile or a serial killer (or other more twisted folk) indulge in their sociopathic behaviour add any value to their human experience? It only reinforces their drive and addiction to deviant behaviours. A person who pulls the heads off cats for fun in their spare time does not ‘get relief’ and then acts normal in society because we let him pull the heads off cats when nobody is looking. It only allows them to graduate to more sadistic forms of behaviour, with more sophisticated victims because the action of engaging in such a behaviour reinforced it. Remember that poor sex robot, Samantha, that needed repairs after being repeatedly molested while on display at a recent tech fair? Nobody was concerned about Samantha, and even I am more concerned about what kind of people think it is ok to exhibit such lewd behaviour and get away with it. Sergi Santos, the doll’s developer iterated, “People can be bad. Because they did not understand the technology and did not have to pay for it, they treated the doll like barbarians.” Sergi Santos and “Samantha.” Would those people behave in such a lewd manner in any other circumstance? Did the availability of the technology and the visual appeal of the humanoid, coupled with the fact that this was actually a sexual object with no repercussions elicit such behaviour? Would they do it again? Would it change the way they see other women in public? Would they feel more confident about attacking a real woman in public like they did the doll? Would it give a molester enough practice to do it to a real person? Would it create more outlandish behaviour with a don’t care attitude? Does having this outlet affect how people behave in real life? Lower your gaze? I posed the question to my colleagues, “Does a Muslim man have to lower his gaze if he sees a female robot?” Depending on the sophistication of the humanoid robot, it is very hard to tell them apart from real humans at first glance. It makes the experience more real. It makes you think and feel as if you are doing something to a real person. That’s just the problem. We are reinforcing negative behaviours. We are allowing people to become more depraved and sadistic by making it ethical to be unethical. ChihiraAico, a lifelike android robot built by electronics manufacturer Toshiba, is able to talk, sing, gesture and cry like a 32-year-old Japanese woman. The Islamic perspective is simple. It’s not about whether it is a robot or a human. You, as a human being, are answerable for your behaviour and your own intentions. If you think that engaging in something will lead to fitna (strife, distress, temptation, affliction), stay away from it. So yes, the answer was, “…he should lower his gaze”. The wisdom behind it is that you are a spiritual being, and should rise above the base desires of the lower self. If you feel that looking at something or someone, whether it is real or perceived, human or humanoid, will cause you to behave in a way that is contrary to what your are striving for as a spiritual being, then avoid the temptation. It is not liberating to become a savage, debased human being, with no sense of morality or character. It will not benefit you to engage in barbaric acts of debauchery, regardless of the nature of the victim, so do not be fooled into thinking that the lack of consciousness in the bot will cover up your lack of conscience in real life. Is that what the creators of these humanoids want? A world filled with depraved humanoids who want to be human and humans who have lost their humanity and just become slaves to their addictions? You may or may not be legally answerable for defiling a robot in real life, but we are still answerable for our intentions and our characters.

Losing your Humanity…

31

losing-your-humanity-1781752a73e6

2018-04-11

2018-04-11 14:16:04

https://medium.com/s/story/losing-your-humanity-1781752a73e6

false

1,380

null

Ethics

ethics

Ethics

7,787

Zahara Cassim

People & Impact

dcff9ac51301

cassim_13433

20

31

20,181,104

null

0

null

0

null

2018-05-08

2018-05-08 09:40:45

2018-05-08

2018-05-08 09:41:35

1

false

en

2018-05-08

2018-05-08 09:41:35

1

178250580371

1.641509

1

0

If you are a fashion designer and you are working on a new collection basis the idea that you have in your head, you are also at the same…

5

The Marriage of Convenience — Artificial Intelligence & Fashion If you are a fashion designer and you are working on a new collection basis the idea that you have in your head, you are also at the same time now sure if that idea has been done before by anyone else or not. How to do come out of this situation? Try IBM`s AI tool called Cognitive Prints, a dedicated tool for the fashion industry which will help designers to actually understand if the idea that they think is unique has been done before. You have to take the photo of a dress/sketch and use the tool to search the similar garments done before. The tool can search for images with specific results like Chinese collars, printed shirts or bell bottom jeans. It can also search design patters based on any image data set by a user, images which will have sunsets, swimming pool, vintage cars etc. The cognitive prints makes it easy for the designers by augmenting the design cycle. The AI powered search engine is the joint effort of IBM and The Fashion Institute of Technology (FIT) has been developed on the back of training on 100,000 print swatches across 10 years of fashion week entries. The AI tool users can search the images by year, designer or inspiration (Indian Street Wear). This will help designers to tackle the whole plagiarism debate that has been going on for a while in the fashion industry. The team of Cognitive Prints is already working on adding several new features to the existing tool`s abilities. They are working on features which allows designers to change the background color, or changing the stripes to polka dots on a fabric. Artificial Intelligence in fashion is now been extensively used, there are apps which allow you to check makeup using face recognition, dresses on body types etc. The designer duo Shane and Falguni Peacock has been using Watson, IBM`s AI platform to search over 600,000 images in order to create a new collection. Tommy Hilfiger uses AI to identify fashion trends in real time for a faster production turn around, Amazon has been using AI designer which will generate customized designs for online buyers. The trend of using AI in fashion will only grow from here.

The Marriage of Convenience — Artificial Intelligence & Fashion

20

the-marriage-of-convenience-artificial-intelligence-fashion-178250580371

2018-05-08

2018-05-08 10:37:32

https://medium.com/s/story/the-marriage-of-convenience-artificial-intelligence-fashion-178250580371

false

382

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

FashGroupe

The most happening fashion, lifestyle and entertainment related content on web.

cb40d36211f

fashgroupe

17

2

20,181,104

null

0

null

0

d8630d2b9608

2018-03-20

2018-03-20 11:42:24

2018-03-20

2018-03-20 11:45:31

5

false

en

2018-04-17

2018-04-17 20:53:32

5

1782c0a989d1

3.067296

4

0

It’s taken us a year to find him, and we feel like every nanosecond was worth it. This month Kirk Ballou joins FLUX as our CTO and Chief…

5

Blockchain “Titan” Kirk Ballou joins FLUX as CTO & Chief Blockchain Architect FLUX CTO Kirk Ballou It’s taken us a year to find him, and we feel like every nanosecond was worth it. This month Kirk Ballou joins FLUX as our CTO and Chief Blockchain Architect, giving us the chops and experience needed to launch our vision of a “prosperous future for all,” merging blockchain technologies with hardware so we can listen to Mother Earth. Kirk (follow him on Twitter here) has over 10 years of experience leading digital development for Fortune 500 companies like National Geographic CNN, Red Bull, & Microsoft hewing these heavyweights into new frontiers through his company Touch Titans. A contributor to open-source blockchain projects such as Zclassic and Bitcoin Private, along with his technological and business acumen will help FLUX scale out and round out our development team between the United States, Hong Kong and Israel. Among his accomplishments are winning the Stellar Build Challenge, Adobe’s Open Screen Fund, and AT&T’s Calling All Innovators. “I’ve known Blake for a dozen years and we have had the chance to work symbiotically in other startups. It started with Flux as an advisory position last fall and then the vision of the team inspired me to join Flux as I got to know the team, history and our collective capacity to achieve results,” says Ballou. “For me it was obvious where I want to spend the next five years: We are in an exceptionally special place right now where blockchain applications are meeting hardware, is meeting unlimited kinds of sensors, is meeting a community that wants to listen and learn from natural systems,” he adds. FLUX Network Ballou is architecting “If you just take the siloed data of agriculture, worth trillions, well just imagine when you get a million farmers use FLUX to share best practices, trading data through a peer-to-peer marketplace, well you turn the current system on its head;, leaving room for a whole new economy based on sustainable practices and a fairer distribution of wealth for all.” “This is a future I want a hand in creating — to be able impact the base of the pyramid, the 3 billion underserved by current economic systems globally and the majority of smallholder farmers that rely on growing their own food to survive,” says Ballou, “and farming is just the first. Imagine the market opportunities and impact in forestry, fish farming, water remediation.?” At FLUX, Ballou will first start in making our vision of urban farming for all, a reality on the blockchain. He will link growers using our first product built on Flux’s open hardware, MICO, to a peer-to-peer network. will they can exchange recipes and best practices for growing any kind of crop and medicinal herb. They can even then sell these products and services on the network. FLUX CEO Blake Burris! “Out of all of the candidates we considered,We met dozens of people, but we chose Kirk because he’s already experienced in both blockchain and large- scale apps and platform development, and he is ready to apply his smarts to disrupting the agriculture system so we can grow a sustainable food supply for all around the world;, starting with Africa and India,” says FLUX CEO Blake Burris. Ballou is currently leading strategic partnerships and Flux global development teams in Colorado, Israel and Hong Kong. Follow Kirk! https://twitter.com/kirkballou https://medium.com/@kirkballou https://www.linkedin.com/in/kirkballou/ If you would like to join our mission, get in touch: careers@fluxiot.com Flux (www.fluxtoken.io) is a platform technology that applies big data processing and AI to Life Sciences. It’s like a stethoscope for Mother Earth.

Blockchain “Titan” Kirk Ballou joins FLUX as CTO & Chief Blockchain Architect

8

blockchain-titan-kirk-ballou-joins-flux-as-cto-chief-blockchain-architect-1782c0a989d1

2018-04-17

2018-04-17 20:59:40

https://medium.com/s/story/blockchain-titan-kirk-ballou-joins-flux-as-cto-chief-blockchain-architect-1782c0a989d1

false

592

News and announcements about the FLUX Protocol project.

null

FLUX Protocol

blake@fluxtoken.io

flux-network

BLOCKCHAIN,AGTECH,IMPACT INVESTING,CRYPTO,AGRITECH

fluxtoken

Agriculture

agriculture

Agriculture

12,051

Karin Kloosterman

Working on Flux, a “prosperity” grow technology that applies big data processing and AI to agriculture. We’re rooting for Planet Earth (www.fluxtoken.io)

8edc88c5d0ae

karinkloo

260

296

20,181,104

null

0

null

0

null

2017-09-01

2017-09-01 11:22:04

2017-09-01

2017-09-01 11:28:34

4

false

en

2017-09-01

2017-09-01 11:28:34

4

1784b02ad121

3.296226

2

0

Since sound and music are time based media, events (that are triggering sounds) are occurring in varying patterns of frequencies and…

1

Sonic State Machines Ramon Llull: Ars Magna (1305) Since sound and music are time based media, events (that are triggering sounds) are occurring in varying patterns of frequencies and repeating occurrences. This sound experiment is treating with the events of sounds as entities that can be transferred or permutated forward and backwards using random probabilities. Sequencers in electronic music and time based media do nothing else than distributing events within small cells of discrete timeframes. This method is purely combinatoric, that might involve ancient techniques such as permutation, repetition and other basic algorithmic manipulations. Some people compose those cells by hand (just like most composers in the last century), others let algorithms take part in the process, leading to unexpected generative results. Manipulating sequences with a machine companion reminds me to some of the techniques of the 13th century philosopher Ramon Llull. As it is pointed out by Umberto Eco in his brilliant book The Search for the Perfect Language, Llull’s masterpiece Ars Magna introduced an abstract language of a reduced alphabet to create cosmic questions and seek answers for a reconfigurable universe. His intent was to convert muslims to christians through praying, reasoning and pure logic (instead of war) so he studied arabic language, combinatorics and other unfamiliar-at-the-time practices. Ars Magna was using a mechanic, reconfigurable circular geometry that could be rotated (using papers with a pin in the middle), which led to a large set of constellations of the symbols (each symbol meant different principles). This attend to seek answers through such logical combinatorics reminds me to the current trend of using neural networks and machine learning in order to find meaningful answers regarding to interrelationships within the quantified universe. Again, we find the rhyming nature of history in those practices: we regularly like to build soft machines, state machines, learning machines to achieve the devine & the ultimate and propose questions that can be answered only by using some extended cognitive process. Some computer scientists have adopted Llull as a sort of founding father, claiming that his system of logic was the beginning of information science. Regarding to (machine) learning, mathematics and classification: with the discovery of his lost manuscripts Ars notandi, Ars eleccionis and Alia ars eleccionis, Alchetron enciclopedia says “Llull is recognized as a pioneer of computation theory, especially due to his great influence on Gottfried Leibnitz. Lull’s systems of organizing concepts using devices like trees, ladders and wheels, have been analyzed as classification systems”. Markov Graph of transition probabilities between states A, B and C Learning machines (especially in reinforcement learning), generative music, autonomous poetry are all using Markov Chains as their state swapping machine components. For simple generative purposes, I built the following Pure Data patch on the top of a first order Markov model. In the patch, lists of numbers define the probabilities of the next state. Most abstractions are built with rjLib. First order, 4 state markov chain with probability lists For the first state, a 0 1 1 0 would mean, the next step can’t be itself, but can be the 2nd and 3rd state, it also can’t be the 4th. For the 2nd state, a 1 0 1 0 means, it can continue with the 1st & 3rd states, but not with itself, nor the 4th state, etc. First order markov chain sequences can be designed like this, on a meta-level of the sequencing process. The patch works with weighted probabilities on a smaller level, too: each sound sample occurrence is defined with a percentage, that leads to the probability of the actual sound production. Probability distribution of sonic triggers in the sequence (0: not to play, 50: to play or not to play, with 0.5 probability, 100: play) For example: in the sequence that consists of 8 elements, a list like 100 0 0 0 0 0 0 50 means, when reaching the first position, the sound will be triggered for sure, 100%. Then, no sound, and the last position will be on or off, with a weight of 50%. Some sound sculpting is involved in the process, such as pitch & gain changing in order to make the resulting sound more diverse. As a rule of thumb, it is good to follow: avoid empty and boring repetitions. Take care of each sound occurrance: modulate, change their gain, with some LFO, permutate other parameters along time to achieve an organic sound.

Sonic State Machines

12

sonic-state-machines-1784b02ad121

2017-10-19

2017-10-19 12:26:04

https://medium.com/s/story/sonic-state-machines-1784b02ad121

false

688

null

Machine Learning

machine-learning

Machine Learning

51,320

Agoston Nagy

coding, interaction design, workshops

1fb5146c84d

_stc

148

104

20,181,104

null

0

null

0

null

2017-10-19

2017-10-19 15:00:16

2017-10-19

2017-10-19 15:02:43

1

false

en

2017-10-19

2017-10-19 15:02:43

1

178571bd3fd0

0.369811

0

Wouldn’t we all aspire to achieve a perfect 20/20 vision, and have it remained so for the rest of our lives? Well, that dream could just…

3

Bionic Lens Could Give You Superhuman Vision Wouldn’t we all aspire to achieve a perfect 20/20 vision, and have it remained so for the rest of our lives? Well, that dream could just become a reality in the near future! Don’t believe it? Read more…

Bionic Lens Could Give You Superhuman Vision

0

bionic-lens-could-give-you-superhuman-vision-178571bd3fd0

2018-05-09

2018-05-09 23:04:36

https://medium.com/s/story/bionic-lens-could-give-you-superhuman-vision-178571bd3fd0

false

45

null

Science

science

Science

49,946

Evolving Science

Inspiring innovation and scientific research for the advancement of mankind. www.evolving-science.com

7bd43d6083eb

EvolvingScience

25

185

20,181,104

null

0

null

0

null

2018-09-12

2018-09-12 09:18:45

2018-01-15

2018-01-15 09:38:38

13

false

en

2018-09-17

2018-09-17 13:41:36

10

17866c7aeac9

7.675472

1

0

Learn about the hottest trends at CES 2018 that encourage a brave new tech world

4

CES 2018: Tomorrow’s Trends to Solve Today’s Problems Learn about the hottest trends at CES 2018 that encourage a brave new tech world This year’s Consumer Electronics Show (CES) kicked off on January 9, and this time the event is more breathtaking to watch than ever. The most popular tech trends at CES 2018 become more relevant and desirable to see in the near future as mobility is spreading everywhere, and everything has become a gadget. Consumers expect to use products and services from anywhere — and often on the go. Connected devices are no longer only phones, laptops, wearables, smart coffee machines, or even talking fridges. We’re stepping further with smart houses, connected city infrastructure, intelligent assistants, and highly automated vehicles. These dates of CES have brought us the greatest hype around automation, connectivity, and augmented reality we cloud only dare to perceive. And most importantly, the new trends are about solving our everyday problems. Autonomous technology is turning delivery services upside down Delivery services are often a headache for consumers, and the negative responses can be enough to drive any delivery provider crazy. This could be a cold pizza, bad quality goods, or smaller size of the ordered clothes. Customers won’t use a service again if their experience is bad, which means lost money and harm to a service provider’s image. The latest news from CES 2018 ensures some optimism. Toyota has presented a practical solution to this problem. Its e-Pallete Concept is an autonomous driving pod that delivers products and services. For instance, it can bring goods from the grocery store that you ordered. Moreover, it can bring the entire mobile shelf with these goods to choose the best one right from there. Also, you can be sure your pizza will be hot because a mobile kitchen will warm it on-the-go. Toyota e-Pallete can bring goods from the grocery store that you ordered. Moreover, it can bring the entire mobile shelf with these goods to choose the best one right from there. But Toyota promises to use its e-Pallete Alliance to improve a wide range of services apart from just delivery. These self-driving pods could be van-based restaurants or physical Amazon stores on wheels. Customers could avoid ordering the wrong size of clothing, for example, with the help of mobile showrooms. A rolling van could even become a mobile hotel room that you could order if you’re having trouble finding a free room in a city. Also, e-Pallete opens a wide range of new mobility services a part of listed above, just take a look: Toyota’s e-Pallete Concept is a blank canvas that lets companies make services mobile and customer-centric. Service providers should be able to connect an entire fleet of e-Pallete autonomous vehicles and manage them remotely. This promises to turn automated delivery into a lucrative opportunity for businesses, which can capitalize on the connected and mobile experience in consuming services and products. It also connects various industries, making automotive software providers the main drivers of their autonomous future. Toyota’s e-Pallete Concept is a blank canvas that lets companies make services mobile and customer-centric. Drones are playing the piano, so what? They could be doing much more Probably your attention was also captured by one of the most resonating CES 2018 highlights which is the astonishing performance of drones playing the piano. Yeah, it was cool, but what’s next? Who really needs a huge drone fleet that plays music? Probably not everyone. But if we look deeper into the robotics technology behind this performance, we see some true potential. What we really should take away from this show is that drones can now navigate with centimeter precision to destinations no larger than the key of an on-floor piano. Moreover, they can do this in synchronized groups of hundreds of units. With the addition of automated path planning, sensors, cameras, and intelligent navigation solutions like Sky Atlas for safe flights to avoid restricted no-fly areas, we would receive an easily controlled swarm of little helpers for a range of tasks. Among others, these tasks include disaster management, rescue operations, building inspections, equipment repairs in dangerous areas, precise and dependable deliveries to anywhere, and rapid data collection. Drones may be used for advertainments and commercials, farming and agricultural purposes, military and airspace security. The total profit of drones usage across the industries would be counted for billions. Drones can now navigate with centimeter precision to destinations no larger than the key of an on-floor piano. Moreover, they can do this in synchronized groups of hundreds of units. Augmented reality builds trust between drivers and driverless cars With fast-emerging autonomous driving technology, the issue of trust in artificial intelligence cannot be overestimated. The reason why trust is needed is simple — when the car itself performs all driver’s tasks it essentially becomes responsible for the lives of both the driver and those who share the road with the self-driving vehicle. This is no wonder that safety is among the top concerns drivers have about the future self-driving cars adoption. A person sitting behind the wheel of an autonomous car can’t just trust the AI, at least not right now. A self-driving car has to show its intention to make a move before actually making it. And this is where augmented reality (AR) can truly shine and gain recognition among the consumers. WayRay, a Swiss augmented reality software provider, is moving toward a solution for the issue of trust in autonomous cars. They’re introducing an AR-based heads-up display that holographically projects a car’s intentions to turn left or right, speed up, brake, or make some other maneuver right onto the car’s windshield. This advanced HUD also shows information about the environment, route, points of interest, available features, potential obstacles, and car performance to assure the driver that everything’s okay. WayRay is making augmented reality a critical part of the Human Machine Interface that breaks the barrier between people and full metal cockpit of an autonomous car. WayRay introduced an AR-based heads-up display that holographically projects a car’s intentions to turn left or right, speed up, brake, or make some other maneuver right onto the car’s windshield. An intelligent home assistant? Give me two, please. A war that hasn’t even begun Staying home alone isn’t boring anymore. Moreover, you no longer need to run from room to room to turn off the light, increase the volume for your favorite song, or even heat the oven to cook a holiday dish. Just ask your intelligent assistant to do these things for you and, voila, everything’s taken care of. Alexa, turn off the light in the bedroom, please. Siri, turn the volume up. That’s my favorite song. Hey Google, find a roast chicken recipe and prepare the kitchen for cooking. What could be easier than making something happen just by saying it? We just have one warning: don’t confuse the name of your assistant, as they have feelings too. This year, the top trends at CES show that the market for intelligent assistants and smart home devices is heating up. Soon, the war to sneak into your house with the best home pod app will begin, and mobile development will have a new point of focus for manufacturers who want to bring more value to clients and businesses. In addition to that, smart assistants are finding a natural fit in other environments, including in the car. Toyota has announced an original in-car system for next year with Amazon Alexa on board without any additional hardware or software to be installed. The integration will be so natural that you’ll call your car a second home even while driving long miles at night. Toyota has announced an original in-car system for next year with Amazon Alexa on board without any additional hardware or software to be installed. Voices, voices everywhere. Infrastructure and cars are really talking The idea of the smart city is stuck in the heads of enthusiasts from many different industries. This time, Ford has intervened in the dispute over what smart connectivity should look like. The company has announced a partnership with Qualcomm that’s intended to give birth to a cellular V2X technology to redefine the modern city infrastructure. Ford is preparing a cloud-based connected platform for mobile services, smart infrastructure, and connected vehicles to unite efforts in preparing for a highly connected world. Connected cars are already on the road, but what are they communicating with while rolling the lane? Basically, the infrastructure of even the biggest cities isn’t ready yet to communicate with highly connected technologies. Ford is preparing a cloud-based connected platform for mobile services, smart infrastructure, and connected vehicles to unite efforts in preparing for a highly connected world. Ford isn’t the first company with an idea to design a comprehensive data platform for collaborative development. Initially, HERE has presented its own Open Location Platform with the same purpose of uniting industries to bring about a common connected future. Baidu is also on their way to presenting a platform for crowdsourced development in the location and smart infrastructure domains. ___________________________________________________________________ As evidenced at CES 2018, the biggest breakthroughs today are appearing at the junction of industries. The new tech world is all about partnerships between fields that previously were considered irrelevant to each other. The importance of platforms that could connect different industries is rising, so as the demand for platform developers to commute the pipelines. The automotive industry is gaining a new role as a hub that helps businesses in various fields transform their products and services into on-the-go solutions and adapt to the future mobile-first and autonomous-first world. ___________________________________________________________________ Don’t hesitate to ask our experts how trends from CES 2018 could help you solve the problems you’re struggling with. Originally published at www.intellias.com on January 15, 2018.

CES 2018: Tomorrow’s Trends to Solve Today’s Problems

3

ces-2018-trends-that-will-shape-technology-market-intellias-blog-17866c7aeac9

2018-09-17

2018-09-17 13:41:36

https://medium.com/s/story/ces-2018-trends-that-will-shape-technology-market-intellias-blog-17866c7aeac9

false

1,663

null

Self Driving Cars

self-driving-cars

Self Driving Cars

13,349

Intellias Automotive

Software engineering services for OEM and Tier 1 companies

65d3ac492e92

intellias

25

558

20,181,104

null

0

null

0

null

2018-04-02

2018-04-02 08:03:43

2018-04-02

2018-04-02 08:29:19

7

false

zh-Hant

2018-04-11

2018-04-11 12:18:13

10

1787ef189c83

1.442453

2

0

人工智慧的發展，對於產業的影響不容小覷。新媒體作為利用電腦及網路的新興媒體科技，相較於傳統媒體之形式，其受到人工智慧的影響，肯定比起傳統媒體較會引發更大的質變。

5

新媒體如何運用AI增加經營效率與價值? 人工智慧的發展，對於產業的影響不容小覷。新媒體作為利用電腦及網路的新興媒體科技，相較於傳統媒體之形式，其受到人工智慧的影響，肯定比起傳統媒體較會引發更大的質變。 “Two people in elegant shirts brainstorming over a sheet of paper near two laptops” by Helloquence on Unsplash 新媒體借重社群媒體的資料結構分析，人工智慧等相關資料收集工具，更能夠帶給新媒體環境更有價值的附加功能。本文列舉幾項人工智慧可以帶來何種附加價值，無論是應用在社群媒體或是其他新媒體環境，皆能作為新媒體產業應用的參考。內容自動生產在社群媒體中的內容，是由使用者自行生產，然而對於企業端來說，若要運用社群媒體來傳遞訊息，企業主仍需要自行生產內容、制定內容企劃等。這也說明為何現今社群小編比比皆是的現象，透過社群小編對於社群趨勢的洞察，制定相關的內容政策，讓該企業品牌能夠在社群媒體間不斷的發酵與傳遞。除了社群媒體之外，傳統媒體如電視、廣播更是需要針對平台上的內容精心設計，制定內容產製程序，以及季度或年度計畫，為的就是讓內容跟得上當下的議題。然而，計畫總趕不上變化，如何收集當下的資訊，對使用者傳遞最即時的訊息，是人工智慧可以輔助的。至於人工智慧自動產生內容的原理是什麼呢？可以簡單的以隱藏層(Hidden Layer)來看，如下圖所示，當我們input一個內容架構給電腦時，電腦會透過神經網絡架構跟不同類目的組合，最後產生一個電腦生成的內容。這項應用已經被實現，Aiva是一個人工智慧的作曲家，在input中工程師會輸入許多古典樂的樂譜。而在隱藏層(Hidden Layer)，Aiva學習了音樂生產的基本結構和模式，最後Aiva在瞭解所有古典音樂的結構後，自動產生了原創古典音樂。想聽聽Aiva所創作的古典樂？音樂如此複雜的原創內容，都可以透過人工智慧的方式自動產生，社群媒體上的影像圖片以及文案，確實能夠在可預知的未來看到實際應用的普遍案例。 AI追蹤社群媒體中的聲量與輿情新媒體或是社群媒體在經營時，大眾如何討論你的品牌是一件重要的事情，因此許多網路媒體聲量與社群監聽(social listening)成為經營者必須瞭解的方法與工具。但是如何在社群媒體中追蹤數萬條貼文以及每個用戶的討論內容，過濾掉一些不必要的雜訊，找到媒體經營者真正要的聲音，作為後續經營策略的參考，這些人工智慧都是可以協助幫忙的。坊間有許多社群監聽工具，有些需要付費，以下列出免費的資源來供讀者參考：中研院輿情分析系統：本系統運作分為四個階段 1. 爬取新聞語料：利用爬蟲程式蒐集最近的新聞，本系統目前採用的是蘋果日報和中國時報的新聞。 2. 中文剖析系統：將蒐集到的新聞前處理後，使用中文剖析系統建立每個新聞語句的結構樹。 3. 建立關鍵字資料庫：從結構樹中抽取字詞之間的關係，依照字的詞性以及搭配詞的詞性建立關鍵字資料庫。 4. 建立情感分析資料庫：利用 E-HowNet 中的情緒相關字眼，分析每個語句中的情感分佈，建立情感資料庫。快速找到對的KOL 新媒體經營若只有單方面的傳遞訊息，則無法正確的打動消費者或是讀者的心。數位行銷的慣用手法就是找到針對自身主題相關的意見領袖(KOL)，來為品牌或是新媒體平台做代言與背書。但如何找到正確的KOL是件需要經驗的功夫。過往的媒體經營者會根據該KOL的個人特質以及所吸引粉絲的族群進行分析，檢查其是否與自身產品有相關性。再者，針對該KOL的網路聲量以及過往有的行銷活動經驗來進一步決定是否該邀請他來為自身代言。如今，人工智慧也能針對分析品牌的DNA找到符合該品牌的KOL。因為媒體經理人雖然可以透過經驗與手邊數據判斷該KOL適不適合，但仍沒辦法確保萬無一失的狀態。 https://insightpool.com/marketing-platform/ Insight Pool透過自家的搜尋系統，整合品牌資訊進而比對超過六億筆KOL在社群媒體上的資料，如臉書、Instagram、Twitter等，去與品牌做最高相關性的適配。如此一來，不但節省了經營者在選用KOL的時間與成本，更能夠有效率的針對後續的活動進行規劃。即時又準確的競爭者分析所謂知己知彼，百戰百勝。對於社群行銷，或是新媒體的經營更是重要的一環，不能不知道競爭者目前的狀態。 Unmetric是一個可以幫助經營者瞭解競爭對手在社群上狀態的平台，該平台會針對競爭對手在社群中的資料進行收集與分析，經過統計資料的整理後以視覺化的方式呈現出來。 https://unmetric.com/ 該資料經過統整後，讓經營者可以一目了然的知道目前與自身產品或平台相關的競爭對手都在討論些什麼議題、做了什麼活動以及目前的聲量。 Unmetric使用了人工智慧，縮短洞察競爭對手資料的時間，交叉比對了競品的資料並且不斷挖掘異常值，來追蹤競品的動態。也使用了人工智慧來消除數據上歧義，從而增加系統的洞察力，並標出異常值以提醒經營者們的注意。此外，第二個領域是模式的匹配，以查看對手在社群媒體上內容運行的方式，並預測出非公開使用者討論的聲量，如貼文的觸擊率等。甚至，假如競品的貼文因為社群媒體中的某個人，而有顯著的討論聲量時，Unmetric也會將這些視為經營者必須知道的重要指標。人工智慧不但能夠提升新媒體與社群的經營效率，透過人工智慧相關工具的輔助，更能夠彰顯出在現今社群潮流變化萬千的趨勢。傳統的經驗已經逐漸沒辦法追趕上變化的速度，而這時，人工智慧不妨成為經營者重要的好幫手。參考資料： New AI-Based Tools Are Transforming Social Media Marketing With the vast amounts of unstructured social data, the myriad of social media influencers and the growing number of…www.forbes.com Creative machines: How close are we to AI-generated content marketing? - ClickZ We typically think of artificial intelligence (AI) within our industry in terms of processes and calculations. Media…www.clickz.com 猜你感興趣： AI如何解決中國13億人「吃」的問題？ AI發展快速，除了我們時常聽到的下圍棋、安全辨識、智慧助理外，到底還能對人類現實生活中有什麼樣的幫助呢？這篇文章將以農業的角度來探討AI的幫助。medium.com 人工智慧已成趨勢，你真的瞭解它嗎？人工智慧已經不是新名詞。早在半世紀前就以充斥在人們的想像裡，本文帶你一窺AI基礎概念。medium.com

新媒體如何運用AI增加經營效率與價值?

11

新媒體如何運用ai增加經營效率與價值-1787ef189c83

2018-04-11

2018-04-11 12:18:14

https://medium.com/s/story/新媒體如何運用ai增加經營效率與價值-1787ef189c83

false

104

null

人工智慧

85

Ray Wang

媒體｜科技｜社會：政治大學傳播所，寫寫觀點和學習心得

19b621c8f533

wangjui

141

111

20,181,104

null

0

null

0

null

2018-01-09

2018-01-09 02:00:01

2018-01-09

2018-01-09 15:25:09

7

false

en

2018-01-09

2018-01-09 18:07:34

12

178805cca2ab

4.517925

7

0

fuzz·y (adjective) — difficult to perceive clearly or understand and explain precisely; indistinct or vague.

5

An Introduction to Fuzzy String Matching fuzz·y (adjective) — difficult to perceive clearly or understand and explain precisely; indistinct or vague. Starting out as a developer, one might not think of the term ‘fuzzy’ as applicable to anything they are doing. In fact, you might not think of the term at all (I didn’t. Except while looking at cat gifs on other people’s blogs). However, we use ‘fuzzy’ to describe life regularly, because life can be extremely vague and general. As an object-oriented programmer, we strive to minimize the gap between our code and the real world through accurate models of this world. So why wouldn’t fuzzy apply? Of course, it already does. Super cool vaguely original content. Fuzzy logic? The concept of ‘fuzzy logic’ was developed in the 20th century, elaborating on Jan Łukasiewicz’s proposition of many-valued logic in 1920. Jan specificlly pioneered negation and implication; you might know implication as an if statement. Many-valued logic is necessary because it allows for mathematical calculations around the ambiguous nature of life. “Fuzzy logic… provides an effective conceptual framework for dealing with the problem of knowledge representation in an environment of uncertainty and imprecision.” The importance of fuzzy logic has only become more apparent as science digs into computers and programming is taken further. It has become especially useful in the context of artificial intelligence. You may be thinking ‘But wait! Booleans are everywhere in programming. Isn’t that at odds with many-value logic?’ Well, yes and no. We use many- (AKA infinite-) valued logic regularly when we code already. Every time you use an if statement, adding in multiple elsif clauses, you are using a form of infinite logic. You are accounting for multiple possible scenarios other than true or false, up to a potentially infinite amount of times. However, there are many other situations where fuzzy logic can become immediately helpful. Edit Distance & Fuzzy String Matching “Fuzzy String Matching is basically rephrasing the YES/NO ‘Are string A and string B the same?’ as ‘How similar are string A and string B?” Fuzzy logic came in to play for me when I was working on a basic command line interface at the Flatiron School. If someone types a command or something incorrectly, it can break your program. The first step is to implement proper error handling, e.g. a message to let the user know their error and allow for another try. However, that isn’t forgiving when you’re a messy typer. You can have your program match the first three characters, but there is still room for mistakes there. Human error is consistently inconsistent. How do we account for incorrect letters, or extra ones, that can appear anywhere in the string? Enter edit distance. An edit distance matrix from calculating the Levenshtein Distance. The two possible paths for the above matrix. “=” Match; “o” Substitution; “+” Insertion; “-” Deletion At its most basic, edit distance calculates how similar (or not) two strings are. This is calculated through the number of operations needed to transform one string into the other — e.g. with my name “Julien” and the common spelling “Julian”, only one operation is needed, a substitution of “a” for “e.” The original algorithm can be attributed to Vladimir Levenshtein, who passed away in September of last year. It’s known as the Levenshtein Distance. It’s a recursive function that calculates the edit distance for every prefix and suffix. The algorithm results in a matrix of all possibilities. Edit distance has found widespread use in the mapping and comparing of genomes. It has also become extremely useful in the namesake of this article — fuzzy string matching. Fuzzy string matching, also known as approximate string matching, can be a variety of things; Regular expressions are a form of it, as are wildcards in the context of SQL. It is any form of attempting to match one string to another one. Fuzzy string matching with regards to edit distance is the application of edit distance as a metric and finding the minimum edit distance required to match two different strings together. We encounter this on a daily basis in our interactions with computers — does the red line under a misspelled word ring a bell? Or maybe this one: All of these implement some form of fuzzy string matching. The Levenshtein Distance is the most common metric, but there are other variations on the algorithm — Sellers, Damerau-Levenshtein, Hamming, and more. They all have different ways of computing the same thing. Ok, how do I use this to my advantage? The Levenshtein Distance in Ruby. Much of the heavy lifting has already been done for us, as these algorithms have already been expressed in code in every conceivable language across the internet. I posted the Ruby version above, but for everyone else: Find the Levenshtein Distance expressed in other languages here. In fact, the reason I found myself down this edit distance wormhole was thanks to the Ruby gem Amatch by Florian Frank. Check it out! It offers you a selection of algorithms to choose from depending on the data you’re inputting. Some are more suited to larger bodies of text, while others aren’t. Amatch testing. The world of fuzzy string matching has come a long way. There are a lot more advanced ways that incorporate these concepts into their fuzzy string searches, and there is more room for efficiency (through MIT claims the algorithm is as efficient as it’ll be). Dig in! Sources Yager, Ronald R., and Lotfi A. Zadeh. An Introduction to Fuzzy Logic Applications in Intelligent Systems. Kluwer, 1998 https://en.wikipedia.org/wiki/Many-valued_logic http://www.levenshtein.net/ https://www.r-bloggers.com/fuzzy-string-matching-a-survival-skill-to-tackle-unstructured-information/ https://en.wikipedia.org/wiki/Approximate_string_matching https://en.wikipedia.org/wiki/%C5%81ukasiewicz_logic https://www.youtube.com/watch?v=ocZMDMZwhCY http://rosettacode.org/wiki/Levenshtein_distance#Ruby http://www.occasionalenthusiast.com/wp-content/uploads/2016/04/levenshtein-formula.png

An Introduction to Fuzzy String Matching

125

an-introduction-to-fuzzy-string-matching-178805cca2ab

2018-04-09

2018-04-09 16:10:44

https://medium.com/s/story/an-introduction-to-fuzzy-string-matching-178805cca2ab

false

919

null

Programming

programming

Programming

80,554

Julien Tregoat

creating decentralized applications @ThunderToken

ee41e5f07ab8

julientregoat

73

81

20,181,104

null

0

null

0

null

2018-09-18

2018-09-18 15:17:16

2018-09-18

2018-09-18 19:53:23

2

false

en

2018-09-18

2018-09-18 19:53:23

3

17882d5ce899

5.515409

1

0

Artificial Intelligence will be one of the basic technologies that will be used to build the future.

5

Why You Need to Study Artificial Intelligence, Even As A Non-Technical Person Artificial Intelligence will be one of the basic technologies that will be used to build the future. You’ve probably heard the terms Artificial Intelligence (AI)and Machine Learning (ML) recently in the news. It seems to be the new buzzwords, just like cryptocurrency not long ago. Bitcoin is not making headlines anymore, making people question, was it only a clickbait buzzword and is AI the same? What are these new technologies and why should we, the common people, the non-technical people, care? Reality is— our society is on the brink of a technological revolution, and Artificial Intelligence is at the core the Fourth Industrial Revolution which will fundamentally alter the way we live and work. It will affect all of us. Leading technology giants Facebook, Google, Amazon are all excited about Artificial Intelligence. Google CEO Sundar Pichai emphasized in 2017 just how important AI is to the future of the company by saying: “I’m really happy with how we are transitioning to an AI-first company. The Google Assistant is one of our first steps towards that future…Advances in machine learning are helping us make many Google products better… Beyond that, we continue to set the pace in machine learning and AI research.” — Alphabet 1Q17 Earnings Call Amazon CEO Jeff Bezos did not hold back either in 2016: “It’s hard to overstate how big of an impact AI is going to have on society over the next 20 years.” at the Code Conference in California Even though we hear about it in the news, and see headlines, how many of us really know — what is AI? Why is it important? Let’s face it, there are not many people who truly understand what AI and ML are, it’s relatively new, but as it will have a huge impact on our lives in the future, we should aim to have at least a basic understanding of it. I would suggest to pay attention especially if you are still in the early days of your career. Chances are, it will deeply influence the future of work and your career as a result, no matter the industry you are in. So — what is AI? I have been really interested in everything related to the future and especially the technology that will shape it. But I have to admit, I also lacked a clear understanding of AI, even though I’ve known for a very long time about the massive importance of it. During my University years, I always attended workshops around entrepreneurship and got the habit of going to one of the biggest tech conference in the Nordic, Slush, every year. Even though I studied business, I always tried to widen my perspective and educate myself in those topics that weren’t directly related to my studies. It is a shame business and technology studies are not integrated more, as it often is in the “real” world. So, I have been learning on my own, mostly because, well, I can. As I wrote in my previous article, it is up to us how we use the power that is access to unlimited information that is available thanks to the internet. I really find technology fascinating and over time, got over the limiting belief that only tech people need to know this stuff. I’m not a tech-head — what I am is a person of curiosity, ideas and dreams. We cannot wait until someone or something (university) will give us the information we need, we need to educate ourselves. Taking matters in my own hands to learn about these concepts that are already major and will only get more significant in the future, I did what any other person would do — I Googled it. Here are some of the results to what is AI: · AI is the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. · Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other animals. · Artificial intelligence (AI) makes it possible for machines to learn from experience, adjust to new inputs and perform human-like tasks. · Artificial intelligence (AI) is an area of computer science that emphasizes the creation of intelligent machines that work and react like humans. From this simple Googling experience, I already noticed that AI means different things to different people, and I had to go deeper in the attempt to better understand it. The graduation speech of my generation. Google is a good start. You can find many online courses and workshops organized all over the world if you live in a bigger city. Some of them might cost a lot of money since it really is a hot topic at the moment. Free quality online education Finland style Here’s where the Finnish education delighted me once again. I learned about an online course created by the University of Helsinki and Reaktor, called The Elements of AI. I saw many people in my professional network endorsing it on social media, so I didn’t hesitate for a moment to take this course. I took the first module yesterday and I’m impressed. It’s mind blowing how this quality of teaching is free and accessible to anyone. The teaching material itself is amazing and equally importantly, the way it’s delivered is what the future of higher education can and should be. The course is divided into 6 modules: · What is AI? · Solving problems with AI · Real world AI · Machine learning · Neural networks · Implications Upon completion, you can earn an extra 2 ECTS toward your studies and a digital certificate, which looks like this: This course will not make anyone an expert, but as you can tell from the modules overview, it provides a solid foundation to understanding the topic and the ability to join the discussion. Innovation is everyone’s job I said previously that you should pay attention if you are just now starting out your career. Why? Because you need to be in charge of your own knowledge and readiness to face the changing job market. Currently, the education system cannot keep up with rapid changes, but thank goodness we can use our own minds and the internet, to be ready. No matter your role in your future work place, what will be radically different from organizations in the past is how innovation happens on all levels. Imagine if Apple would have only relied on product (technology) innovation, instead of innovative approaches to packaging, retail sales, customer access, partner agreements etc. It probably would not have become the world’s first trillion-dollar company. You don’t need to become an AI expert to come up with new ways how AI can help solve problems and make our lives better. You need to understand the core of the underlying technology to join the conversation. Gone are the days when only certain people within an organization were in charge of innovation. The organizations that will survive the Fourth Industrial Revolution encourage and empower all of their employees to innovate. Diversity is key in building a better world with technology. We need different points of views and more gender equality in the future. How can we build technology that will be used by everyone, but if not everyone is involved in the process of imagining, designing and building it? Understanding the basics of technology, such as code, AI, ML means we no longer have to accept the bystander role in our digital society. Even if we think of ourselves as non-technical people. If this post (or my any other work) has in any way inspired you to keep learning about technology, entrepreneurship and self-development, please leave me a message here or find me on social media. I already have a small study group online where we encourage and inspire each other, but I want to grow the community, if there’s interest. Looking forward to connecting with more people and continuing to learn together!

Why You Need to Study Artificial Intelligence, Even As A Non-Technical Person

1

why-you-need-to-study-artificial-intelligence-even-as-a-non-technical-person-17882d5ce899

2018-09-19

2018-09-19 02:26:02

https://medium.com/s/story/why-you-need-to-study-artificial-intelligence-even-as-a-non-technical-person-17882d5ce899

false

1,360

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Sandra Lusmägi

Founder of Wanderwork.com, futurist and global citizen.

6bbf14026f1b

sandralusmgi

38

269

20,181,104

null

0

null

0

8803a17b09f

2017-09-09

2017-09-09 01:32:28

2017-09-09

2017-09-09 01:51:00

4

false

en

2017-09-09

2017-09-09 17:32:31

1

1789cb26b501

1.111321

1

0

Created for MDes Seminar, taught at Carnegie Mellon University by Molly Steenson

3

Sketchnotes: Week 3 Created for MDes Seminar, taught at Carnegie Mellon University by Molly Steenson I’ve been mulling over this idea lately about the paradox of freedom in modern times, and reading about AI this week and how little knowledge, control, and rights we have over our data and online activity made me think about how much of our freedom we freely surrender for the sake of modern technology. “Freely surrender,” though sometimes it feels like a form of inescapable bondage. “In truth that which you call freedom is the strongest of these chains, though its links glitter in the sun and dazzle your eyes.” —Kahlil Gibran, “On Freedom”

Sketchnotes: Week 3

16

sketchnotes-week-3-1789cb26b501

2017-12-22

2017-12-22 16:22:15

https://medium.com/s/story/sketchnotes-week-3-1789cb26b501

false

109

By Master of Design & Master of Professional Studies Students at Carnegie Mellon

null

Seminar One: Context and Perspectives for Design in Flux, CMU School of Design, 2017

steenson@cmu.edu

seminar-one-interaction-service-design-concepts

INTERACTION DESIGN,CMU,SYLLABUS,UX,AI

maximolly

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Hajira

Graduate student in Carnegie Mellon University School of Design

ed308f23f654

arijah_cmu

35

34

20,181,104

null

0

null

0

null

2017-12-13

2017-12-13 22:12:49

2017-12-13

2017-12-13 22:30:24

4

false

en

2017-12-13

2017-12-13 22:30:24

2

178ac5cc1677

8.624528

0

Artificial intelligence is quickly invading our world. AI, as it is commonly referred to, is intelligence that is demonstrated by machines…

1

Artificial Intelligence and Psychology Artificial intelligence is quickly invading our world. AI, as it is commonly referred to, is intelligence that is demonstrated by machines and that mimics the cognitive functions that we often associate with the human mind. It was first explored in the 1950’s, birthed out of a workshop at Dartmouth College, the next decade was unproductive, leading the U.S. government to cut funding in order to fund “more productive projects”. This cut let to what was called the AI winter, a period of almost a decade in which AI research was untouched. Following this “winter”, AI was revived in the 1980’s and by the late 1990’s AI began to be used for logistics, data mining, medical diagnosis and other areas because of increasing computational power, new ties between AI and other fields and a commitment by researchers to mathematical methods and scientific standards. Today we have seen an increase in artificial intelligence in other aspects of our environment. Almost all of what we see in artificial intelligence today is focused on machine learning. Machine learning is the input of new data to increase the knowledge base of the machine system. This is done in many ways, mainly using intricate and extensive algorithms. Artificial intelligence is a very general term, it’s often used to refer to concepts that are intangible and hard to “see”. There is a reason for this. Artificial intelligence is “whatever hasn’t been done”. As machines become increasingly capable, more tasks that are considered to require intelligence are removed from the definition, this is known as the AI effect. Artificial intelligence is sometimes sorted into two categories; weak AI and strong AI. Weak artificial intelligence is an artificial intelligence system that is specialized in one very specific area or made to perform one or two very specific tasks. This is the kind of intelligence that we are all familiar with, because most of us have it right at our fingertips. Iphone’s virtual personal assistant, “siri”, is an example of weak artificial intelligence because while siri can learn new things about you, your needs and behaviors, it’s function doesn’t go beyond that. So don’t worry, your Siri isn’t going to develop its own personality and start defying you, well not in the near future anyway. Strong artificial intelligence is indicated by a general base of knowledge and one that, like the human brain, is flexible and adaptable. This kind of machine intelligence would be able to use it’s wealth of knowledge to adapt to new or unfamiliar situations, much like the human brain. Arend Hintze, a professor at Michigan State University classified this type of artificial intelligence into four categories: Type 1: Reactive machines: These systems have the ability to react to unexpected situations. It can identify it’s environment and make predictions but it has no “memory”, meaning it cannot draw on past experiences to inform future knowledge. This system is only able to react to its environment Type 2: Limited memory. These AI systems can use past experiences to inform future decisions. Observations are used to inform actions happening in the not-so-distant future, though these observations are not stored permanently. Type 3: Theory of mind: This kind of system would be able to recognize that others have their own beliefs, desires, and intentions that impact the decisions that they make. This kind of artificial intelligence doesn’t exist yet. Type 4: Self-awareness. This kind of AI system would have a sense of self and a sense of consciousness. Machines with self-awareness understand their current state and can use the information to infer what others are feeling. This type of AI also does not yet exist. These types or stages of artificial intelligence identify what we see now but also the goals for artificial intelligence. Currently, we are starting to see machines that can be classified as type two come into the light. This “type” would include some systems of autonomous vehicles. It is clear to see that the people at the forefront have big aspirations for what artificial intelligence could be. Intelligence as an Innovation People in the tech industry talk about artificial intelligence, as we know it today, as the next big innovation for our society, but is it? Because intelligence is such a broad term, it’s hard to know what parts of it will become the next big thing. For example, we’re seeing a boom in trends that have artificial intelligence at their core, like autonomous vehicles, in-home personal assistants, ect. None of those however, are large enough or influential enough to stand alone as the next big innovation. The thing that will make artificial intelligence an innovation will be its integration into the everyday technology that we already use today. Almost anything that we do could have intelligence integrated into it, which would change the world around us. This creates a huge market for artificial intelligence. As conglomerates explore the different ways of implementing it into their products, the artificial intelligence market will grow. As companies continue to propel this trend forward, they will begin to figure out how to integrate intelligence in a way that sustains the brand that they have built. While each company will have its own way of doing this, but each company will have to adapt to the new c-scape that AI makes. In the terms of Larry Kramer, it will be imperative that companies focus on the four C’s, consumers, content, curation, and convergence. In my opinion, “consumer” is the most important of these, it will be increasingly important for any company exploring AI to keep their intent focus on the consumer above everything else. This then leads to the other three C’s, these companies will have to figure out the content, or in the case of AI, the tasks it will perform. They will then have to figure out how to present it to the consumer and how it will change the landscape of their technology and its uses, curation and convergence. However the companies decide to do it, it does seem as though artificial intelligence can continue to grow. Based on some of the supertrends outlined by Edward Cornish in his book; Futuring, it seems that artificial intelligence plays into a few of the trends that drive our current society and dictate our future. One of the supertrends outlined is technological progress, which essentially is what artificial intelligence is based on. Artificial intelligence needs technological progress to evolve but also provides a pathway to habilitate progress. The second supertrend that promotes artificial intelligence is economic growth, as opportunity arises to integrate artificial intelligence into society, this will create opportunity for immense economic growth. As long has private companies continue to push the boundaries of what AI is capable of, it is inevitable that the government will get involved, setting regulatory rules. Artificial intelligence is mainly reliant on big data to extract meaning from analytics and numbers, but how that big data is collected could be disrupted by government regulation in order to protect citizen privacy. Psychology and Intelligence It is clear that those working with artificial intelligence have big plans for the future, our future, but I’d like to examine just how this could affect the future of our society. Inevitably there will be forces that block the expansion of artificial intelligence, those forces could be economic, technological, ethical, ect., but in my opinion the biggest blocking force to the expansion of artificial intelligence will be human emotion and psychology. Intelligence is a multi-faceted concept, in which a lot of things are intertwined. There are tons of different kinds of intelligence to be measured. Just a few of them include: naturalist intelligence, musical intelligence, logical-mathematical intelligence, existential intelligence, interpersonal intelligence, bodily-kinesthetic intelligence, linguistic intelligence, intrapersonal intelligence, and spatial intelligence. These types of intelligence refer to humans and the human experience. We think of humans as being the only beings to possess all of these types of intelligence, but what would happen if that changed? Machine learning has scratched the surface on many of these types of intelligence. In the case of Logical-Mathematical Intelligence, it’s possible that machines have already surpassed us. They have also touched linguistic intelligence, musical intelligence, and in some cases, like in autonomous vehicles, spatial and kinesthetic intelligence. Some might even argue that artificial intelligence has also acquired naturalist intelligence because it is able to distinguish between people. So then the question to be asked is; where’s the line between human and machine? The answer seems somewhat obvious, humans have emotional intelligence and machines don’t. Humans have the ability to connect with each other, have self-awareness, and are able to think deeply about life. What would happen though, if machines developed aspects of emotional intelligence. If a machine develops the ability to connect with humans, become self-aware or think deeply about life, how will this affect humans? This all may sound like the plot of a science-fiction novel, though the work of Arend Hintze proves that this is ultimately the end goal for artificial intelligence, at least through his eyes. It’s safe to assume that even the biggest technology optimists may feel unnerved by the thought of machines achieving that level of intelligence, and the rest of us are left absolutely terrified by the mere thought of it. It’s that feeling, the uncomfortableness, that will ultimately be the biggest blocking force to artificial intelligence. Once AI surpasses that milestone, companies will be left to guide their consumers through the landscape of uncertainty that is going to follow. They will have to figure out how to ease the very real anxiety of the customers that they are trying to keep. A Look into the Future Machines have taken over, human civilization is left in ruins… just kidding. Though to some, that’s what it feels like. Artificial intelligence has grown immensely since 2017, when the rest of this paper was written, and it now encompasses many aspects of human life. Smart homes, autonomous vehicles, virtual personal assistants, smart pets and even personal robots have become popular, and more human needs are met than ever before. Machines, thorough artificial intelligence, now have a sense of self and has learned how to connect with not only humans but other machines. This has left many people feeling uneasy, the thought of connecting with a machine is outside any schema they have created, and therefore they are having a lot of trouble adjusting to the changing world and the threat of competing with machines for jobs. These people, while cautious, are not completely against the new technology trends because they have seen how much better life can be with artificial intelligence integrated into it. The prevalence of artificial intelligence has manifested itself mainly in the area of digital personal assistants, the new technology that now knows you better than you know yourself, which now serves as a tiny life coach to the people who use it. As a company that has invested itself in this trend, you want to prove to the skeptics that they are worrying for the wrong reasons. As more and more conglomerates move into this new media space, your competition is increasing and you are fighting for your piece of the pie. You must focus on the consumers. Most importantly you will need to keep the trust of the consumers. There are many ethical problems that could arise and the key to dodging a fatality is avoiding mistakes before they happen. First, there are privacy issues at stake. Privacy is something that our culture holds sacred and disrupting that is going to be fatal to the company. At the very least, the consumers must feel like they have privacy. Curation and convergence are extremely important to consider as well. These will affect how the consumer sees the trend and how it changes the way they use it. As more companies move into this media space, it will change the landscape of it. Companies may use this space for advertising, using the mentor dynamic of the technology to increase brand loyalty. This will change the landscape. While this seems like a major opportunity, how does it affect the ethics of the consequences that stem from it. Will this crush smaller brands that don’t have the money to compete with the conglomerates? Or will brands even still exist. It’s all very hard to say because the future of artificial intelligence is fairly uncertain. So how are you going to be Moses, the person that leads people out of uncertainty to the promised land?

Artificial Intelligence and Psychology

0

artificial-intelligence-and-psychology-178ac5cc1677

2018-04-01

2018-04-01 13:35:03

https://medium.com/s/story/artificial-intelligence-and-psychology-178ac5cc1677

false

2,100

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Julie Downing

null

9ec4a53bf598

juldowning24

0

1

20,181,104

null

0

null

0

null

2018-01-20

2018-01-20 12:58:05

2018-01-20

2018-01-20 13:34:36

1

false

en

2018-01-20

2018-01-20 13:34:36

3

178ade32b239

1.207547

0

We act as if the world is made of persons and things. It might be obvious what ‘things’ could be. But it is less obvious what a ‘person’…

5

The Badge of Personhood We act as if the world is made of persons and things. It might be obvious what ‘things’ could be. But it is less obvious what a ‘person’ could be. For example, legal systems recognize corporations as legal ‘non-human’ persons. But this is just the beginning of the rabbit hole. We have the obvious: pet owners attributing personhood to their pets, and often regarding them as members of the family. And then on a darker note we have the fact that colonialism and slavery were possible because the colonizers saw the conquered masses as non-human — savages who need to be civilized. Not recognizing someone’s personhood is often an act of evil. And yet we do it all the time. Hence the popular notion that the problem we face is that too many people value ‘material’ things more than ‘persons’ — who have ‘real’ value. Further down the rabbit hole, you will find that dolphins and even the rivers they swim in could be given the badge of personhood and the privileges and protections that go with it. In the coming decades, we will further expand the community of ‘persons’. The voice assistants on our phones and devices like Amazon’s Alexa and Google Home are the first indicators of what’s possible and what’s coming our way. Virtual and augmented reality will further add to the growing pile of possibilities. There are sure to be a lot of questions to answer and a lot of reflection to be done. This is why the future is fascinating and terrifying at the same time.

The Badge of Personhood

0

the-badge-of-personhood-178ade32b239

2018-01-20

2018-01-20 13:34:36

https://medium.com/s/story/the-badge-of-personhood-178ade32b239

false

267

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Soul Sailor

Soul Sailor is the game of life.

528cc6f1f209

soulsailorsays

13

7

20,181,104

null

0

null

0

null

2018-01-15

2018-01-15 11:10:31

2018-01-16

2018-01-16 14:41:33

1

false

en

2018-09-12

2018-09-12 13:06:50

10

178bdc8bce58

4.49434

1

0

A necessary conversation for governments, citizens, journalists and scientists

5

Algorithmic transparency and public policies A necessary conversation for governments, citizens, journalists and scientists [this article was originally published on Valigia Blu, in italian] source: Diliff, Wikipedia cc-by-3.0 James Vacca, member of the New York City Council, the legislative body of the city, has recently proposed a bill that aims at making the algorithms used for public decisions more transparent. The project gathered a lot of interest, and public hearings scored record presence, and it is proceeding in its route to being approved by mayor De Blasio. The discussion during the hearings are really worth listening. New York uses a number of software instruments to take decisions, or at least to suggest them, in terms of public policy. The use of cutting-edge technology is highly appreciated if looking at efficiency, but it is no news that, if left unchecked, can hide insidious problems. Vacca, during his opening speech and the presentation of the bill, refers to two highly relevant examples. The first is the application of an algorithm to determine the fire department services, used also to decide the number of police force in different districts. Despite Vacca has been a high-level administrative figure for years, he explains, he has not been able to get an answer to some quite clear questions: what are efficiency criteria? How is the decision taken? “There’s a formula” (RAND formula)— is the only explanation given — without the possibility to clarify which variables were considered, and without the possibility for citizens to audit the process. Same happens when a teenager tries to get accepted to the his or her preferred high-school, and is assigned to the sixth or seventh choice. Rightfully, Vacca asks: “why shouldn’t one be able to understand — and therefore to contest — the decision?” There are obviously many other applications: from public housing to justice. In the case of algorithms used in court rooms, there have already been several uses and widely discussed controversies: Eric Loomis was charged six years by a proprietary algorithm, that indicated in him a propensity to recidivism, without any right to contest the decision. Algorithms are used intensively in predictive policing, estimating the risk of escape or crime-committing. Problem is, as well explained by ProPublica, that these techniques introduce a racist bias. Clearly, these softwares are used to respond to complex issues (even just in a sole quantitative dimension) with an increased efficiency. But it is exactly where injustices seem to lie, that it would be fair for citizens to understand the rationales. The importance of algorithms in our lives has increased steeply, and it will more and more in the future, even though this risks to be a niche topic, a missing discussion. The critiques to opaque algorithms, in terms of responsibilities are mounting, rightfully, towards the Silicon Valley, where they are in between consumers and corporations. However, what are the questions we have to ask ourselves, when they are handling the relationship between citizens and public administrations? How should this relationship be regulated? What are the responsibilities of policymakers in terms of transparency and accountability? I know it might sound too far from the Italian current from the state of affairs, but I believe it’s better to reconsider. To make an example: Bank of Italy, a few weeks ago, published a research that re-established the targets of Renzi’s 80 euros tax-rebate, using machine learning techniques. It’s a very interesting study that aims at increasing the efficiency of this measure, by establishing the target in a data-driven manner, redefining who’s considered “in need”. The estimates are that 29% of the cost of the policy (around 2 billion euros), was addressed to non-ideal households. But what is “ideal”? The key characteristic, for a policy that aims at stimulating consumption, is the propensity to consume of a household. The technique used is that of decision trees, and we can already appreciate a concern of the authors for communicability and transparency of this process. Going back to New York and Vacca’s proposal, we have to say that the problem is not only relevant, but also complex under many point of views: firstly technological, but also legislative, economic, ethical and of security. Algorithms are often sold by third parties that would never want to share the code. The relationship between suppliers and public administrations is obviously commercial, so based on the selling of proprietary software: this process, though, excludes citizens from access to knowledge. From data input used to the methods, and therefore to the possibility to appeal. And it is fundamental that citizens gain this access, because, as it has been clear for a long time, neutral algorithms do not exist. Vacca sums up this idea in a sentence that I find very effective: “algorithms are a way to encode assumptions”. But the topic here is even more subtle, because opacity is not only a product of private interest: algorithms are by nature complex objects, for most of us. But if we reckon that access to knowledge, in terms of public policy, is something to protect, it necessary to act. But how? It is not a trivial issue: transparency does not mean publishing a few thousands lines of code. It would for sure be a starting point, it would enable hackers and journalists to work on it. But the real transparency is not in code: it more generally lays in the accessibility to the decision process, to the assumptions made. First of all, the code must be documented. Open source without documentation is not necessarily open, as stated by one of the discussants in the talk. That is just the beginning. Among the proposals, for example, there are simulation systems in which the citizens can understand what changes based on changes on the outputs. Talking about data, another fascinating aspect of the issues comes up: if transparency is ideal, we might say there is a trade-off with privacy. How could we (as citizens and as policymakers) take care of this? During the discussion, a representative of open-source world explains how open technologies can help not only to better accountability, but also security levels. In a post-Equifax world, it is straightforward to assess the necessity to have clearly established relationships between third-parties and public administrations, when sensible data (and hence lives) of millions of citizens depend on opaque or vulnerable IT structures. I reckon we can make a parallel comparison with the discussions about the Freedom of Information Act, which responds to the citizens requests for public data: it must involve not only tech experts, jurists and policymakers, but also journalists, activists, citizens. There is no simple answer, but if we want to build an open and inclusive democracy, thinking about what ‘government’ might mean in the age of information, I believe it is a conversation worth having.

Algorithmic transparency and public policies

1

algorithmic-transparency-and-public-policies-178bdc8bce58

2018-09-12

2018-09-12 13:06:50

https://medium.com/s/story/algorithmic-transparency-and-public-policies-178bdc8bce58

false

1,138

null

Algorithms

algorithms

Algorithms

7,319

Enrico Bergamini

Ferrarese a Bologna, orgogliosamente emiliano. Amo Internet nelle sue declinazioni. Ogni tanto scrivo. Faccio cose con i dati.

dc448ebe420d

enricobergamini

158

330

20,181,104

null

0

null

0

null

2018-09-04

2018-09-04 04:14:45

2018-09-08

2018-09-08 17:21:25

1

false

en

2018-09-10

2018-09-10 07:22:03

1

178c7ea2e795

2.743396

0

Recently, I talked to some friends about food and if there was some technology that can recognize a food picture. Not only recognize it as…

2

Google and Artificial Intelligence Recently, I talked to some friends about food and if there was some technology that can recognize a food picture. Not only recognize it as a food but also what kind of food it is. Then, I remembered that Google had launched a camera app called the Lens App that when you open your camera, you can point at various objects or places to tell you what they are and offer reviews. This was released on October 4, 2017. Image: Google When I looked it up again, Google Lens has been updated to be a better one. Besides our camera app, Google Lens will also be implemented on Google Assistant and Google Photos. Thus, it can even do a broader type of things. Recognizing the style of clothing and furniture. If you see a piece of clothing that you like while shopping, Google Lens can identify it and give reviews about similar things. It is also possible to recognize furniture that you found interesting while visiting your friend’s house Search around us. Google Lens will detect where you are at and shows you the places nearby just by pointing your camera around you. Another thing Google Lens can do is Smart Text Selection which enables your phone’s camera to highlight a text that is being pointed at and copy it to be used on your phone. On Google Assistant, there will be an icon at the bottom right-hand corner. If you tap it and point your smartphone camera around, you can get suggestions on the things you point at such as show times outside a cinema. You can see the movie’s details, immediately order a ticket, and even add the event to your calendar. Within Google Photos, Google Lens can identify buildings or landmarks, presenting users with locations, directions, and other information about them. These functionalities that made finding information much easier have been popular for a long time. Over time, the technology has become a lot more advanced and a lot of people started trying to make it too. There are also other apps that can do similar things such as Bixby Vision for Samsung and Image Analysis Toolset (IAT) (not available in Indonesia). All the things that those apps can do are possible by using the technology of Artificial Intelligence (AI). Currently, AI is one of the most significant technologies that is being researched by many computer scientists. Google has started to implement AI on many of its software and products. In this area particularly, AI enables the app to learn things and quickly determines or recognizes what is being shown. Google’s vision is to bring the benefits of AI to everyone by helping us solve big and small problems. AI also provides new ways of looking at existing problems, from healthcare issues, environmental issues or even advancing scientific discovery. Google AI At Google, we think that AI can meaningfully improve people's lives and that the biggest impact will come when everyone…ai.google Google also has an open-source machine learning platform called TensorFlow. It provides an open-source software library for dataflow programming across a range of tasks using a symbolic math library. Machine learning apps such as neural networks use this. By design, TensorFlow is able to tackle a much wider range of machine learning problems than its predecessor (DistBelief). It allows users to design deep neural networks and run them on a single smartphone or across thousands of computers in data centers. There are a lot of possibilities of AI use in the future. In some movies, they show that AI can mean the end of the world, ruled by robots. However, I think that’s not the case. It promises a bright new future for us. Despite that AI is still not that smart to be able to understand all the things around us, it is making progress every second. I’m sure that in the near future, AI will have been developed to the point that it changes the way technology works for the good of mankind.

Google and Artificial Intelligence

0

google-and-artificial-intelligence-178c7ea2e795

2018-09-10

2018-09-10 07:22:03

https://medium.com/s/story/google-and-artificial-intelligence-178c7ea2e795

false

674

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Vincentius Aditya Sundjaja

null

c3ba12cf1bc9

vasundjaja

0

2

20,181,104

null

0

null

0

null

2018-01-07

2018-01-07 20:05:03

2018-01-29

2018-01-29 10:23:30

5

false

en

2018-04-07

2018-04-07 10:39:24

3

178cbc254bc6

5.761635

2

0

How will AI affect our lives?

5

When Artificial Intelligence becomes unrecognizable Click here to share this article on LinkedIn » My first encounter with an “AI” was ELIZA. It was not only me, lots of people played the program in the past. There were different variation scripts of ELIZA but the DOCTOR script is the most widely known. In my case, this was a program on the C64. It was presented as a psychotherapist wherewith you could interact by typing in conversation. Due to the clever programming, it was entertaining and sometimes even convincing: “how does she know that?” But after a short while, the repeated texts, the patterns, and structure of the logic started to be noticeable. Since it was programmed in BASIC, I could stop the program and investigate the source. I remember being surprised how relatively short the program was. It used a small dictionary and a bit of pattern matching. The clever bit, besides the pattern matching, was that the program used your answers to create more questions. And if it got stuck, it said something like: "can you tell me more about that?" Just like a real psychotherapist. The curious thing about the ELIZA program was that some people really got into it as if she was a real psychologist. This even got its own name: the ELIZA effect; when people contribute human thoughts and feelings to computers (and maybe also robots?). ELIZA was developed by Joseph Weizenbaum in 1964–1966. A famous anecdote is that his secretary, while knowing about the program, still got confused as if ELIZA was a real person or not. ELIZA passed the famous Turing test as one of the first psychotherapeutic chatbots. The whole point behind the creation of ELIZA was to prove how superficial the relation between human and computer was. And Joseph was actually shocked by the fact that people took his program as a real person. People could spend hours telling the program all kinds of personal stuff. When Weizenbaum later told his secretary that he had access to all the logs of the program, she was furious about the invasion of her privacy! Weizenbaum found it bizarre that users actually believed ELIZA understood and could help with their problems. It seems that ELIZA was resonating with something in the human psyche but this was absolutely not the creator’s intention. His first reaction was to erase the program completely because he saw it as a threat, but then he went on and wrote a book about his experiences in 1972 (Computer Power and Human Reason). In this book, we find that the creator of one of the first and convincing chat AI programs speaks out against artificial intelligence. It is also interesting to note that the AI hype of these days is not new. During the days ELIZA was created, generous funds from the military-industrial complex were spent on AI research. Now 50-something years later, we live during another AI hype. Lots of money is poured into it again, and numerous start-ups are working to create new innovative AI solutions. We have learned a lot in the past, and currently computers are vastly superior to the ones in the sixties. And more importantly; we also have become used to the terminology and ideas. Artificial intelligence has been the holy grail in the world of Informatics for a long time. Nowadays it seems that if you just program something that can learn a little bit then it’s already called an AI (you know it is a hype if they put the term AI on anything that contains a little bit of interaction). By the old definition of AI, we are nowhere close. This means having all these labels with AI on it flying around is a bit ridiculous. In a positive mood, I would call them pseudo-AIs. A short while ago, I watched an anime series (Japanese animation) called Real Drive. I hadn’t read anything about it, so I didn’t know what it was about or its background story; I just started to watch it based on a short preview on YouTube. It’s a friendly anime that’s happening on an artificial island in the future and it revolves around a kind of cyberspace known as 'the Metal' that uses the ocean as a metaphor. In the series some characters have cyberbrains, and others (like the main character) don't. Some characters have few prostheses while some have so many prostheses that they are basically cyborgs. And some characters are androids. But this is the crux; they all look just like normal humans. (Well, besides the police that for some reason are cylindrical robots.) Can you spot the human? So, when you follow the anime through the episodes, you meet the characters (especially newly introduced ones) that you are not sure in which category they fall; are they human, half-human, nearly-human or android? Who has a normal brain, cyberbrain or an AI? It leaves you guessing and that is a weird feeling. Something evenly weird happened when I phoned the customer support of my TV and internet provider regarding the delivery of a modem. While I had received a notification that it was signed-for and delivered, it was nowhere to find; it was not with my neighbours or me! After the mandatory waiting time, I was put through. Someone started to talk with me, telling stuff and asking questions. Due to the way he spoke, his monotonous voice and the things he said (very script-like), some weird feeling came over me: for the very first time in my life, I was not sure if I was talking to an actual human or a very smart AI bot. My suspicion went back and forth until we discussed the signature on the delivery-form. He described the shape and agreed that it was a weird shape to use as a signature on the delivery-form. Because he could describe the shape and had an opinion about it (something that is very hard to do for a program), from that moment I was convinced, that I was talking to a human brain. It is a weird feeling if you don’t know you're communicating with a human or AI. And this will become reality in the future, maybe sooner than later. A term that is often used in these kind of situations is the ‘uncanny valley’ feeling. Some time ago the robotics professor Masahiro Mori coined this concept: when a robot, android or a CGI character is very realistic, but not realistic enough, then it leads to weirdness or even revulsion. However in the case of the Anime, all characters and persons looked and sounded completely human; it is just that you are not sure they are human. So it hasn't to do with the uncanny valley concept; everybody looked convincingly human. It isn't also the ELIZA effect, when people contribute human thoughts and feelings to computers or robots, because you are not interacting with a computer or robot. It is more a kind of Turing test but then during real life! But it gives rise to another interesting question: should we care? If we can't tell if our conversation partner is human or not; then, maybe it doesn't really matter? Last year, I got hailed by a chatbot on a well-known big blue retail site and I was shocked by the stupidity of the bot. Come on: it is 50 years after ELIZA and then you deliver a chatbot that is way dumber than ELIZA! It also shows how clever ELIZA was constructed. For all its faults and simplicity, ELIZA is the standard that a chatbot or AI at least must pass. ~If you liked the article, click the 💚 below so more people can see it! Also, you can follow me on Medium, so you get updates regarding future articles!~

Cogito ergo sum — It thinks therefor it is

26

cogito-ergo-sum-it-thinks-therefor-it-is-178cbc254bc6

2018-04-07

2018-04-07 10:39:25

https://medium.com/s/story/cogito-ergo-sum-it-thinks-therefor-it-is-178cbc254bc6

false

1,306

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Hans Bruins

Thinker, inker, keyboard extrovert, truth-seeker and ignorance disintegrator

daf1848139a0

hans.bruins

74

5

20,181,104

null

0

null

0

null

2018-07-11

2018-07-11 12:38:04

2018-08-01

2018-08-01 05:08:29

2

false

en

2018-08-01

2018-08-01 05:08:29

0

178d757cebc1

2.258805

0

What actually is the internet of things?

5

Internet Of Things : IOT What actually is the internet of things? The IOT is a web of physical objects or “things” that are connected to each other allowing them to collect and exchange of information with the help of sensors i.e; it is inter networking of devices and software. It allows objects to be controlled remotely which creates great opportunities for physical world to get integrated into computer based systems which enhance efficiency and accuracy. The basic components of newer version of IOT includes autonomous control and ambient intelligence i.e.responsiveness of a computer system in presence of a group of people. The current state of IOT Unlike the other game changing revolutionary technology such as quantum computing, which now a days working in mechanical field to fight against the huge complexities, the IOT is already affecting human lives. It offers advanced technological connectivities and is heralding the revolution in the machine to machine to machine communication. Application of IOT Manufacturing: It enables rapid manufacturing as it connects all machinery together as central networking system. Agriculture: It is one of the first industry to use the idea of IOT.Wireless sensors are used to give information regarding weather forecasting or collect climate changes and other environmental causes and issues. Client: Consumer connected devices which includes smart TV’s,smart speakers,toys and other smart applications.Smart meters,commercial security systems and smart city technology such as those use to monitor traffic and weather forecasting are most common examples of industrial and enterprise IOT devices. IOT device management A number of possible changes that can make it difficult for an organisation to deploy a successful IOT system and its connected devices.Many of these can be address with IOT device management either by adopting standard protocols or using services offer by vendor. IOT devices management software are also available from commercial traveller i.e; vendor including amazon,microsoft,bosch software innovations. How IOT can be secured? The IOT is connecting people,places and devices at a rapid pace.With the search of connected devices comes the demand and necessity to implement secutiy features for IOT devices. Qualcomm technologies has long heritage of providing mobile security solutions. Today,are security solutions are found in billions of commercial devices around the world utilising are proven mobile solutions for consumer and industrial IOT But it suffers platform fragmentation i.e; some use newer version of IOT whereas others use older version of IOT which leads to lack of technical standards which make developing applications difficult.The IOT has enormous beneficial opportunities for customer and consumer but its evolutionary changes have been affected by product development challenges,such as rapidly changing requirements,consumer expectations,pricing and heavy competitions. And it continues to be the same As widespread consumer adoption of connected devices increases,so does the pressure on customers to create differentiated,high end devices that are powerful,yet power efficient,human efficient and highly secured. This is precisely what IOT platforms do for us.It enables devices and objects to observe,identify and understand the situation or the surroundings without been dependent on human help.

Internet Of Things : IOT

0

internet-of-things-iot-178d757cebc1

2018-08-01

2018-08-01 05:08:29

https://medium.com/s/story/internet-of-things-iot-178d757cebc1

false

497

null

Internet

internet

Internet

29,573

MANTHAN SOLANKI

null

a6448b7bcd0f

technologist1403

0

1

20,181,104

null

0

null

0

19ca4e270259

2018-05-16

2018-05-16 08:02:53

2018-05-16

2018-05-16 11:26:38

18

false

en

2018-05-24

2018-05-24 18:38:17

15

178dfdfd48ce

8.446226

4

0

Body tracking or pose recognition has been a field of interesting work for a long time. Usage includes gaming(Kinect), wearable-free…

3

AI for Skeletal Tracking Using Synthetic Data Body tracking or pose recognition has been a field of interesting work for a long time. Usage includes gaming(Kinect), wearable-free actions, activity detection and prediction( person of interest =P). Turns out it is not tech that is easy to come by unless you have a Kinect. So we decided to rebuild our version of it. Just to see how Microsoft achieved that kind of accuracy. The Deal: We are going to make an AI which can track human body parts/pose in 3D, ie x,y and z. So you know how far the person is and in what pose. As part of the training data required we are going to generate millions of images on our system in contrast to expensive methods used in studios. The AI is ‘tricked’ into learning artificially generated data. While it operates with real world data to which, surprisingly, it has shown remarkable results. Here is a sample of our work : Touch a cuboid in a virtual environment hands free 3D cameras 3D cameras offer much more information than 2D, because of point clouds. More data, better pattern recognition. example data from a 3D sensor SO, lets gets started. In our case we used a custom ToF based camera made in 1byZeroLabs Enter Microsoft's Kinect The Kinect v1 is based on structured light and v2 is based on ToF (Time of flight ) sensors. Both give 3D data as ouput. Their pose estimation system is based on “Real-Time Human Pose Recognition in Parts from Single Depth Images”, by Jamie Shotton Et al. The Kinect has been used a lot not just for gaming but for academic purposes also. It was the first time 3D cameras became available to normal consumers at a reasonable cost. LiDAR has been the choice for high resolution 3D scans but it comes at a high cost. The body tracking API in Kinect enabled many possibilities. I was curious how they made it work. So we decided to re implement it and check it out for ourselves. Most notable was their use of synthetic data. There are few studies on using synthetic data[2] and possible applications. Making Synthetic data for humans: Got to love open source. There are some excellent tools for doing just the thing. I chose Blender for the graphic pipeline and rendering, and Make Human for creating humans with skeletal rigs. Make Human[3] The interface for Make Human Alright so we got the human part done. We can export the model to Blender with rigs and skeleton. So how do we tell the machine which is what ? We color! ( label ) Body part map to wrap into our human model We color each part of the body , 42 , in our case. We label these parts for use later when training our system. Blender : Blender is an opensource 3D engine for creating and editing anything 3D. Another important feature we want to probably use later is that the pipelines are exposed via Python. This allows us to automate our rendering pipelines for huge projects. Allowing us to modify content or maybe reuse assets or bring variation to 3D content on the fly. Blender Interface We can have loads of variations : But since we are using a 3D camera. We need 3D data! Luckily Blender’s node based render engine allows us to use the z-buffer. Piping this data to be normalized and rendered as an image. So for every depth image we generate the corresponding labelled image. Here is what it looks like : Multiple types of body[6] The pure white areas denotes ‘infinitely FAR’, whereas, the darker gradients means those pixels( or area) are closer to the camera. In short we got ourselves 3D data where relative distance is represented by the intensity of the pixel. (0.0 to 1.0) Enter MOCAP ( Motion Capture ): So if our AI is going to learn parts of our body. It has to ‘see’ and understand realistic poses we do in our day to day lives or for a specific application (gaming for one;we mostly use specific actions like kick, punch, jump, dodge etc.) Thankfully the Make Human models are already riggable during export so we can plug it into our multi color human. Getting MOCAP turned out to be very expensive. Animated movies use sensors around the body to capture information. Like : Those sensors/receptors attached capture motion information BUT, memes these days >.> SO, we used Carnegie Mellon University’s Motion Capture Database[5]. We selected loads of actions to train our system. Pairs of images are created during render; The depth image and its ground truth. In the end we get life-like actions on our synthetic humans. ground truth on left, synthetic depth image on right Microsoft's Method: I will not go into detail about the paper[1]. The crux of it is the feature engineering and the use of Randomized decision forests. Feature engineering : Feature function[1] Randomized decision forests: Representation of a forest[1]. Randomized decision forest are chosen for their speed and efficiency especially on a GPU. A forest is an ensemble of trees. Each split node consists of a feature f and a threshold t. To classify pixel x in image I, the current node is set to the root, and then the previous feature equation is evaluated.The current node is then updated to the left or right child according to the comparison f(I, x) < t, and the process is repeated until a leaf node is reached. At the leaf node reached in tree t, a learned distribution Pt(c|I, x) over body part labels c is stored. The distributions are averaged together for all trees in the forest to give the final classification. part wise probability distribution Very democratic, must say. But, It also needs a huge database( which isn't a problem now with our home brewed synthetic data) and huge availability of RAM( I wasn't able to make this an out-of-core operation). My 64 GB RAM is puny for a million images decompressed into the RAM. For forests, the data has to be entirely loaded into the RAM before sampling from them. This posed a challenge for me due to lack of resources. ( it took Microsoft lots of hardware and a week to get this model trained with millions of images) Enter Deep Neural Networks and Fully Convolution Networks: Convnets (Convolution Networks)are driving huge advances. They are inspired from the receptive signals studied and modeled from the visual cortex of a mouse. They have proven well in local tasks with structured output. Semantic Segmentation : This allows us to classify parts of the image. So we can identify multiple objects within one image itself. The pixel wise prediction is what makes this possible. FCN 8 results from [8]. [8]FCN or Fully Convnets are used for pixel by pixel based segmentation. Probability distributions are directly connected to ‘what part does that pixel belong to’.Locations in higher layers correspond to the locations in the image they are path-connected to, which are called their receptive fields. Convnets are built on translation invariance. Their basic components (convolution, pooling, and activation functions) operate on local input regions, and depend only on relative spatial coordinates. Figure from [8] :FCN heatmaps show where the highest probability is. The dense body part convnet: Among a lot of nets I’ve tried, the work from Alireza Et al.[7] was the most interesting. After a little preprocessing the results were amazing. Here is the net : from [7] Input 250x250 normalized depth image. Deconvolution kernels fused with information from lower layers help generate densely classified pixel depth. With input from our 3D sensors made with love in 1byZeroLabs we were able to interface and get predictions : Output after segmentation. Notice how the AI segments even when your back is shown. Improvements and Possibilities 1. Convert to C++: Its crazy how much you can optimize if you use C++. Takes time to build, but cant beat performance once its done. I initially converted parts of the code into C++ libraries (.so and .dll). This was so that the heavy I/O, pre-processing etc. could be faster and parallelized. The Point Cloud Library is extremely useful for Point cloud operations. I was able to manipulate and run some de-noising operations at very high speeds once my bindings were based on C++. 2. Plug into Unity3D: Unity3D is very powerful game engine with a good level of abstraction if you are new to it. We managed to call our C++ libraries from Unity3D allowing us to use the results of our AI and drive some action. 3. Redo the render pipelines and automation scripts Using the tools available now in Unity3D allows us to automate the render pipelines and add variations much more easily. This is something I’ve been meaning to look into. Mixamo, includes and huge library of rich Motion data. Something that is also easily plug-able to the Unity environment. 4. Apply temporal features to train the AI for better recognition Just theorizing here: The current algorithm predicts frame by frame. There is no previous memory of the on going action. This works well, but adding memory (LSTM) or temporal features COULD improve the accuracy over a couple of frames. 5. Using the AI only for fallbacks Optical flow can be implemented to help performance. The body parts can be predicted and then tracked using optical flow algorithms till the confidence values of the tracked areas fall below threshold. 6. Possibilities with other synthetic data You can basically implement this process on any ‘view’. If you just want to train your model for hands/fingers ( LeapMotion much ? =P ) you need to fix the camera view on the head and simulate finger movements. Or, you can generate data for identifying and motion planning a robot’s arm as it looks through its camera. Thanks and Acknowledgements I would like to thank the open source community and the collection of wonderful and very powerful software available to us due to their contributions. Vimal Kaul, from Zayyon labs who had extended tremendous support for carrying out this research. Alireza Shafaei, for being kind enough to help me out through emails. To our team in 1byZeroLabs, and special thanks to Alekh, Ananth Sridhar and Sriram Harvind for helping out when it was getting rough. References [1] Jamie Shotton Et al. “Real-Time Human Pose Recognition in Parts from Single Depth Images” [2]Weichao Qiu, “Generating Human Images and Ground Truth using Computer Graphics” [3]Make Human http://www.makehumancommunity.org [4]Blender https://www.blender.org/ [5]CMU Mocap database http://mocap.cs.cmu.edu/ [6]Bengio, Yoshua, et al. “Curriculum learning.” Proceedings of the 26th annual international conference on machine learning. ACM, 2009. [7]Shafaei, Alireza, Little, James J.. “Real-Time Human Motion Capture with Multiple Depth Cameras”. 13th Conference on Computer and Robot Vision, 2016. [8]Jonathan Long,Evan Shelhamer and Trevor Darrell, “Fully Convolutional Networks for Semantic Segmentation” [9]L epetit, V., Lagger, P., Fua, P. Randomized trees for real-time keypoint recognition. In Proceedings of CVPR (2005). [10] V. Ganapathi, C. Plagemann, D. Koller, and S. Thrun, “Real-Time Human Pose Tracking from Range Data,” in ECCV,2012. [11]Jingwei Huang,David Altamar, “Pose Estimation on Depth Images with Convolutional Neural Network” Stanford university. [12] Unity3D, https://unity3d.com/ [13]Point Cloud Processing http://pointclouds.org/ [14] OpenCV https://opencv.org/ [15]Mingyuan Jiua, Christian Wolfa, Graham Taylorc, Atilla Baskurta,, “Human body part estimation from depth images via spatially-constrained deep learning” [16]Gilles Louppe, “Understanding Random Forests” , from theory to practice

AI for Skeletal Tracking Using Synthetic Data

18

ai-for-body-tracking-using-synthetic-data-178dfdfd48ce

2018-05-25

2018-05-25 13:27:57

https://medium.com/s/story/ai-for-body-tracking-using-synthetic-data-178dfdfd48ce

false

1,801

1byZ tech blog

null

1byZeroLabs

null

1byZeroLabs

null

1byzerolabs

null

1byZeroTechLabs

Machine Learning

machine-learning

Machine Learning

51,320

Kevin Roy

Jack of all trades. Master of none ?

562001cb61d7

kevz93g

39

53

20,181,104

null

0

null

0

null

2017-10-07

2017-10-07 12:13:56

2017-10-07

2017-10-07 12:18:18

16

false

en

2017-10-07

2017-10-07 12:18:18

0

178ec8e24621

1.791509

0

It’s nearly Halloween again and it’s nice to collect the mad realities of today’s data world, this time through the lens of memes. All of…

3

Mad World Of Analytics It’s nearly Halloween again and it’s nice to collect the mad realities of today’s data world, this time through the lens of memes. All of the pictures explain themselves rather well so I’ll keep my text to a minimum. On the other hand, don’t spare your own comments especially if you agree or disagree with them. It’s madness out there.

Mad World Of Analytics

0

mad-world-of-analytics-178ec8e24621

2018-03-12

2018-03-12 06:26:28

https://medium.com/s/story/mad-world-of-analytics-178ec8e24621

false

64

null

Data Science

data-science

Data Science

33,617

Dominic Ligot

Applied Analytics

9777f22b2997

docligot

10

20

20,181,104

null

0

null

0

7a72d0826cdc

2017-10-11

2017-10-11 07:58:21

2017-10-11

2017-10-11 08:06:07

1

false

it

2017-10-11

2017-10-11 08:06:07

0

17903ae8511e

2.120755

0

Moneyfarm acquisisce la tech company Ernest

5

L’intelligenza artificiale entra nella gestione dei risparmi Moneyfarm acquisisce la tech company Ernest Ernest Bot Moneyfarm, il gestore digitale del risparmio, ha acquisito il chatbot di personal finance Ernest e le tecnologie impiegate per il suo sviluppo e il suo funzionamento. Ernest è un personal banker alimentato da intelligenza artificiale che combina la tecnologia di elaborazione delle lingue naturali con l’apprendimento automatico. Moneyfarm diventa così il primo gestore digitale a esplorare l’utilizzo dei chatbot per la gestione dei risparmi e degli investimenti. Ernest è un progetto partito a Londra nel 2016 per iniziativa di tre sviluppatori italiani (Cristoforo Mione, Lorenzo Sicilia e Niall Bellabarba). Si tratta di un consulente finanziario personale, un “financial wellness coach” che, sfruttando i dati ricavati dalle transazioni bancarie dell’utente, è in grado di fornire, interagendo attraverso Facebook Messenger, statistiche e consigli personalizzati per una migliore e più efficiente gestione delle proprie finanze personali. Attualmente dispone di due modalità di funzionamento: risponde alle domande e invia notifiche proattive. Ernest permette di integrare i diversi conti di una persona così da offrire una panoramica unica di tutte le fonti di reddito e di spesa e creare con semplicità report personalizzati per monitorare, per esempio, l’evoluzione nel tempo della spesa in diverse categorie di beni o servizi. Una funzionalità particolarmente interessante in un contesto come quello attuale nel quale l’approvazione in corso della normativa PSD2 stabilirà che le banche dovranno, previa autorizzazione, rendere accessibili i dati relativi ai conti correnti dei propri clienti a terze parti selezionate da questi. Moneyfarm ha in programma di combinare la tecnologia di Ernest con la propria piattaforma che offre servizi di investimento. Con l’acquisizione della tecnologia di Ernest, Moneyfarm sarà in grado di fare un passo avanti nel suo percorso per accompagnare i clienti lungo l’intero ciclo di vita finanziario: dalla generazione dei risparmi fino all’investimento. La tecnologia Ernest, che ha la capacità di imparare e interagire con le abitudini finanziarie quotidiane del cliente, si integrerà con quella di Moneyfarm per permettere alla società di offrire una consulenza finanziaria ancora più personalizzata e costruita intorno alle abitudini di spesa di ciascuno. L’acquisizione di Ernest accresce inoltre il know how di Moneyfarm nel campo dell’intelligenza artificiale, know how che potrà essere impiegato per sviluppare molti altri aspetti del prodotto offerto da Moneyfarm, sempre nell’ottica di fornire ai clienti la migliore soluzione personalizzata per proteggere e far crescere i propri risparmi. Giovanni Daprà, co-fondatore e CEO di Moneyfarm, ha commentato: “L’intelligenza artificiale e un’interfaccia utente interattiva ci aiuteranno a migliorare la nostra capacità di profilazione dei clienti e quindi a offrire un servizio migliore, più completo e sempre più personalizzato. Stiamo lavorando all’integrazione della tecnologia Ernest nella nostra offerta di prodotti, grazie alla quale saremo in grado di assistere il ciclo di vita completo dei risparmi del singolo cliente, dal primo stipendio fino al pensionamento.” Niall Bellabarba, co-fondatore di Ernest, ha dichiarato: “L’acquisizione della tecnologia Ernest da parte di Moneyfarm offre un’ottima opportunità per portare la nostra visione ad un nuovo livello e creare un gestore finanziario virtuale più avanzato, basato sull’intelligenza artificiale.”

L’intelligenza artificiale entra nella gestione dei risparmi

0

lintelligenza-artificiale-entra-nella-gestione-dei-risparmi-17903ae8511e

2018-01-16

2018-01-16 23:26:22

https://medium.com/s/story/lintelligenza-artificiale-entra-nella-gestione-dei-risparmi-17903ae8511e

false

509

We hack your business to improve it…starting with a chatbot

null

hackbiz

null

hackBiz

hackbizit@gmail.com

hackbiz

MOBILE APP DEVELOPMENT,BUSINESS,HACKING,CHATBOTS,CHATBOT DEVELOPMENT

hackbiz_it

Chatbots

chatbots

Chatbots

15,820

hackBiz

La filosofia hacker è agire e pensare in modo diverso allo scopo di trovare soluzioni alternative ed innovative a problemi noti.

f84d831795f9

hackbiz

3

7

20,181,104

null

0

null

0

null

2018-08-28

2018-08-28 12:18:30

2018-08-28

2018-08-28 13:01:54

0

false

en

2018-08-28

2018-08-28 13:01:54

1

1790ce228ce1

0.615094

0

Millennials - this generation has been heralded for its collective innovation, love for tech and desire to learn new skills.

5

Millennials and Data Science Millennials - this generation has been heralded for its collective innovation, love for tech and desire to learn new skills. This set of persons have a great thirst for taking on challenges and on this journey, we meet different persons, network and grow. Jungle saw Data Science for Beginners and was one of first participants to send a mail and register. She is a banker who is hungry to learn and become better at what she does. She wants to be at the top of her game. She has learnt how to use R to analyze data. https://youtu.be/Bp9ygyBlN2w Be like Jumoke. Register and take the Beginners Course for Data Science today. After 5 weekends, you will be able to analyze data without anyone’s help. Registration is on for the September class. Please call Ayodeji on 0905 029 9919 or reach him on Whatsapp on 0815 909 8464 for more information. You can also send an email to info.futureofworkafrica@gmail.com to register.

Millennials and Data Science

0

millennials-and-data-science-1790ce228ce1

2018-08-28

2018-08-28 13:01:54

https://medium.com/s/story/millennials-and-data-science-1790ce228ce1

false

163

null

Data Science

data-science

Data Science

33,617

The Future Of Work

"The best way to create the future is to create it." This project seeks to prepare professionals for the future through the Application of Data Science and AI.

5d6e7364e6fa

info.futureofwork

6

2

20,181,104

null

0

null

0

ecca28658d43

2017-11-19

2017-11-19 18:24:00

2017-11-19

2017-11-19 18:30:35

1

false

en

2017-12-08

2017-12-08 13:37:15

4

1790dd8ecf8e

2.630189

24

2

0

Get ready to make room for Siri, Alexa, Cortana or Google at your conference room table. AI is about to head to the office.

5

Welcome to the Meeting, Alexa Get ready to make room for Siri, Alexa, Cortana or Google at your conference room table. AI is about to head to the office. Already these “voice assistants” are learning to recognize different voices within one conversation, and respond accordingly. When you ask Google Home about your calendar it will know to reply about the events on your calendar and not your wife’s calendar. Likewise when your wife queries Google home, it will respond using her data. Imagine what this could mean in the office. If you could include a voice assistant in your videoconferences and meetings, then each participant could use the voice assistant to interact with their own calendar, email, and task list as the meeting progressed. Imagine that Mary, Jane, Bob, and Steve are all on the same video call. Instead of fumbling around with calendars and availability times to schedule the next meeting, one of them could just say “OK Google, when is the next time that the four of us can meet for 30 minutes?” Or “Alexa, when is Bill Jones available to meet with the four of us?” The voice assistant would respond with dates and times, from which you could pick. And when you do pick a time, the meeting would be added to the calendar for Bill Jones and everyone at the current meeting. Makers of voice assistants have already started integrating videoconferencing into their systems. Google Home can initiate a voice call and Alexa uses Amazon’s new conferencing service to connect participants via the echo show. Extend this just a little bit more and voice assistants could be active participants in our video calls, standing by to complete the mundane tasks so the meeting stays on track. No longer will each meeting participant have to type To-Dos and Follow-Up items into their computers. Participants could simply ask Google Business — or whatever Google will call its Google Home once it hits the conference room — to add a task to their respective To-Do list. Or to mark an item complete if the task gets completed in the meeting or is deemed to no longer relevant to the project at hand. A wonderful side effect of having voice assistants join the workplace would be the elimination of laptop use in meetings and videoconferences. I am fully convinced that if people did not have their laptop computers with them — under the guise of doing some work in a meeting — that meetings would be about 50% shorter and about 200% more efficient. If you think such a scenario is way off in the future, you are wrong. The geometric growth associated with machine learning is already at a steep point in the growth curve, having accumulated knowledge, algorithms and methodologies over the past 20 years. It’s only a matter of months until Zoom, Cisco, Lifesize, or some other videoconferencing company announces an alliance or integration of an API that will allow users to include a voice assistant in their videocall. Or perhaps these voice-enabled assistants will simply be expanded more deeply into their company’s native conferencing application. Skype for Business and Cortana might go very well together! I think we are about two years out (or less) from being able to walk into a conference room and to simply say “I’m ready to start the meeting” and a voice enabled-assistant will make all the connections necessary to connect with the people who’ve accepted your meeting invite, regardless of the platform they are currently logged in and using. In fact voice-enabled assistants may be the only way that we can deal with some of the multi-platform confusion that is inherent in the modern workplace. (see Christianson’s Law of Communication Platforms). Hello Siri and Cortana. Goodbye interns. This article was originally published at Let’s Do Video

Welcome to the Meeting, Alexa

112

welcome-to-the-meeting-alexa-1790dd8ecf8e

2018-04-09

2018-04-09 12:55:20

https://medium.com/s/story/welcome-to-the-meeting-alexa-1790dd8ecf8e

false

644

Technology and Tales from the road

null

The Free-Range Technologist

jscottchristianson@mac.com

the-freerange-techologist

VIDEOCONFERENCING,TEACHING,PROJECT MANAGEMENT,EDUCATION,FUTURE OF WORK

jscottmo

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

J Scott Christianson

UM Teaching Prof, State Tech College Regent, PMP, CISSP, Technologist & Entrepreneur.

b46e4e2c5bd3

JScottMO

256

186

20,181,104

null

0

new_labels = (1 — ε) * one_hot_labels + ε / K

1

null

2018-09-09

2018-09-09 13:37:36

2018-09-10

2018-09-10 07:35:30

12

false

en

2018-09-23

2018-09-23 07:20:51

15

17915421f77c

6.278302

1

0

In this story, Inception-v3 [1] is reviewed. By rethinking the inception architecture, computational efficiency and fewer parameters are…

5

Review: Inception-v3 — 1st Runner Up (Image Classification) in ILSVRC 2015 In this story, Inception-v3 [1] is reviewed. By rethinking the inception architecture, computational efficiency and fewer parameters are realized. With fewer parameters, 42-layer deep learning network, with similar complexity as VGGNet, can be achieved. AlexNet [2]: 60 million parameters VGGNet [3]: 3× more parameters than AlexNet GoogLeNet / Inception-v1 [4]: 7 million parameters With 42 layers, lower error rate is obtained and make it become the 1st Runner Up for image classification in ILSVRC (ImageNet Large Scale Visual Recognition Competition) 2015. And it is a 2016 CVPR paper with about 2000 citations when I was writing this story. (SH Tsang @ Medium) Image Classification Error Rate in ILSVRC 2015 ImageNet, is a dataset of over 15 millions labeled high-resolution images with around 22,000 categories. ILSVRC uses a subset of ImageNet of around 1000 images in each of 1000 categories. In all, there are roughly 1.2 million training images, 50,000 validation images and 100,000 testing images. About The Inception Versions There are 4 versions. The first GoogLeNet must be the Inception-v1 [4], but there are numerous typos in Inception-v3 [1] which lead to wrong descriptions about Inception versions. These maybe due to the intense ILSVRC competition at that moment. Consequently, there are many reviews in the internet mixing up between v2 and v3. Some of the reviews even think that v2 and v3 are the same with only some minor different settings. Nevertheless, in Inception-v4 [5], Google has a much more clear description about the version issue: “The Inception deep convolutional architecture was introduced as GoogLeNet in (Szegedy et al. 2015a), here named Inception-v1. Later the Inception architecture was refined in various ways, first by the introduction of batch normalization (Ioffe and Szegedy 2015) (Inception-v2). Later by additional factorization ideas in the third iteration (Szegedy et al. 2015b) which will be referred to as Inception-v3 in this report.” Thus, the BN-Inception / Inception-v2 [6] is talking about batch normalization while Inception-v3 [1] is talking about factorization ideas. What are covered: Factorizing Convolutions Auxiliary Classifiers Efficient Grid Size Reduction Inception-v3 Architecture Label Smoothing As Regularization Ablation Study Comparison with State-of-the-art Approaches 1. Factorizing Convolutions The aim of factorizing Convolutions is to reduce the number of connections/parameters without decreasing the network efficiency. 1.1. Factorization Into Smaller Convolutions Two 3×3 convolutions replaces one 5×5 convolution as follows: Two 3×3 convolutions replacing one 5×5 convolution By using 1 layer of 5×5 filter, number of parameters = 5×5=25 By using 2 layers of 3×3 filters, number of parameters = 3×3+3×3=18 Number of parameters is reduced by 28% Similar technique has been mentioned in VGGNet [3] already. With this technique, one of the new Inception modules (I call it Inception Module A here) becomes: Inception Module A using factorization 1.2. Factorization Into Asymmetric Convolutions One 3×1 convolution followed by one 1×3 convolution replaces one 3×3 convolution as follows: One 3×1 convolution followed by one 1×3 convolution replaces one 5×5 convolution By using 3×3 filter, number of parameters = 3×3=9 By using 3×1 and 1×3 filters, number of parameters = 3×1+1×3=6 Number of parameters is reduced by 33% You may ask why we don’t use two 2×2 filters to replace one 3×3 filter? If we use two 2×2 filters, number of parameters = 2×2×2=8 Number of parameters is only reduced by 11% With this technique, one of the new Inception modules (I call it Inception Module B here) becomes: Inception Module B using asymmetric factorization And Inception module C is also proposed for promoting high dimensional representations according to author descriptions as follows: Inception Module C using asymmetric factorization Thus, authors suggest these 3 kinds of Inception Modules. With factorization, number of parameters is reduced for the whole network, it is less likely to be overfitting, and consequently, the network can go deeper! 2. Auxiliary Classifier Auxiliary Classifiers were already suggested in GoogLeNet / Inception-v1 [4]. There are some modifications in Inception-v3. Only 1 auxiliary classifier is used on the top of the last 17×17 layer, instead of using 2 auxiliary classifiers. (The overall architecture would be shown later.) Auxiliary Classifier act as a regularization The purpose is also different. In GoogLeNet / Inception-v1 [4], auxiliary classifiers are used for having deeper network. In Inception-v3, auxiliary classifier is used as regularizer. So, actually, in deep learning, the modules are still quite intuitive. Batch normalization, suggested in Inception-v2 [6], is also used in the auxiliary classifier. 3. Efficient Grid Size Reduction Conventionally, such as AlexNet and VGGNet, the feature map downsizing is done by max pooling. But the drawback is either too greedy by max pooling followed by conv layer, or too expensive by conv layer followed by max pooling. Here, an efficient grid size reduction is proposed as follows: Conventional downsizing (Top Left), Efficient Grid Size Reduction (Bottom Left), Detailed Architecture of Efficient Grid Size Reduction (Right) With the efficient grid size reduction, 320 feature maps are done by conv with stride 2. 320 feature maps are obtained by max pooling. And these 2 sets of feature maps are concatenated as 640 feature maps and go to the next level of inception module. Less expensive and still efficient network is achieved by this efficient grid size reduction. 4. Inception-v3 Architecture There are some typos for the architecture in the passage and table within the paper. I believe this is due to the intense ILSVRC competition in 2015. I thereby look into the codes to realize the architecture: Inception-v3 Architecture (Batch Norm and ReLU are used after Conv) With 42 layers deep, the computation cost is only about 2.5 higher than that of GoogLeNet [4], and much more efficient than that of VGGNet [3]. The links I use for reference about the architecture: PyTorch version of Inception-v3: https://github.com/pytorch/vision/blob/master/torchvision/models/inception.py Inception-v3 on Google Cloud https://cloud.google.com/tpu/docs/inception-v3-advanced 5. Label Smoothing As Regularization The purpose of label smoothing is to prevent the largest logit from becoming much larger than all others: where ε is 0.1 which is a hyperparameter and K is 1000 which is the number of classes. A kind of dropout effect observed in classifier layer. 6. Ablation Study Ablation Study (single-model single-crop) Using single-model single-crop, we can see the top-1 error rate is improved when proposed techniques are added on top of each other: Inception-v1: 29% Inception-v2: 25.2% Inception-v3: 23.4% + RMSProp: 23.1% + Label Smoothing: 22.8% + 7×7 Factorization: 21.6% + Auxiliary Classifier: 21.2% (With top-5 error rate of 5.6%) where 7×7 Factorization is to factorize the first 7×7 conv layer into three 3×3 conv layer. 7. Comparison with State-of-the-art Approaches Single-Model Multi-Crop Results With single-model multi-crop, Inception-v3 with 144 crops obtains top-5 error rate is 4.2%, which outperforms PReLU-Net and Inception-v2 which were published in 2015. Multi-Model Multi-Crop Results With multi-model multi-crop, Inception-v3 with 144 crops and 4 models ensembled, the top-5 error rate of 3.58% is obtained, and finally obtained 1st Runner Up (image classification) in ILSVRC 2015, while the winner is ResNet [7] which will be reviewed later. Of course, Inception-v4 [5] will also be reviewed later on as well. References [2016 CVPR] [Inception-v3] Rethinking the Inception Architecture for Computer Vision [2012 NIPS] [AlexNet] ImageNet Classification with Deep Convolutional Neural Networks [2015 ICLR] [VGGNet] Very Deep Convolutional Networks for Large-Scale Image Recognition [2015 CVPR] [GoogLeNet / Inception-v1] Going Deeper with Convolutions [2017 AAAI] [Inception-v4] Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning [2015 ICML] [BN-Inception / Inception-v2] Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift [2016 CVPR] [ResNet] Deep Residual Learning for Image Recognition My Reviews Review: Batch Normalization (Inception-v2 / BN-Inception) -The 2nd to Surpass Human-Level Performance in ILSVRC 2015 (Image Classification) Review: PReLU-Net, The First to Surpass Human-Level Performance in ILSVRC 2015 (Image Classification) Review: GoogLeNet (Inception v1) — Winner of ILSVRC 2014 (Image Classification) Review: VGGNet — 1st Runner-Up (Image Classification), Winner (Localization) in ILSVRC 2014 Review of AlexNet, CaffeNet — Winner of ILSVRC 2012 (Image Classification)

Review: Inception-v3 — 1st Runner Up (Image Classification) in ILSVRC 2015

5

review-inception-v3-1st-runner-up-image-classification-in-ilsvrc-2015-17915421f77c

2018-09-23

2018-09-23 07:20:51

https://medium.com/s/story/review-inception-v3-1st-runner-up-image-classification-in-ilsvrc-2015-17915421f77c

false

1,306

null

Machine Learning

machine-learning

Machine Learning

51,320

SH Tsang

PhD, Researcher. I share what I've learnt and record what I've done. I hope you like my sharings. :)

aff72a0c1243

sh.tsang

195

16

20,181,104

null

0

null

0

f3225cc85e15

2018-02-25

2018-02-25 04:07:05

2018-03-01

2018-03-01 00:27:24

1

false

en

2018-10-03

2018-10-03 02:29:10

34

1791ad7dc3ae

2.769811

198

1

Google has some of the most talented AI research scientists, data engineers and data scientists in the world. Sundar Pichai the CEO of…

5

Google AI Interview Questions— Acing the AI Interview Google has some of the most talented AI research scientists, data engineers and data scientists in the world. Sundar Pichai the CEO of Google has focused to realign Google into an AI-first company. Google has weaved AI into all or most of its products from Gmail to Autonomous driving with the mass of data it possess. At Acing AI, the aim is to help you to get into Data Science and AI. I have profiled some of the best technology companies and written articles about AI interviews at Microsoft, Amazon, Netflix, LinkedIn, Ebay, Twitter, Walmart, Apple, Facebook, Zillow, Salesforce, Uber, Intel, Adobe Tesla and most recently IBM. This has led to being the top writer in Artificial Intelligence on Medium. The AI interview preparation guides Part 1, Part 2 go over the details which help you ace any AI interview. Acing AI Portfolios helps you to showcase your AI work. Expert interviews and analyses gives you a sneak peak into the lives of AI/Data Science Leaders and analyses of AI tech companies. Google AI related research has three major collections: Data Infrastructure and Analysis, Google Brain Team and Google AI Residency. Three areas with most number of AI publications at Google: Machine Intelligence Machine Perception Natural Language Processing Interview process Google’s technical interview process is a standard technical interview process. It consists of Phone screen/s followed by onsite interviews. For technical interviews they have their interview guide: here. TensorFlow Important Reading(About Google AI) TensorFlow: A system for Large Scale Machine learning. Tools that Google uses both Hardware and Software: AI Tools Unofficial Google Data Science Blog AI/Data Science Related Interview Questions What is the derivative of 1/x? Draw the curve log(x+10) How to design a customer satisfaction survey? Tossing a coin ten times resulted in 8 heads and 2 tails. How would you analyze whether a coin is fair? What is the p-value? You have 10 coins. You toss each coin 10 times (100 tosses in total) and observe results. Would you modify your approach to the the way you test the fairness of coins? Explain a probability distribution that is not normal and how to apply that? Why use feature selection? If two predictors are highly correlated, what is the effect on the coefficients in the logistic regression? What are the confidence intervals of the coefficients? K- mean and Gaussian mixture model: what is the difference between K-means and EM? When using Gaussian mixture model, how do you know it is applicable? (Normal distribution) If the labels are known in the clustering project, how to how to evaluate the performance of the model? You have a google app and you make a change. How do you test if a metric has increased or not? Describe the process of data analysis? Why not logistic regression, why GBM? Derive the equations for GMM. How would you measure how much users liked videos? Simulate a bivariate normal Derive variance of a distribution How many people apply to Google per year? How do you build estimators for medians? If each of the two coefficient estimates in a regression model is statistically significant, do you expect the test of both together is still significant? Source: Glassdoor Reflecting on the Questions Google is known for its dense interviews. There is a mix of questions from practical perspective, general ML perspective as well as a theoretical perspective. A well read candidate with a little bit of luck can surely make it into one of the most prestigious AI companies in the world. For a more consumable list of questions: 20 Google AI Interview Questions Thanks for reading! 😊 If you enjoyed it, test how many times can you hit 👏 in 5 seconds. It’s great cardio for your fingers AND will help other people see the story. Subscribe to our Acing AI newsletter, I promise not to spam and its FREE! Acing AI Newsletter - Revue Acing AI Newsletter - Reducing the entropy in Data Science and AI. Aimed to help people get into AI and Data Science by…www.getrevue.co

Google AI Interview Questions— Acing the AI Interview

1,185

google-ai-interview-questions-acing-the-ai-interview-1791ad7dc3ae

2018-10-03

2018-10-03 02:29:10

https://medium.com/s/story/google-ai-interview-questions-acing-the-ai-interview-1791ad7dc3ae

false

681

Acing AI provides analysis of AI companies and ways to venture into them.

null

Acing-AI-203608690388104

null

Acing AI

acingai101@gmail.com

acing-ai

ARTIFICIAL INTELLIGENCE,DATA SCIENCE,MACHINE LEARNING,STARTUP,TECHNOLOGY

Acing_AI

Machine Learning

machine-learning

Machine Learning

51,320

Vimarsh Karbhari

Engineering Manager | Udacity Deep Learning & AI(part1) Alumnus | Editor/Founder of Acing AI

825fa70f9e24

vimarshk

2,009

730

20,181,104

null

0

null

0

null

2018-01-05

2018-01-05 04:27:36

2018-01-13

2018-01-13 09:03:38

1

false

id

2018-01-13

2018-01-13 09:03:38

6

179247685bae

2.830189

4

0

Mengarang Lirik Lagu Berbahasa Indonesia Menggunakan LSTM di PyTorch

3

Mengarang Lirik Lagu Indonesia Menggunakan LSTM Hasil LSTM dengan kata awal “galau” Setelah membaca artikel understanding LSTM nya Christoper Olah yang membahas cara kerja LSTM secara rinci dan The Unreasonable Effectiveness of Recurrent Neural Networks nya Andrej Karpathy yang berhasil melatih LSTM yang dapat digunakan untuk “mengarang” berbagai macam teks, saya sempat dapat ide untuk membuat RNN yang dilatih untuk membuat teks lirik lagu berbahasa Indonesia. Sayangnya waktu itu belum ada GPU yang bisa dipakai. Sekarang setelah berhasil merakit deep learning rig, akhirnya project ini bisa dimulai. TL;DR Untuk mengarang lirik lagu (atau lebih tepatnya membuat model bahasa pada level karakter) saya menggunakan model char-rnn dan melatihnya banyak teks lirik lagu berbahasa Indonesia. Berikut ini adalah beberapa teks lagu karangan model yang sudah dilatih: Dapat dilihat dari hasil yang ditampilkan diatas bahwa“mengarang” sebenarnya bukan kata yang tepat untuk mendeskripsikan lirik lagu yang dihasilkan karena kalimat yang dihasilkan tidak masuk akal. Membuat neural network yang dapat menghasilkan kalimat yang dapat di mengerti oleh manusia masih merupakan salah satu topik riset AI. Namun untuk kata yang dihasilkan, sebagian besar merupakan kata alami yang biasa digunakan pada lirik lagu. Struktur teks yang dihasilkan juga terdapat kemiripan dengan teks lirik lagu pada umumnya. Hasil lainnya dapat dilihat di: generated — Google Drive More generated examplesdrive.google.com Dataset Untuk membuat model yang dapat mengarang teks lirik lagu Indonesia, tentunya kita memerlukan dataset yang berisikan kumpulan lirik lagu Indonesia dari berbagai macam judul dan artis. Sayangya dataset semacam itu belum ada, jadi harus didapatkan dengan cara lain. Cara mudah untuk membuat dataset tersebut adalah dengan melakukan crawling di situs web penyedia lirik lagu Indonesia. Dari beberapa situs yang ada, saya memilih liriklagu.co.id karena layout nya mudah untuk di crawl. Dari proses crawling berhasil didapatkan: 3.012 Artis 19.576 Lagu 12.264.283 Karakter Setelah itu dokumen-dokumen hasil crawling disatukan menjadi satu file .txt besar untuk mempermudah akses. Dataset nya dapat di download disini: dataset.txt Dataset Lirik Lagu Indonesia drive.google.com Model Character-level RNN atau Char-RNN merupakan RNN yang dilatih untuk memprediksi karakter selanjutnya dari karakter yang diinputkan. Walaupun terdapat “RNN” di namanya, yang sebenarnya digunakan adalah Long Short-Term Memory (LSTM) yang merupakan pengembangan dari RNN. Secara garis besar Char-RNN mirip dengan klasifikasi menggunakan neural network biasa, namun karena RNN mempunyai memory / state, prediksi akan dipengaruhi oleh karakter-karakter yang sudah diprediksi sebelumnya. Hal ini menyebabkan Char-RNN dapat belajar struktur kata dan struktur teks lirik lagu. Secara garis besar modelnya seperti ini: Embedding Layer LSTM Layer (3 layer) Linear Layer Cross Entropy loss (Soft Max + Negative log likelihood) Input berupa kumpulan indeks karakter. Embedding akan menghasilkan fitur yang lebih baik dibandingkan dengan input one-hot-vector dari indeks karakter. Karena tujuan dari Char-RNN merupakan klasifikasi biasa, maka digunakan Cross Entropy Loss. Untuk menghasilkan lirik lagu, char-rnn akan digunakan untuk memprediksi karakter yang akan muncul berikutnya dengan memasukkan kata awal. Prediksi dari karakter terakhir pada kata awal akan digunakan untuk menghasilkan karakter berikutnya. Pemilihan karakter berikutnya didapatkan menggunakan sampling sehingga karakter yang di pilih bukan selalu karakter yang mempunyai kemungkinan yang paling tinggi. Karakter yang dipilih kemudian akan dimasukkan kembali sebagai input. Proses memasukkan karakter hasil sebagai input diluang sampai jumlah karakter yang diinginkan dicapai. Untuk source code lengkapnya nya dapat dilihat di github: dieka13/indonesian-lyrics-generation indonesian-lyrics-generation - Indonesian Lyrics Generator With LSTMgithub.com Training Training dilakukan menggunakan GPU Nvidia GTX 1060. Proses training dilakukan dengan ADAM optimizer dengan learning rate awal 1e-4 selama 5 epoch. Total waktu training kurang lebih 28 jam, kurang lebih 6 jam untuk setiap epoch nya. Hasil Char-RNN berhasil mempelajari struktur kata dan teks lirik lagu Indonesia. Hal ini dapat dilihat dari kata dan struktur teks yang dihasilkan, contohnya adanya kalimat back to reff:, (korus), back to # yang biasa muncul pada teks lirik pada umumnya. Evaluation / Next Project(?) Dataset tidak di preprocessing sama sekali sehingga memunculkan hasil yang tidak diinginkan seperti adanya bahasa inggris dalam teks lirik (walaupun ada kemungkinan terdapat lagu yang memiliki sisipan lirik berbahasa inggris) dan lirik lagu religi yang tidak sesuai jika disisipkan pada lirik lagu mainstream. Menggunakan informasi tambahan seperti nama artis dan genre untuk membuat dataset dan model yang lebih spesifik. Mengarang lagu dari judul menggunakan sequence-to-sequence.

Mengarang Lirik Lagu Indonesia Menggunakan LSTM

57

mengarang-lirik-lagu-indonesia-menggunakan-lstm-179247685bae

2018-04-17

2018-04-17 11:48:48

https://medium.com/s/story/mengarang-lirik-lagu-indonesia-menggunakan-lstm-179247685bae

false

697

null

Machine Learning

machine-learning

Machine Learning

51,320

Dieka Nugraha K

ML enthusiast

58dc1a6b3c97

diekanugraha

23

24

20,181,104

null

0

null

0

b468e053644a

2018-02-17

2018-02-17 23:37:44

2018-02-17

2018-02-17 23:39:20

0

false

en

2018-02-17

2018-02-17 23:39:20

1

1792831c29bc

0.101887

0

In this episode, David talks about the history of automating work and the challenges of automating services.

3

S01E05 CX — History of automating work and experiences In this episode, David talks about the history of automating work and the challenges of automating services. Episode

S01E05 CX — History of automating work and experiences

0

s01e05-cx-history-of-automating-work-and-experiences-1792831c29bc

2018-03-03

2018-03-03 23:42:58

https://medium.com/s/story/s01e05-cx-history-of-automating-work-and-experiences-1792831c29bc

false

27

a double entendre where point can be interpreted both as the moment in time of or the meaning to struggle — our focus is on the nexus of user experience and artificial intelligence

null

the point of struggle

gonzo@ziff.io

the-point-of-struggle

UX,AI,CUSTOMER SUCCESS,PRODUCT DESIGN,DESIGN THINKING

pointofstruggle

Business

business

Business

153,000

David "Gonzo" Gonzalez

Data Scientist, Storyteller, LEGO Coach

573cab224fc

datagonzo

240

4

20,181,104

null

0

null

0

null

2018-03-18

2018-03-18 05:23:00

2018-03-18

2018-03-18 06:39:35

2

false

en

2018-03-18

2018-03-18 07:32:47

3

1796e0ade573

1.764465

2

0

18 years a go, I just finished my study @ Theoretical Physics Department, Brawijaya University. My Bachelor of Science thesis is…

5

Road to Theory of Everything 18 years a go, I just finished my study @ Theoretical Physics Department, Brawijaya University. My Bachelor of Science thesis is “Generalisasi Persamaan Maxwell Dalam Medan Gauge Non-Abelian dan Analognya Pada Relativitas Umum”, Generalisation of Maxwell’s Equations in Non-Abelian Gauge Field and Their Analogies in General Relativity. There are two major streams to explain Theory of Everything, Standard Model and String Theory. This theoretical research is a journey to unite all forces in universe into a generalisation theory that can can explain anything of everything in path of Standard Model. To describe the mathematics formulation of the Standard Model, physicists use group theory. “In mathematics and abstract algebra, group theory studies the algebraic structures known as groups. The concept of a group is central to abstract algebra: other well-known algebraic structures, such as rings, fields, and vector spaces, can all be seen as groups endowed with additional operations and axioms. Groups recur throughout mathematics, and the methods of group theory have influenced many parts of algebra. Linear algebraic groups and Lie groups are two branches of group theory that have experienced advances and have become subject areas in their own right.” — Wikipedia, Group Theory To represent algebraic groups, typically used tensors. “In mathematics, tensors are geometric objects that describe linear relations between geometric vectors, scalars, and other tensors.” — Wikipedia, Tensors After this research, I never used tensors again, except when implementing Markov Chain Algorithm for the purpose of detecting typo correction of hotel name for Tiket.com. Until someday, I knew Tensorflow that introduced by the Google Brain team within Google’s Machine Intelligence Research organization. “TensorFlow is an open source software library for numerical computation using data flow graphs. The graph nodes represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) that flow between them.” — Github, TensorFlow Realising these things, I decided on my career to start focusing on exploring the unification of forces in Theoretical Physics and the development of generic (but powerful) machine learning algorithms using tensors base as representations of their mathematical formulas. So, the next articles will be studies of Group Theory and TensorFlow. Welcome to my world.

Road to Theory of Everything

2

road-to-theory-of-everything-1796e0ade573

2018-03-18

2018-03-18 11:37:27

https://medium.com/s/story/road-to-theory-of-everything-1796e0ade573

false

366

null

Machine Learning

machine-learning

Machine Learning

51,320

Handaru

I’m a Product Strategist. I love data and street food. As a Theoretical Physicist, I believe the power of imagination.

8fad38ba2265

handaru

416

2,074

20,181,104

null

0

null

0

null

2017-10-04

2017-10-04 07:44:52

2017-10-04

2017-10-04 08:08:09

2

true

en

2017-10-04

2017-10-04 08:08:23

0

1797425669c5

3.315409

5

0

What happened?

5

Would you get in a car that was Programmed to Kill you? What happened? Your parents were involved in a car accident this morning between two autonomous vehicles on their way to lunch. The Other driver, a young female, survived unscathed. The logs from the car reveal your parents had been travelling at a constant speed, with no obstacles in its path when the other car altered course and smashed into the side of them. Who is she? The young female is a member of a wealthy family, which meant that she had abortive collision insurance installed which crashed the car in such a way which ensured her survival without consideration of other cars that don’t have the same insurance. Why? The young woman’s vehicle sensed the collision and implemented an escape programme which calculated every possible escape scenario. The car on her left hand side had collision abortive insurance which meant the car was unable to veer that way. Unfortunately, your parents were the closest car without it. How? Welcome to the future and my biggest fear for the impending automation of industry and the growth of AI — human selfishness. This is a huge philosophical question which harks back to the trolly car problem The worry that machines and robots will take over the world is overblown and misleading, the reality is that they won’t. Most likely, at most, they will assist us in our everyday tasks, improving our performance far beyond what is currently imaginable. That’s not to say AI shouldn’t be questioned or kept in check The deeper question though delves into the possibilities that are about to emerge which creates far more questions that must be answered. Active assists on vehicles are great, anything which improves the performance of a vehicle beyond human capabilities should be utilised as extensively as possible as to ensure the lowest possible loss of life. But AI cannot be developed in such a way that preferential treatment is given to people who are able to afford programmes that preserves their life above others when faced with the probability of catastrophe As with the above, the extension of those possibilities to alternative realms is entirely imaginable. Autonomous vehicles are the easiest to imagine due to the fact they will be rolled out widely within the next decade. iRobot provides a window to future probabilities In the opening scene the robot saves the life of a one human more likely to survive than the other. This showcased the logical reasoning which I believe would be most sensible, but what if the the opposite could also be true? What if the wealthiest members of society wore hardware which superseded the logical reasoning of computers meaning they would save them regardless of the alternative possibilities? This brings forth a whole other level of moral reasoning which juxtaposes the current conversation of what happens if machines kill humans. Our fate could be decided by what we can afford And that is what we must contend with. Detrimental effects of automation are an inevitability of progress, but casualties that arise will be scrutinised fervently as their injuries will not have been caused by the actions of somebody else, they will have resulted as a direct consequence of the actions of a computer. The jump is stark Human progress has went from autonomy of self, to control of machines through to autonomy of machines which take control. The first two we were responsible for the consequences, the latter we are passengers to our own fate We have ceded control in hope that the time regained from undertaking menial and tedious tasks allows us to focus our effort on things that matter more, but it is essential that we first create a memorandum of rights which is universally applicable. In order to progress we need to be clear with the established protocol to understand the implications of technological development. Oversight after the fact will not be good enough And that is what we must understand. We need to know why something is going to happen before it does with automation. We need to see and hear about what will happen before it does because it is all predictable. It is predictable because every eventuality is predicated by the code we give it. Autonmous vehicles at the end of the day are still governed by the laws that we fix. Whoever writes the code could be playing god We shouldn’t fear AI or autonomy but we must ready for the way in which humanity utilises the most important tool of the 21st century At the end of the day we will only have ourselves to blame

Would you get in a car that was Programmed to Kill you?

15

would-you-get-in-a-car-that-was-programmed-to-kill-you-1797425669c5

2018-01-30

2018-01-30 05:58:42

https://medium.com/s/story/would-you-get-in-a-car-that-was-programmed-to-kill-you-1797425669c5

false

777

null

Self Driving Cars

self-driving-cars

Self Driving Cars

13,349

Chris Herd

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

da7b665f3cc7

ChrisHerd

31,328

3,629

20,181,104

null

0

null

0

e5dde54c0aec

2017-09-24

2017-09-24 21:16:03

2017-09-24

2017-09-24 21:19:53

3

false

en

2017-09-24

2017-09-24 21:19:53

0

179800051621

2.682075

0

The questions for artificial intelligence arises with the day dreams and imaginer. People have imagined of machines doing all their tasks…

1

What is Artificial Intelligence? The questions for artificial intelligence arises with the day dreams and imaginer. People have imagined of machines doing all their tasks very long ago with human like abilities. Which is also described in many historical pictures, paintings, stories and in many science fiction movies like automata etc. Leonardo da Vinci sketched designs for a humanoid robot in the form of medieval knight around the year 1495. Model of robot by Leonardo da Vinci Three laws for the artificial intelligence were derived 1. It must not harm the human beings or through any interactions. 2. It must obey the instructions given by the human beings. 3. It can protect its existence until the first and second laws does not conflict. And later in 18th and 19th century more progress made in this multi-disciplinary field. And in 19th century Alan Turing who was a computer scientist, mathematician, cryptologist, philosopher and theoretical biologist invented a Turing machine which uses mathematical algorithms to decode the information into the human readable data. And he is also known as the father of computer sciences and artificial intelligence. Presently more and more focus is on developing actual artificial intelligence which is something like equals to or may be more efficient than natural intelligence NI or human intelligence. Google Alpha Go develops an AI which beats the World GO champion. And like that Elon musk AI start up beats the Dota 2 Visual game. And after every passing day AI is becoming more powerful, intelligent and workable. And many other leading organizations also using machine learning and artificial intelligence algorithms like facial recognition, audio recognition and pattern studies in their products like mobile applications and smartphones like Google using its artificial intelligence and machine learning embeded Google assistant in its devices and in Android OS. Which makes your mobile experience more easiest than ever before. It makes your calendar and reminders management, notes organization, yours inquires and search queries easy and organized. You can ask Google assistant anything like any questions or weather forecast of past, present or future it will answer your searched queries very much efficiently. Other mobile apps like Prisma a photography editing tool which edits your photos after processing your picture and facebook which throws the ads on the basis of user usage which also uses the machine learning algorithms. And one biggest example of it is the Google search engine which is the far more efficient search algorithms ever build and implemented by the Google which shows all of your searches very efficiently and organized. You may have noticed that whatever you searched on Google it shows almost the exact matches of your searches. How’s that possible? It is because Google search using machine learning which is the sub field of artificial intelligence algorithm which make it all happened. But all these are not exactly the artificial intelligence. The actual artificial intelligence has yet to arrive very soon in near future which can do all of your tasks from daily working routines, cooking, hospitality, medications and surgery in hospitals in short it can do any tasks which it would be instructed to do. And that AI is self-learning on the basis of machine learning algorithms. Which can learn the data and information itself and performs the tasks accordingly. And that AI may can lead us as it may have ability to become more efficient than NI or natural intelligence.

What is Artificial Intelligence?

0

what-is-artificial-intelligence-179800051621

2017-09-24

2017-09-24 21:19:55

https://medium.com/s/story/what-is-artificial-intelligence-179800051621

false

565

Thinking on different aspects of humanity, living and sciences.

null

SOACH

hamzaabdullah009@gmail.com

soach

TECHNOLOGY,TECHNOLOGY NEWS,DEVELOPMENT,SCIENCE,MEDICAL

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Hamza Abdullah

ecca0fe6b54f

hamzaabdullah

42

1

20,181,104

null

0

null

0

null

2018-06-07

2018-06-07 05:51:04

2018-06-25

2018-06-25 11:22:13

1

false

en

2018-09-12

2018-09-12 09:49:30

22

1798197b648e

5.279245

2

0

An underutilized scientific theory that deserves a startup.

5

Hacking the Forgetting Curve An underutilized scientific theory that deserves a startup. In 2013, I blogged about a study technique based on an insight that there’s an optimal time to review what you learned. Review too early you’re wasting your time, review too late you’ve forgotten too much and have to relearn it. Review at the right time — when you still remember 80% of it — you reap the benefits of fast recall and longer-term memory at the same time. It’s the most efficient way to retain knowledge. The human memory follows a pattern of forgetting. German psychologist Hermann Ebbinghaus first discovered it in 1885 and named it the Forgetting Curve. In 1992, Polish researcher Piotr Woźniak wrote computer programs based on it to more accurately predict the curve’s knowledge retention. Since then, not much progress has been made. Relevant apps like Anki still have UIs stuck in 1995. Duolingo has some elements of it, but not in a significant way. Perhaps more pressingly, people are not aware of this “memory hack” that basically makes us more knowledgeable. Over the years I’ve always had this idea floating in my head, but it never occurred to me that it’s worth a startup, probably because it doesn’t have a clear way to make money. I’ve always hoped that someone would make something out of it. It’s about helping people remember things better. It’s about helping people learn the knowledge that matter. For God’s sake, it’s about making people smarter. We’re learning everyday Learning doesn’t stop after graduation — whether you like it or not. We’re still learning everyday. When you go on social media, you’re learning about your friends and acquaintances. When you read a news article, you’re learning about the state of the world. You read a Medium article, you’re learning about the ideas in it. You read some insightful comments on Hacker News, you’re learning about people’s ideas about an idea. You watch a video on YouTube that explains what blockchain is, you’re learning. You watch a video on your news feed that shows you how to make Russian tomato soup (it’s Borscht), you’re also learning. Some learnings are more important than others. The important learnings should probably stay in your long-term memory. Others, maybe not. Whatever you and I decide — we’re more helpless than we think we are. Our forgetting curve is killing our efforts We’re also forgetting everyday. Whether it’s an advice you read yesterday on the benefits of meditation, or a mindless comment by a stranger on your weird-looking hair, they all face the same destiny — to be forgotten. Around two-third (66.3%) of knowledge you attain today will be forgotten tomorrow. That is a scary figure. Perhaps you want to be better at your job. Maybe you’re a student and you simply want to ace that exam. It seems tragic to me that everyday we are given the knowledge that will materialize our goals only to be lost tomorrow. Fortunately with today’s scientific research and machine learning, we can more accurately than ever predict (thus recover from) when you’re going to forget the knowledge that matter to you. The anatomy of human forgetting is no longer a mystery. It has peaked interests in academia and the psychology community for decades. Yet, few outside the circle have even heard of the concept of the forgetting curve. The science and technology are here. We just need a good application. My motivation to solve this problem It is common wisdom that reviewing often would yield a better memory of knowledge acquired, but how practical is that? To make sure you absolutely will not forget any important knowledge, you can review everyday, but even med-school students struggle with that. What about just weekly or monthly revisions? Sure, but keeping track of everything you’ve learned is no easy task. When I was a student, I knew the benefits of reviewing notes, but in reality I was so often just overwhelmed. I had so many notes, which should I read? The next thing you know, I was procrastinating until the week before exam and had to relearn most of the materials. It was painful. Now as a founder, I also read many books and blog articles. I watch lectures and interviews online, and I attend sharing sessions of other founders offline. There is a lot of wisdom being passed around, but unfortunately, they too, are prone to be forgotten. Helping people and myself become more knowledgeable is a passion of mine for life. I guess having spent 4 years and nearly all my resources on my first ever project StudyBloc (pitch: Quora for universities) proved that. I’m just going back to my old but important topic. Challenges 1. Design for repetition Journalist Gary Wolf wrote in a piece he contributed to Wired Magazine: Our capacity to learn is amazingly large. But optimal learning demands a kind of rational control over ourselves that does not come easily. Even the basic demand for regularity can be daunting. If you skip a few days, the spacing effect, with its steady march of sealing knowledge in memory, begins to lose its force. Getting people to review things in general is hard. It’s boring. It seems redundant and feels like a lot of work. However, that sentiment is also based on the fact that people don’t know the benefits of efficient spaced repetition. A little education topped with delightful user experiences and gamification can solve this problem. If we can help people manage knowledge in a way that’s not overwhelming, reviewing will become a no-brainer. 2. Design for outcome How can we know that you’ve truly learned something? According to Robert Bjork from UCLA’s Learning and Forgetting Lab: Long-term memory can be characterized by two components: retrieval strength and storage strength. Retrieval strength measures how likely you are to recall something right now, how close it is to the surface of your mind. Storage strength measures how deeply the memory is rooted. The amount of storage strength you gain from practice is inversely correlated with the current retrieval strength. In other words, the harder you have to work to get the right answer, the more the answer is sealed in memory. The biggest conflict in designing any education app is between giving instant gratification and giving solid, long-term learning. With no immediate sense of achievement, people drop out too early; with no actual learning, well, we’re wasting everybody’s time. Perhaps quantifying the learning curve and engagement level of any material can solve this problem. If we can allow people to incrementally progress through difficulty based on their personal learning curves to the content, we can ensure true learning and stickiness at the same time. 3. Machine learning Everyone operates with different learning and forgetting curves. The good news is that researchers in academia have figured out really accurate ways to predict them. We just need to implement them in a real, consumer product. Depending on parameters such as difficulty of material, ability of person, and revision history of user, we can use machine learning models coupled with psychology theories to help people learn more effectively. How does that work in real-life? I don’t know, but it sounds like an exciting challenge. I guess the best way to figure it out is to dive right in. :) Credits If you’re interested in this topic, I encourage you to read Gary Wolf’s article in Wired Magazine first. He dives deep into the topic and summarizes it really well. Many points in this article are inspired by his article. I’ve read (or at least tried to read) many research papers on this topic, and Rob Lindsey’s PhD thesis paper stood out the most — in clarity as well as application insight. His paper has inspired me to dive into the problem. Lastly, I’d like to thank Nicolas Bustamante for the wonderful summary on this topic he wrote on Medium, which I read at the beginning of my research. It’s a good one to get started.

Hacking the Forgetting Curve

2

hacking-the-forgetting-curve-1798197b648e

2018-09-12

2018-09-12 09:49:30

https://medium.com/s/story/hacking-the-forgetting-curve-1798197b648e

false

1,346

null

Learning

learning

Learning

37,342

Harrison Lo

Creator of things, thinker of thoughts.

fba4471500eb

harrison0723

162

106

20,181,104

null

0

null

0

null

2018-03-29

2018-03-29 14:02:53

2018-03-29

2018-03-29 14:09:33

1

false

en

2018-03-29

2018-03-29 14:09:33

1

179d5d893f3d

2.543396

2

0

In last few weeks or so there is a lot of buzz about privacy breaching from your best friends Google and Facebook (and alikes — I am sure…

5

The biggest problem of FB and Google is not what you know they know about you. It’s what you don’t. In last few weeks or so there is a lot of buzz about privacy breaching from your best friends Google and Facebook (and alikes — I am sure there will be more of that coming). But nothing of this is in any way shocking if you at least a bit follow shady principles they were quite obviously using for years. Why did they do that? It’s their business model — follow the money they always said. Ok, to be honest, If I would be Google of Facebook I would also try to build the profile around you. And this means getting single point of data I can grab from you. Having a profile about you of course can help them to target ads better and be more efficient. But it also gives them a huge advantage in other fields like AI or behavioural and cognitive knowledge (do a home work and watch Ex Machine movie if you haven’t yet). Just an example what they surely collect is: What you write into the input box and then delete: well, was it something inappropriate — this says a lot about you. Every time you put wrong email into login field: now we have another email from you, also we know that you can do these mistakes. More obvious, but you use Facebook on a controlled matter, but you act way more loosely with your quite secretive Instagram account, that is… well well, connected your FB account. And also add WhatsApp here (right, this one is end-to-end encrypted ;) But the biggest problem is the data they collect and you have no clue about. Let me make you paranoid with two cases: 1. Browser fingerprinting provides all these shady things about you So without getting too much into tech details, browser fingerprinting tries to identify you as an single browsing person and then follow what you do on internet — where and what you browse. Google and FB are able to follow you even if you are not logged in to FB (or Google services). Even more, they can also follow you if you use separate browsers for the tasks and what’s even more problematic, they have great tools to follow you if you try to protect yourself from these technics. Now just imagine that you have one of these lonely moments and you go to some adult site, browsing a bit around. Well, you might write on FB that you prefer gentle loving, but you online behaviour tells a bit different story. And FB knows ;). Now extend this to some things that you might find even more private, like using Tor and you feel safe with it. Well think again. Learn more about browser fingerprinting. 2. Face recognition everywhere Ok I would do extensive face recognition on every pic ever posted on FB. This would give me a great opportunity to follow people all around the globe even on a shadow spots (so where they can’t follow you — yet; or correction — Google can though Android and FB though one of their apps). So imagine a nice Japanese tourist taking a photo of some monument in Barcelona. It also catches you there as an innocent passenger. She excitingly posts this pic on FB and well, FB probably know you are there too! You might not have problem with this, at least not yet. But think again about building full scale profiles of humanity. You owe this data, right? This is not something you gave them for free, right? (Don’t kill me, I have no time to edit this post)

The biggest problem of FB and Google is not what you know they know about you. It’s what you don’t.

2

the-biggest-problem-of-fb-and-google-is-not-what-you-know-they-know-about-you-its-what-you-don-t-179d5d893f3d

2018-03-30

2018-03-30 12:43:18

https://medium.com/s/story/the-biggest-problem-of-fb-and-google-is-not-what-you-know-they-know-about-you-its-what-you-don-t-179d5d893f3d

false

621

null

Privacy

privacy

Privacy

23,226

Jernej Adamic

CEO @zenodys, IoT heavy, Zen worshiper.

c424136bbbc0

jernejadamic

269

712

20,181,104

null

0

null

0

null

2018-01-06

2018-01-06 16:36:33

2018-01-07

2018-01-07 13:11:27

3

false

en

2018-01-07

2018-01-07 13:11:27

7

179e12445c5a

7.142453

4

2

0

I am neither a data scientist nor a programmer. However, driven by curiosity and the amount of resources available online, I embarked on a…

5

From scratch — An LSTM model to predict commodity prices Image by François Deloche, via Wikimedia Commons I am neither a data scientist nor a programmer. However, driven by curiosity and the amount of resources available online, I embarked on a mini-project to build a machine learning system that can predict a commodity’s price at some future time period. The particular case being - Brent crude prices for the next month. This article is an explanation of how I went about the process and the model that was finally built. How can you use this article? I hope it serves as an introductory guide for you to work with LSTM models. Why are LSTMs important — because they are the practical way of implementing Recurrent Neural Networks (RNNs) and RNNs hold a lot of promise as explained in this wonderful article by Andrej Karpathy. You may also use this code on a new dataset for your own application. I would be happy if it is of any help to the reader. I have uploaded the code and the dataset used for this project in a Github repository. Do go through the dataset which is a .csv file so that you understand this example better. If you need more information on these concepts, I am listing below the resources that helped me a lot and have also served as inspiration for me. You can go through them for in depth understanding. Frank Kane’s book on Python programming Dr. Jason Brownlee’s website Siraj Raval’s youtube channel As previously mentioned, I am not a data scientist or a programmer, however I had completed the very popular Andrew Ng Coursera course two years back and have a basic understanding of machine learning. Also as an engineer I have learnt C programming in the past and can understand and do programming. In my opinion though, even without past formal training in machine learning you can build this system I am describing in this article. Of course some basic coding skills are helpful for such a project. My idea for this project was pretty straightforward. The prices of commodities are obviously linked to the global economy in general and of course supply-demand dynamics. Both these aspects are reflected in the behavior of price movements of commodities. So, there must be a relationship between the past price movements of Brent crude and other commodities with the future price of Brent over a short horizon, like one month. Brent price for next month = f(Price trend of Brent crude and other commodities up till this month) Thus I explored, using this data available to the current month how well can the price for the next month be forecast. On doing some research and reading online it was evident that most such implementations are being done on Python - the programming language. For those who are new to all this, Python is an open source programming language, very popular in the machine learning universe. It is a high level language, i.e, there are many libraries available from which you can use functions already written for complex tasks, which means you can focus more on your actual end goal. This is excellent for business managers who are more interested in quickly getting business value than in getting entangled in time consuming for loops! Let us go through the Python code segment by segment and understand how this is implemented. If you want to replicate this, simply copy pasting the code will work. # load required libraries from numpy import concatenate from pandas import read_csv from pandas import DataFrame from pandas import concat from matplotlib import pyplot from sklearn.preprocessing import MinMaxScaler from sklearn.metrics import mean_squared_error from keras.models import Sequential from keras.layers import Dense from keras.layers import LSTM import numpy as np The above segment starts with importing the libraries that will be used in the rest of the code. “numpy” and “pandas” are frequently used Python libraries for mathematical calculations. “matplotlib” is a library for plotting graphs. “sklearn” is again a library for mathematical functions. “keras” is a machine learning library and which is key for this project. The “keras” library is what we use to create the LSTM model and train it. Next we define certain global values that will be used by the code. num_features = 53 #Number of features in the dataset lag_steps = 1 #Number of lagged time features to be generated label_feature = ‘POILBRE’ #The column in dataset that model is being built to predict Here the number of features are nothing but the number of columns in the dataset. Each column is a commodity with its prices in the rows, there are 54 columns in the dataset but the first column is just the month and year data. The lag_steps is the number of months of history that we want to be used as input for the prediction model, here we are using one month back. The label_feature is the commodity whose price we want to predict, in this case the price of Brent oil. Next we define a function that we use to prepare our dataset using the number of lag_steps we have set. # This function arranges the dataset to be used for surpervised learning by shifting the input values of features by the number # time steps given in lag_steps def sequential_to_supervised(data, lag_steps = 1, n_out = 1, dropnan = True): features = 1 if type(data) is list else data.shape[1] # Get the number of features in dataset df = DataFrame(data) cols = list() feature_names = list() for i in range(lag_steps, 0, -1): cols.append(df.shift(i)) # This will be the shifted dataset feature_names += [(str(df.columns[j])) + ‘(t-%d)’ % (i) for j in range(features)] # Names of the shifted features for i in range(0, n_out): cols.append(df.shift(-i)) if i == 0: feature_names += [(str(df.columns[j])) + ‘(t)’ for j in range(features)] # Names of the shifted features else: feature_names += [(str(df.columns[j])) + ‘(t+%d)’ % (i) for j in range(features)] # Names of the shifted features agg = concat(cols, axis=1) agg.columns = feature_names if dropnan: agg.dropna(inplace=True) return agg Next we read in the .csv file containing our dataset, convert it using the function defined above and also scale the data so that all columns are with values 0 to 1, this is important to train the model. We move the label column that we would like to predict to the end of the dataset. # Reading in the dataset which is in .csv format, has column headings and has an index column dataset = read_csv(“Dataset.csv”, header = 0, index_col = 0, squeeze = True, usecols = (i for i in range(0, num_features+1))) supervised_dataset = sequential_to_supervised(dataset, lag_steps) # Move label column to the end of dataset cols_at_end = [label_feature + ‘(t)’] supervised_dataset = supervised_dataset[[c for c in supervised_dataset if c not in cols_at_end] + [c for c in cols_at_end if c in supervised_dataset]] # Dropping the current timestep columns of features other than the one being predicted, which will be the label or y supervised_dataset.drop(supervised_dataset.columns[(num_features*lag_steps) : (num_features*lag_steps + num_features -1)], axis=1, inplace=True) #print(supervised_dataset.shape) # Used for debugging scaler = MinMaxScaler(feature_range=(0, 1)) supervised_dataset_scaled = scaler.fit_transform(supervised_dataset) # Scaling all values Then we split the dataset in a 80:20 ratio. The first 80% of the data will be used for training the LSTM model and the remaining 20% for testing and validating the trained model. Reshaping is carried out because the LSTM model requires input data in 3D format. split = int(supervised_dataset_scaled.shape[0]*0.8) # Splitting for traning and testing train = supervised_dataset_scaled[:split, :] test = supervised_dataset_scaled[split:, :] train_X, train_y = train[:, :-1], train[:, -1] # The label column is separated out test_X, test_y = test[:, :-1], test[:, -1] train_X = train_X.reshape((train_X.shape[0], 1, train_X.shape[1])) # Reshaping done for LSTM as it needs 3D input test_X = test_X.reshape((test_X.shape[0], 1, test_X.shape[1])) Now we come to the part of defining the LSTM network and training it. In machine learning training works by randomly initiating the model and calculating the loss from the model’s prediction, this loss is then minimized by updating the weight of the model using a method called gradient descent — which basically finds the minimum loss model. Here we use mean squared error to calculate loss and an optimizer called ‘adam’. The model trained and a graph of the training is displayed. # Defining the LSTM model to be fit model = Sequential() model.add(LSTM(85, input_shape=(train_X.shape[1], train_X.shape[2]))) model.add(Dense(1)) model.compile(loss=’mean_squared_error’, optimizer=’adam’) # Fitting the model history = model.fit(train_X, train_y, epochs=200, batch_size=175, validation_data=(test_X, test_y), verbose=2, shuffle=False) # Plotting the training progression pyplot.plot(history.history[‘loss’], label=’train’) pyplot.plot(history.history[‘val_loss’], label=’test’) pyplot.legend() pyplot.show() In the above segment you can observe certain parameters of the model being defined. The LSTM layer has 85 cells and a single cell in the output layer. The model is trained for 200 epochs with a batch size of 175. The more the epochs the more the model will be trained for, however it may start over-fitting the training dataset and will be inaccurate with the test dataset. One has to find the right set of parameters which depends on the model and dataset in question, it is also more of an art than exact science involving a lot of trial and error. Plot of training errors over epochs Finally, with the trained model we make a prediction using the features in the testing dataset and compare the predictions against the actual values. The root mean square error is calculated and displayed. # Using the trained model to predict the label values in test dataset yhat = model.predict(test_X) # Reshaping back into 2D for inversing the scaling test_X = test_X.reshape((test_X.shape[0], test_X.shape[2])) # Concatenating the predict label column with Test data input features, needed for inversing the scaling inv_yhat = concatenate((test_X[:, 0:], yhat), axis=1) inv_yhat = scaler.inverse_transform(inv_yhat) # Rescaling back inv_yhat = inv_yhat[:, num_features*lag_steps] # Extracting the rescaled predicted label column test_y = test_y.reshape((len(test_y), 1)) inv_y = concatenate((test_X[:, 0:], test_y), axis=1) # Re joing the test dataset for inversing the scaling inv_y = scaler.inverse_transform(inv_y) # Rescaling the actual label column values inv_y = inv_y[:, num_features*lag_steps] # Extracting the rescaled actual label column rmse = np.sqrt(mean_squared_error(inv_y, inv_yhat)) # Calculating RMSE print(‘Test RMSE: %.3f’ % rmse) pyplot.plot(inv_y, label = ‘Actual’) pyplot.plot(inv_yhat, label = ‘Predicted’) pyplot.legend() pyplot.show() Actual vs predicted for next month’s Brent crude price Is it a good enough prediction? You may be able to tune the model better by adjusting the hyperparameters. It is also good practice to evaluate other predictive models, for this particular case simple linear regression could also be effective. I hope this article has been of help, comments are welcome! You can find out more about me at LinkedIn.

From scratch — An LSTM model to predict commodity prices

22

from-scratch-an-lstm-model-to-predict-commodity-prices-179e12445c5a

2018-06-17

2018-06-17 07:47:56

https://medium.com/s/story/from-scratch-an-lstm-model-to-predict-commodity-prices-179e12445c5a

false

1,747

null

Machine Learning

machine-learning

Machine Learning

51,320

Vinay Arun

A supply chain professional with curiosity in data science and machine learning. https://in.linkedin.com/in/vinayarun

47ef133ad78b

vinayarun

90

41

20,181,104

null

0

null

0

da5918029fd3

2018-04-11

2018-04-11 20:40:23

2018-04-11

2018-04-11 21:05:07

1

false

en

2018-04-12

2018-04-12 22:15:08

1

179f5f74f7d6

2.403774

16

0

It seems like barely a week goes by without a headline warning about the risk of unbridled development of artificial intelligence, also…

5

Artificial Intelligence, Robots and Job Security It seems like barely a week goes by without a headline warning about the risk of unbridled development of artificial intelligence, also known as machine learning. We’re Running Out of Time to Stop Killer Robot Weapons: The Guardian At On-Demand Education Marketplace, we believe mankind must strike a balance between exploiting the benefits of AI and avoiding the most-negative consequences of aggressive development of thinking machines. Elon Musk and others (including the late Stephen Hawking) have spoken out about the dangers of rapidly advancing AI research as we move toward singularity — the point where machines overtake us on the intelligence scale. This month’s 50th anniversary of the release of “2001: A Space Odyssey” probably did little to ease Musk’s concerns. Remember HAL, the film’s terrifyingly polite, lipreading computer (with a Canadian accent) that decides that the crew on a mission to Jupiter is incompetent? “Dave,” HAL says to one of the astronauts. “This conversation can serve no purpose anymore. Goodbye.” Musk, the founder of SpaceX and co-founder of electric-car maker Tesla, says extreme development of AI is more dangerous than the threat of nuclear war. He has repeatedly called for AI research to be regulated. Facebook founder Mark Zuckerberg last year dismissed Musk’s doomsday comments as “pretty irresponsible.” [Which reminds me of an old joke about the future of air travel. “Ladies and gentlemen, this is your captain speaking. Welcome to the world’s first, fully automated passenger aircraft. Please sit back, relax and enjoy the flight for nothing can go wrong, go wrong, go wrong…”] All kidding aside, at ODEM we’re excited about putting AI to work. We’re moving ahead to use blockchain technology and artificial intelligence to improve the quality of interactions between students and educators toward making higher-quality education more accessible and affordable. ODEM will use smart contracts, a blockchain-based category of AI, to streamline and automate the laborious process of organizing and delivering in-person academic programs. We believe that artificial intelligence also has a role to play in guiding students in their choice of academic courses to insulate them from the danger of being displaced in the workforce by the application of artificial intelligence. While history shows us that technological advances tend to create more jobs than they eliminate over the longer-term, some job categories will be more vulnerable to disruption than others. In Thinking Machines: The Quest for Artificial Intelligence and Where It’s Taking Us Next, the author Luke Dormehl says society must do better at educating up-and-coming generations of workers. “Currently, education is stuck in the same Industrial Revolution paradigm it has been in for more than 100 years,” he writes. Dormehl argues that formal education is too focused on standardized training that doesn’t adequately prepare students for a working life in fast-changing industries. “In today’s world, learned skills routinely become obsolete within the decade their learned — meaning that continual learning and assessment is needed throughout people’s lives,” he writes. ODEM is empowering students to actively own their training and education. Our platform provides tools for students to proactively search globally and to register for relevant in-person courses to ensure they stay ahead in the evolving marketplace of skills. ODEM is actively harnessing the power of blockchain technology, latent market forces and the process of higher learning to help ensure that for millions of people around the best is, indeed, yet to come. Fasten your seatbelts. Rich Maaghul CEO

Artificial Intelligence, Robots and Job Security

515

artificial-intelligence-robots-and-job-security-179f5f74f7d6

2018-04-27

2018-04-27 02:27:43

https://medium.com/s/story/artificial-intelligence-robots-and-job-security-179f5f74f7d6

false

584

Unlocking higher education with blockchain technology

null

odemio

null

ODEM

info@odem.io

odem

BLOCKCHAIN,EDUCATION

odem_io

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Richard Maaghul

null

fc6b0ebf6394

richmaaghul

134

11

20,181,104

null

0

from __future__ import absolute_import from __future__ import division from __future__ import print_function import argparse import sys from tensorflow.examples.tutorials.mnist import input_data import tensorflow as tf FLAGS = None def main(_): # Import data mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True) # Create the model x = tf.placeholder(tf.float32, [None, 784]) W = tf.Variable(tf.zeros([784, 10])) b = tf.Variable(tf.zeros([10])) y = tf.matmul(x, W) + b # Define loss and optimizer y_ = tf.placeholder(tf.float32, [None, 10]) # The raw formulation of cross-entropy, cross_entropy = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y)) train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy) sess = tf.InteractiveSession() tf.global_variables_initializer().run() # Train for _ in range(1000): batch_xs, batch_ys = mnist.train.next_batch(100) sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys}) # Test trained model correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) print(sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--data_dir', type=str, default='/tmp/tensorflow/mnist/input_data', help='Directory for storing input data') FLAGS, unparsed = parser.parse_known_args() tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

12

null

2017-10-21

2017-10-21 04:25:55

2017-10-21

2017-10-21 04:26:08

1

false

en

2017-10-22

2017-10-22 05:21:19

8

17a14d6e7794

2.166038

5

0

This post is part of Month to Master, a 12-month accelerated learning project. For October, my goal is to defeat world champion Magnus…

5

M2M Day 354: The easy way to implement a machine learning model This post is part of Month to Master, a 12-month accelerated learning project. For October, my goal is to defeat world champion Magnus Carlsen at a game of chess. Today, I finished writing the small Python script that converts chess games downloaded from the internet into properly formatted data needed to train my machine learning model. Thus, today, it was time to start building out the machine learning model itself. Rather than starting from scratch, I instead looked for an already coded-up model on Github. In particular, I needed to find a model that analogizes reasonably well to chess. I didn’t have to look very hard: The machine learning version of “Hello World” is called MNIST, and it works perfectly for my purposes. MNIST is a dataset that consists of 28 x 28px images of handwritten digits like these: The dataset also includes ten labels, indicating which digit is represented in each image (i.e. the labels for the above images would be 5, 0, 4, 1). The objective is to craft a model that, given a collection of 28 x 28=784 values, can accurately predict the correct numerical digit. In a very similar way, the objective of my chess model, given a collection of 8 x 8 = 64 values (where each value is represented using 12-digit one-hot encoding), is to accurately predict whether the chess move is a good move or a bad move. So, all I need to do is download some example code from Github, modify it for my purposes, and let it run. Of course, there are still complexities with this approach (i.e. getting the data in the right format, optimizing the model for my purposes, etc.), but I should be able to use already-existing code as a solid foundation. Here’s the code I found: Tomorrow, I’ll take a crack at modifying this code, and see if I can get anything working. Read the next post. Read the previous post. Max Deutsch is an obsessive learner, product builder, and guinea pig for Month to Master. If you want to follow along with Max’s year-long accelerated learning project, make sure to follow this Medium account.

M2M Day 354: The easy way to implement a machine learning model

10

m2m-day-354-the-easy-way-to-implement-a-machine-learning-model-17a14d6e7794

2018-02-25

2018-02-25 10:22:20

https://medium.com/s/story/m2m-day-354-the-easy-way-to-implement-a-machine-learning-model-17a14d6e7794

false

521

null

Learning

learning

Learning

37,342

Max Deutsch

Obsessive learner and product builder. Founder at http://OpenmindLearning.com. Guinea pig for http://MonthToMaster.com. Get in touch at http://max.xyz.

86ff34e637cf

maxdeutsch

7,893

1,497

20,181,104

null

0

null

0

null

2018-09-04

2018-09-04 10:47:46

2018-09-04

2018-09-04 10:48:35

1

false

en

2018-09-04

2018-09-04 10:48:35

3

17a544eb4601

2.143396

0

Artificial intelligence or machine intelligence or as its commonly know A.I is changing the way we do things at a pace that is faster than…

2

Areas Uganda needs Artificial Intelligence…like yesterday! Artificial intelligence or machine intelligence or as its commonly know A.I is changing the way we do things at a pace that is faster than many had believed it would a few years ago. As its name suggests it is a phenomenon that enables machines to think for themselves without the direct input from human beings as has been the case since the first computers were made. The concept of computers thinking for themselves creates both excitement and fear in many people, personally I am terrified by the fact that machines can think for themselves, I have watched all the Terminator movies and the possibility of robots one day enslaving humans scares me. However I can not just dismiss the good things that AI has done and continues to do such as those mobile assistants like Siri and Google Assistant or its application in advertising, medicine, gaming just to mention but a few. As a matter of fact, I have often wondered whether we should adopt the use of AI in some of the services we use as Ugandans. Our service sector is generally not good and largely it’s the people to blame for this. Why then can’t we have machines take over some of the activities so as to get us the right kind of service delivery we deserve. I decided to pick out a few areas where I believe A.I can do us some good. Traffic Lights If you have moved through Kampala City during rush hour, it is common to find police officers directing the flow of traffic at a junction with traffic lights. This baffles me a lot and I know I am not alone. The police say that they do this because the lights are not able to make realistic decisions based on the situation. Most times, neither can the officers. Therefore, why not solve this problem by having intelligent traffic lights that can assess the situation and make intelligent informed decisions. National ID replacement God forbid you lose your national ID in Uganda because you will spend the greater part of a year waiting to have it replaced. Since all the information on the registered Ugandans was stored in a database, I would think it must be the human element that makes the process tedious and long. I also believe that having an A.I handling the replacement process of the IDs would take that process from a couple of months to a couple of days and maybe even a few hours. Read: Your side hustle is not too small for the Internet Sim Card registration I have always wondered whether it could be possible to use the data from the national ID registration to verify and register a new SIM card in an instant. I understand that it is a delicate verification process, however, removing the human element and having the super intelligent computers handle this could reduce that verification time to minutes, maybe even shorter. Do you think there are areas where A.I can make service delivery a lot easier and faster?

Areas Uganda needs Artificial Intelligence…like yesterday!

0

areas-uganda-needs-artificial-intelligence-like-yesterday-17a544eb4601

2018-09-04

2018-09-04 10:48:35

https://medium.com/s/story/areas-uganda-needs-artificial-intelligence-like-yesterday-17a544eb4601

false

515

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Daniel Odaka

null

26597769c9e0

dan.odaka

0

1

20,181,104

null

0

null

0

null

2017-09-06

2017-09-06 06:38:39

2017-09-06

2017-09-06 06:45:25

0

false

en

2017-09-06

2017-09-06 06:51:39

2

17a6590a38a

2.324528

0

The syllabus for IBPS PO and SSC online test preparation are all same with general awareness, English language, mathematics and few more…

1

Syllabus they include in IBPS PO and SSC GL The syllabus for IBPS PO and SSC online test preparation are all same with general awareness, English language, mathematics and few more but the pattern of each differs. That the IBPS PO conducts a Common Written Examination(IBPS CWE PO/MT) for highly talented and young job candidate. The exam is organized in two rounds followed by Preliminary Test, Main Exam and Interview and those who qualify all the three phases get placed on the most prominent position of Probationary Officer in government banking organizations. While SSC GL exams syllabus consists of four tiers i.e. Tier-I, Tier-II, Tier-III and Tier-IV. The detailed syllabus of are SSC GL is mentioned below according to Tier’s: 1. Tier 1 · General Intelligence & Reasoning · General Awareness · Quantitative Aptitude · English Comprehension 2. Tier2 · Quantitative Ability · English Language & Comprehension · Collection Classification and Presentation of Statistical Data · Measures of Central Tendency · Moments, Skewness and Kurtosis · Correlation and Regression · Probability Theory · Random Variable and Probability Distributions · Sampling Theory · Statistical Inference · Analysis of Variance · Time Series Analysis · Index Numbers · General Studies (Finance and Economics) 3. Tier 3 Total Marks assigned to the Descriptive paper in CGL TIER 3 is of 100 marks and topics included in exams, is Writing of an Essay/Passage of 250 words and a Letter/Application Writing of approximately 150 words. Letter Writing Precise Writing Application Writing Essay Writing 4. Tier 4 Two tests will be conducted in tier 4 · Computer Proficiency Test (CPT) — CPT will be conducted for all those aspirants who have applied for CSS (Central Secretariat Services). · Data Entry Skill Test (DEST) — DEST will be organized for all those aspirants who have applied for the post of Tax Assistant. If you want to prepare all on your own then you can seek for the Mock Test for IBPS PO sites which are helpful for self preparation. Here is the list of syllabus included in the IBPS PO exams: 1. Quantitative Aptitude Ratio and proportion Averages Time and work Speed, Distance and time Mixture and allegation Stocks and shares Percentages Partnership Clocks Volume and surface Area Height and Distances Logarithms Permutation and combinations Simple and compound interest Equations Probability Trigonometry Profit, Loss and Discount Mensuration Elements of Algebra · Data Interpretation Pie charts Tables Bar & Graphs Line charts 2. Reasoning · Verbal reasoning Analogy Classification Word formation Statement and conclusions Syllogism Statement and assumptions Statement and arguments Coding Decoding Blood Relations Passage and conclusions Alphabet test Series Test Number, Ranking and time sequence Direction sense Test Decision making test Figure series Input/output, Assertion and reasoning Sitting Arrangement · Non-Verbal Reasoning Series test Odd figure out Analogy Miscellaneous Test etc 3. English Language · Vocabulary Homonyms Antonyms Synonyms Word Formation Spelling · Grammar Spotting Errors Phrases and idioms Direct and Indirect speech Active/ Passive voice · Reading Comprehension Theme Detection Passage completion Topic rearrangement of passage Deriving Conclusion 4. General Awareness Current affairs related to national and international issues of last 6 months. Overview of Indian Financial System History of Indian banking system Recent credit and monetary policies Introduction to National financial institutions like RBI, SEBI, IRDA, FSDC etc and of International organizations like IMF, World Bank, ADB, UN etc Abbreviations and Economic terminologies Banking Terms Important Government Schemes on capital & money market. 5. Computer Knowledge Basics of Hardware and software Windows operating system basics Internet terms and services Basic Functionalities of MS-Office (MS-word, MS-Excel, MS-PowerPoint) History of computers Networking and communication Database basics Basics of Hacking, Security Tools and Viruses

Syllabus they include in IBPS PO and SSC GL

0

syllabus-they-include-in-ibps-po-and-ssc-gl-17a6590a38a

2018-05-21

2018-05-21 22:22:34

https://medium.com/s/story/syllabus-they-include-in-ibps-po-and-ssc-gl-17a6590a38a

false

616

null

Data Science

data-science

Data Science

33,617

pathantu

null

c089a58db3f9

pathantu12

0

1

20,181,104

null

0

null

0

ea8807d89dc4

2018-01-11

2018-01-11 19:16:37

2018-01-11

2018-01-11 21:10:45

4

false

en

2018-01-12

2018-01-12 08:36:15

5

17a994f90567

2.75283

1

0

Title: Weather Condition Prediction from Image

5

Week 8— Warmth of Image Title: Weather Condition Prediction from Image Team Members: Berk GÜLAY, Samet KALKAN, Mert SÜRÜCÜOĞLU E-mails Respectively: berkgulay.cs@gmail.com , abdulsametkalkan@gmail.com , mertsurucuogluu@gmail.com Welcome to our latest blog post! Finally, we have completed our project and obtained our latest results (at least for now). :) Before I mention our results and conclusions, I want to give a short info about our latest works from this week and previous weeks. Moreover I am also sure about that you wonder answers of those questions, What will be next?, In which platforms you can follow us? or Where can you find our latest works, datasets, papers? Let me briefly explain all. Over the previous weeks, alongside of completing and comparing our results from different algorithms, architectures and methods, we also prepared a short introductory video presentation for our project. We gathered our conclusions together , tuned our results and arranged all to present you. We organised our codes, dataset, CNN models and extracted feature models to share as well. We wrote a final paper to explain all details which we got over. Good news!!! We also opened our codes to you and you can follow us on Github as well. Next, we will be organizing our codes, manuals and introductory stuff on Github. Furthermore we will gather all of our blog posts together on our own Medium publication and you can find this Medium page in Github-Readme of our project later. Another good news from Twitter, “Deep Learning Turkey” page mentioned about our blog posts and did retweet our videos with our requests on their page. DeepLearningTR mentioned our Medium blogs and projects on their Twitter page Github (Please follow us there!) :=) berkgulay/WeatherPredictionFromImage WeatherPredictionFromImage - ML project to recognize/predict weather condition in given image (WarmthOfImage)github.com Introductory Video Presentation : Here our Project Paper ( with ins and outs ;) ) WarmthOfImage.pdf WarmthOfImage_Project_Paper Let me also briefly show our results and talk about our conclusions after all of these works; This is our latest CNN architecture which gives best result. Our best results. These two algorithms with shown architectures overperform. Some of our test results with respective images, Different test image’s classification results using CNN architecture shown above And lastly our conclusions at the end of Warmth Of Image project… In this work, we used supervised learning methods which are Convolutional Neural Network(CNN), Support Vector Machine, Decision Tree and Random Forest for image classification. CNN gives the best result with %64.63 and Random forest gives close and another best accuracy with %63.90. Also RF is really time efficient and easy to perform with right feature models. Other methods(DT or SVM) are not good enough for our task and slower. The hardest part is trying CNN architectures, because there are so many variations and trying each architecture takes too long(high time cost). If we had strong GPUs, we could try more different architecture and find better architectures maybe. Random Forest works faster(really time efficient and can be used in real time applications) and gives good generalized results at the end. For image description right features and methodologies lead really good results and in future works other descriptors can be researched and used. Cloudy class is tricky and hard to distinguish from other classes because all other weather conditions also include clouds. Taking it out or change with another air type may produce way better weather classification performance. Even if rain detection is hard to perform task, we could succeed it for most of the cases.

Week 8— Warmth of Image

8

week-7-warmth-of-image-17a994f90567

2018-02-12

2018-02-12 21:06:30

https://medium.com/s/story/week-7-warmth-of-image-17a994f90567

false

544

Our Machine Learning project’s blog page. The name of our project is “Warmth Of Image” and official name is “Weather Prediction From Image”. In our project, we aim to predict weather condition in given image/picture by using advanced algorithms and various descriptors or features

null

WarmthOfImage

berkgulay.cs@gmail.com

warmthofimage

MACHINE LEARNING,WEATHER,ARTIFICIAL INTELLIGENCE,PREDICTIONS,IMAGE PROCESSING

null

Machine Learning

machine-learning

Machine Learning

51,320

Berk Gülay

Computer Science&Engineering Student at Hacettepe University — Ankara

cbb8a25fccf7

berkgulay.cs

8

4

20,181,104

null

0

null

0

null

2018-07-20

2018-07-20 05:39:05

2018-07-17

2018-07-17 16:49:03

2

false

en

2018-07-20

2018-07-20 05:40:24

3

17abb7221a95

3.187107

0

With Big Data Analytics and Artificial Intelligence tools, the retail industry is changing its approaches to various facets of its…

5

New Paradigms in Retail with AI and Big Data With Big Data Analytics and Artificial Intelligence tools, the retail industry is changing its approaches to various facets of its operation. While the big brands are the leaders in trying out many innovative technologies, even small retail businesses are investing in these to reap the potential benefits. Big Data and AI Big Data enables business organizations to gain deeper insights into customer behavior, their preferences and expectations, competitor strategies. Retailers can better understand product movement and merchandise assortment needs. They can optimize pricing dynamically depending on various elements using price optimization software. Artificial Intelligence integrates many technologies like voice recognition, machine learning, IoT, robotics etc., to provide an enhanced customer experience. Big Data Analytics Big Data Analytics enables retail firms to gather data from varying sources and mine them to extract nuggets of information. These can then help them gain better perspectives on various aspects of their operations like product assortment, pricing, marketing strategies, customer experience, etc. Customer Experience Big Data technologies can help analyze historical behavior data, trends, specific customer preferences and their shopping patterns, identify high potential customers, and those who can become good future prospects. All these can help retailers design more personalized experiences for customers — for instance, knowing which product to promote to different customers, directing their attention to new products from their favorite brands or favorite product categories, offering special discounts etc. Inventory Analyzing product movement in your own stores across channels, studying competitor strategies, market trends and so on can help retailers make better product stocking decisions. They help improve merchandise assortment plans and inventory management. Pricing Strategies Here too, customer decisions, market trends and prices etc can be analyzed and better pricing decisions can be made in different segments — low price, good quality at competitive price, and premium brands and products with higher prices. There will be different segments of customers for each of these. Price optimization software can analyze relevant data and provide better insights for designing better and more dynamic pricing plans. Price optimization software can help retailers design better pricing, balance low prices and good quality products, design promotional offers, seasonal pricing, offer special discounts and customer specific special offers. They offer a lot more functions, helping you stay competitive in the current market scenario, by coming up with pricing strategies that work best for your business model and your customers. These are just some aspects of the business that big data technologies can help optimize. You can use these technologies to change and streamline other functions like supply chain management, delivery systems etc. Artificial Intelligence Applications Advances in voice and visual recognition, machine learning algorithms, robotics, interactive technologies, faster communication technologies have all resulted in many innovative AI applications that can be used successfully in retail. Robotics Humanoid robots that greet customers personally and provide assistance Virtual Reality and Augmented Reality (AR/VR) Virtual mirrors enable customers in high-end fashion outlets to see themselves in many different garments and different accessory combinations without needing to actually change their clothes. Chatbots Log on to your favorite shop and start chatting with virtual assistants. These can be used in many ways, as sales assistants, as assistants who can help resolve issues etc. These advanced systems can combine machine learning and voice recognition to provide tailored service. Dynamic Catalogs Digital catalogs that ask questions and generates personalized listings that are the best fits for your needs. Virtual Assistants for Online Shopping IBM Watson Websphere technology is being used in many ecommerce sites. For instance, 1800-Flowers.com has used it to implement a system which helps customers make the best gift choices. The virtual assistant does this by getting some information from the customer on what he wants, then matching it to gifts bought by other customers with similar preferences. This helps the system to come up with the best recommendations. No Checkout Shopping Amazon Go, an experimental technology, enables customers to shop in physical stores without ever having to stand in checkout queues. Using customer ID, cameras, motion sensors and visual processing, these shops track customers, identify the products they put into their baskets, calculate the prices and charges their card for their total purchase when they step out of the store. These and more Big Data and AI applications are now remodeling the retail business, changing the way they serve customers and manage their business. Originally published at galido.net on July 17, 2018.

New Paradigms in Retail with AI and Big Data

0

new-paradigms-in-retail-with-ai-and-big-data-17abb7221a95

2018-07-20

2018-07-20 05:40:24

https://medium.com/s/story/new-paradigms-in-retail-with-ai-and-big-data-17abb7221a95

false

743

null

Retail

retail

Retail

16,358

IntelligenceNode Tech Blog

Backstage access to the tech that makes retailers smarter.

d04e34da83c9

life_frida

50

176

20,181,104

null

0

null

0

null

2017-12-22

2017-12-22 04:50:48

2017-12-22

2017-12-22 17:52:54

1

false

en

2017-12-22

2017-12-22 17:56:47

1

17abc88cd52f

4.811321

0

On elevators, women, and changing the world.

5

This picture has deep metaphorical significance to the story. It’s not just a random image I used because I noticed all other Medium posts start with images. Source. The sun shot through the window of my 27th story studio apartment to wake me. Usually my blackout curtains (5 star average on 244 reviews!) kept me in a deep sleep at this time of day, but today was different. Today was a big day. Today was worth breaking the habit of my career and getting up early. I hopped out of bed and stepped over the detritus strewn across the floor, picking from the debris clothes that met a minimum standard of cleanliness. I debated going to put on something special — I still had the suit I wore to my uncle’s funeral. Would that be appropriate? People would die today, so it might be — but I also didn’t want to bother with it. Plus, if I was the only one in a suit, I’d look ridiculous. The elevator was quick to pick me up — a rare treat. The elevator stopped on the 25th floor to admit a young Chinese woman. She had a slender build and long black hair. She was wearing what might be yoga or athletic gear. I couldn’t tell for sure, but it looked good on her. My heart started beating faster and I felt my throat start to tense up. I knew how my retarded classmates must have felt in college when they were called upon to answer a question. I should try this once, while it was still real. Ms. Zimmerman, my therapist, advised speaking honestly to women. She said I should clearly communicate my interest when it was appropriate and suggested that I start by making eye contact. Dr. Ryan, my life-coach, also agreed on the point of eye contact, but he suggested I should start a conversation not so much with honesty, but what he called a “socially appropriate” greeting. Neil Strauss, author of “The Game” — a book on dating, advised opening a conversation with a woman you wanted to date by giving her “a neg”, a kind of backhanded compliment. The theory with the neg being that attractive women would be desensitized to normal conversation starters and compliments, but that they would be shocked and interested in a man daring enough to start a conversation in such an unorthodox way. I squared my shoulders and took a wider stance with my feet as I turned in the elevator to enable natural eye contact with her. Research shows taking up more space in an environment is a masculine and attractive trait. “Nice morning” I said, while trying to think of a neg. Shit, that wasn’t right. Save it! “Good morning; bad hair.” I said with a bit more confidence. She smiled and gave a tiny nod, then took a step forward which broke our eye contact. Was that a neg though, or just an insult? Maybe that was why it didn’t work out. The display on the elevator showed we were approaching floor 10. Not much time. Ms. Zimmerman, time to earn your pay. I stepped forward too, pressing against the closed door, in order to re-establish eye contact. “Look, I’m awkward with women. I know. This is hard for me, but I think you’re pretty. I’m intelligent, creative, well read, and I have lots of money. Let me take you out tonight, to the finest restaurant in town.” “Wǒ bù huì shuō yīngyǔ” she answered with the unmistakable accent of a non-native Mandarin speaker. Ding The elevator had arrived at the lobby. I stood in the opening doorway and tried to pierce her with my stare. “You’ll regret that.” I told her. “Excuse me!?” She answered with the unmistakable accent of a native English speaker. “What did you say to me?” “So you do speak English?” I tried to adopt a wry tone. Had I accidentally stumbled on to the perfect neg for this situation? “Get out of my way” she said while rudely elbowing past me. “Excuse me, excuse me!” I could hear her calling to the concierge. Time to get out of here. “That guy just threatened-” I made it out the front door while she was still shrieking at the concierge and hurried down the street. Funny, even when I failed, I couldn’t help but prove myself right. I knew that advice about picking up girls was dumb. I knew it intuitively, which explained why I had never attempted to follow it before. When I told people at work that I was too nervous to ask a girl out, they always asked “What’s the worst that could happen?” I threw back my head and laughed as I walked, drawing glances from the people bustling by. Yeah, I was probably going to get kicked out of my apartment building or something, but I was glad I had tried picking her up. I had followed the advice, I had made eye contact, I had tried a normal greeting, I had tried a neg, and most of all I had been completely honest. That bitch was going to regret her answer until the day she died. At work, I went through the main gate with everyone else, scanning my badge under the watchful eyes of the security guards. I split off from most of the crowds to go to the elevator banks that served the lower floors. I had to scan my badge again to access my floor — lucky number 13. The elevator opened into an antechamber with yet another security guard. This one smiled, greeted me by name, checked my badge, and had me scan my access token. He waited a second, checked his monitor, and then asked me to step into the full body scanner that stood between me and the antechamber doors. “This is new” I observed to him. “Guess they don’t trust you guys that much, huh?” He said with a laugh. “Guess not.” I forced a chuckle in reply. Perfect small talk. Why couldn’t I pull this off with women? I stepped into the full body scanner, raising my hands to match the icon painted on the scanner, and waited. It was the kind they had at the airport, and the guard looked diligently at his screen for a few moments, no doubt admiring my naked form — I can do a thousand stomach crunches now. “You’re all good” said the guard and there was a beep as the antechamber doors unlocked for me. The scanner was a nice touch, but it was too late. Typical of the business types, they liked to flatter themselves with the conceit that they were not just our superiors in the corporate hierarchy, but that they deserved their lofty org chart perch due to superior intellect. Of course, I had anticipated something like this scanner, and preempted it months before by designing, constructing, and bringing my own lamp to the office. I didn’t need to bring a weapon to the office today, I had one waiting on my desk. I walked through the office space of the 13th floor, unusually quiet this time of day. At the far end of the building was a brown door that might have been a maintenance closet or an electrical access point. The door was marked only with a scanner on the side. I tapped my access token to it, and it beeped as the door unlocked.

Artifice — ONE

0

the-sun-shot-through-the-window-of-my-27th-story-studio-apartment-to-wake-me-17abc88cd52f

2017-12-22

2017-12-22 17:56:48

https://medium.com/s/story/the-sun-shot-through-the-window-of-my-27th-story-studio-apartment-to-wake-me-17abc88cd52f

false

1,222

null

Fiction

fiction

Fiction

84,626

YTL

I like technology, philosophy, engineering, language, and politics. I also like to argue and debate.

623470aceb35

maximumprime

0

1

20,181,104

null

0

null

0

11cce5f166d5

2018-06-08

2018-06-08 15:14:01

2018-06-08

2018-06-08 15:15:08

4

false

en

2018-06-08

2018-06-08 15:15:08

4

17ac091d18e9

5.115094

12

0

Project management is always a difficult and time consuming task which requires solid skills, long-term experience, and deep expertise in a…

5

MANAGING DATA SCIENCE PROJECT Project management is always a difficult and time consuming task which requires solid skills, long-term experience, and deep expertise in a certain domain. What does each project manager have to remember when pushing forward a project related to Big Data industry? Noting value and the importance of Data Science for the advanced software development, many IT companies are investing now more than ever in Big Data and related technologies. According to a SNS Research, the value of global investments in data science technologies will surpass the $57 billion mark by the end of 2017. Both business intelligence (BI) and analytics software market continuously grow and, by the end of this year, will generate global revenue value of $18.3 billion, as reported by Gartner. However, it is important to wisely manage every stage of project development process and safely ensure success by applying the right techniques and leading a team properly. In this article, we will cover main concerns and their solutions in Big Data management. What is CRISP-DM and Why You Need It The most important task of Data Science management is to ensure the highest possible data quality. Various IT companies often try to reinvent the wheel and come up with their own approaches for data mining, but so far, there is still only one appropriate method for doing this which was introduced in Brussels in far 1999. And it’s called the Cross Industry Standard Process for Data Mining, commonly known as CRISP-DM. The CRISP-DM process model is as follows: Business Understanding; Data Understanding; Data Preparation; Modeling; Evaluation; Deployment. Each phase corresponds to the specific activities that usually exist in any project related to DS. Let’s consider the basic benefits you can get by following CRISP-DM principles. Advantages of CRISP-DM The main advantage of CRISP-DM is in its being a cross-industry standard. It means this methodology can be implemented in any DS project notwithstanding its domain or destination. Below, you will find the list of basic advantages of the CRISP-DM approach for Big Data projects. Flexibility No team can avoid pitfalls and mistakes at the beginning of the project. When starting a project, DS teams often suffer from the lack of domain knowledge or ineffective models of data evaluation they have. Thus, a project can become successful only if a team manages to reconfigure its strategy and is able to improve technical processes it applies. Another advantage of CRISP-DM approach is its flexibility. This makes it possible for models and processes to be imperfect at the very beginning. It provides a high level of flexibility that helps improve hypotheses and data analysis methods in a regular manner during further iterations. Long-term Strategy CRISP-DM methodology allows to create a long-term strategy based on short iterations at the beginning of project development. During first iterations, a team can create a basic and simple model cycle that can easily be improved in further iterations. This principle allows to ameliorate a preliminarily developed strategy after obtaining additional information and insights. Functional Templates The amazing benefit of using a CRISP-DM approach is a possibility to develop functional templates for DS management processes. The best way to take as many benefits as possible from CRISP-DM implementation is to create strict checklists for all phases of the work. Microsoft has already built that kind of checklist for DS teams. Team Management In Data Science Software Development Project As the DS market grows, IT companies hire more specialists to develop new projects. According to Evans Data Corporation, 6M developers are working on Big Data projects while you are reading this article. In fact, this number is one-third of all developers worldwide. That is why we need to consider the methods of DS team management. Make Necessary Data Available to Each Specialist DS specialists of every team have to be able to communicate effectively. Therefore, every team member has to have an access to data. It ensures the efficient data collection and obtaining analysis of high quality. Make Sure Everyone Understands the Core Value of Your Company It is crucial for team members to understand where they are going and what they are supposed to achieve. To run the race, you must know where the finish line is. Make sure that all the team members realize what is really important according to the core values company has. Let Your Team Focus on One Task Until the work starts, all roles and responsibilities have to be delegated accurately. Do not let your team members switch between several tasks. Instead, let them focus on one specific task till it is completed. It will help you create a core of in-house professionals specialized in an exact task completion. Hire Responsibly The presence of general DS experience is not enough to take someone aboard. The person, who is considered to be a potential team member, must have an expertise and convenient experience in the domain your project relates to. Use the Right Tools Data processing technologies are continuously improving and evolving. Therefore, it is important to implement centralized platforms that would be able to integrate with currently available tools and improve collaboration between hired talents. Let Your Team Members Learn New Skills When a specialist faces an issue he or she is not familiar with, do not try to delegate the task of finding a solution to another team member if the first one is ready to deal with the situation on his/her own. Let your employees improve their skills and learn new things. Ensure a Timely Big Data Project Delivery To apply Data science successfully to business, it is necessary to build an effective strategy and meet all the deadlines in order to timely perform established tasks. This is where Agile methodology comes in handy. Agile for Data Science Projects In this short paragraph, we will consider basic recommendations regarding using Agile for DS project management. Solving Problems First, Building Features Second The early sprints have to be aimed at understanding what can be controversial in what your team is about to do. Determining primary problems, which require immediate solutions, is a more effective approach in DS than creating additional features to get a wow-effect. Use a Proper Sprint Structure A sprint usually lasts for two weeks, but, sometimes, it can take less time to complete some tasks. Solid sprints do not allow to develop flexible strategies. Set the sprint length depending on the particular situation. Develop a Culture of Fast Experimentation Developing effective data analytics methods is all about getting insights, creating hypotheses and their testing. Yes, you are encouraged to experiment if it can improve your project. The only thing you have to remember that these experiments have to be performed quickly. The effectiveness of any DS project lies in proper setting expectations and estimation of achieved results with regular correction of the previously set direction. Continuous improvement will ensure the success of any Big Data project.

MANAGING DATA SCIENCE PROJECT

326

managing-data-science-project-17ac091d18e9

2018-06-21

2018-06-21 09:04:16

https://medium.com/s/story/managing-data-science-project-17ac091d18e9

false

1,170

Innovative technology boutique. Blockchain. Artificial Intelligence

null

unicsoft

null

Unicsoft

zaa@unicsoft.net

unicsoft

BLOCKCHAIN,ARTIFICIAL INTELLIGENCE,SOFTWARE DEVELOPMENT,DATA SCIENCE,MACHINE LEARNING

unicsoft

Data Science

data-science

Data Science

33,617

Aleksey Zavgorodniy

CEO at Unicsoft. Innovative technology boutique. Blockchain. Artificial Intelligence

388abf6fc5f2

zaa_32329

31

1

20,181,104

null

0

null

0

58e780f214a2

2018-01-13

2018-01-13 00:33:04

2018-01-13

2018-01-13 00:36:50

0

false

en

2018-01-13

2018-01-13 00:36:50

1

17ac542ceff3

1.120755

2

0

Our marketing efforts to date were focused on our community: machine learning and data science practitioners. That’s why most of the…

4

Like these articles? Lead the team that writes them! Our marketing efforts to date were focused on our community: machine learning and data science practitioners. That’s why most of the articles on this blog are relatively technical and tactical with respect to building an AI-first company. We’re expanding our audience in 2018 and hiring someone to lead that expansion. That is, we’re hiring a Marketing and Community Manager to lead the execution of our marketing plan and grow the Zetta community of founders, investors and customers. In this role, you will edit articles on AI, syndicate those articles for publication, win speaking engagements at conferences and organize exclusive events (for Fortune 500 executives, entrepreneurs and investors). You will measure the effectiveness of your work using data about our meetings with entrepreneurs and from industry databases. This is a rare opportunity to be an early member of a venture firm and learn about marketing in the context of investing. Your background includes one or more of: marketing, public relations, journalism or general operations. Your attributes Strong research skills Strong writing skills Effective communicator and critical thinker Demonstrated interest in startups and venture capital Independent self-starter, comfortable thriving in an unstructured environment and taking on a large scope of responsibility. Comfort using Excel for reporting and analytics It’s great if you also have Published articles Previous marketing experience Experience with marketing automation tools Experience with quantitative research methods and company information databases Experience working in a startup environment Application Please send a résumé, a writing sample and a couple of paragraphs expressing your interest in the role work@zettavp.com. All qualified applicants will receive consideration for employment without regard to race, color, religion, gender, gender identity or expression, sexual orientation, national origin, genetics, disability, age, or veteran status.

Like these articles? Lead the team that writes them!

8

like-these-articles-lead-the-team-that-writes-them-17ac542ceff3

2018-06-12

2018-06-12 20:52:36

https://medium.com/s/story/like-these-articles-lead-the-team-that-writes-them-17ac542ceff3

false

297

Zetta invests in intelligent enterprise software. We partner with companies building software that learns from data to analyze, predict and prescribe outcomes.

null

Zetta Venture Partners

info@zettavp.com

zetta-venture-partners

TECHNOLOGY,AI,STARTUP,MACHINE LEARNING,ENTERPRISE SOFTWARE

zettaventures

Machine Learning

machine-learning

Machine Learning

51,320

Zetta Venture Partners

The Intelligent Enterprise Fund

b094da03da9d

Zetta

668

12

20,181,104

null

0

$ yum -y update $ yum -y groupinstall "GNOME Desktop" "Development Tools" $ yum -y install kernel-devel $ mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r) nouveau.img $ dracut /boot/initramfs-$(uname -r).img $(uname -r) $ chmod +x [NVIDIA_driver_file].run $ ./[NVIDIA_driver_file].run $ export PATH=/usr/local/cuda-9.1/bin${PATH:+:${PATH}} $ export LD_LIBRARY_PATH=/usr/local/cuda-9.1/lib64\ ${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} $ cat /proc/driver/nvidia/version #..should output something like this: NVRM version: NVIDIA UNIX x86_64 Kernel Module 387.26 Thu Nov 2 21:20:16 PDT 2017 GCC version: gcc version 4.8.5 20150623 (Red Hat 4.8.5-16) (GCC) $ nvcc -V #..should output something like this: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2017 NVIDIA Corporation Built on Fri_Nov__3_21:07:56_CDT_2017 Cuda compilation tools, release 9.1, V9.1.85 $ cuda-install-samples-9.1.sh ~ $ cd ~/NVIDIA_CUDA-9.1_Samples/5_Simulations/nbody $ make $ cd bin/x86_64/linux/release #..from root testing directory $ ./deviceQuery #..should see something like this: ./deviceQuery Starting... CUDA Device Query (Runtime API) version (CUDART static linking) Detected 1 CUDA Capable device(s) Device 0: "GeForce GTX 1080 Ti" CUDA Driver Version / Runtime Version 9.1 / 9.1 CUDA Capability Major/Minor version number: 6.1 Total amount of global memory: 11169 MBytes (11711807488 bytes) (28) Multiprocessors, (128) CUDA Cores/MP: 3584 CUDA Cores GPU Max Clock rate: 1683 MHz (1.68 GHz) ... $ ./bandwidthTest #..should see something like this: [CUDA Bandwidth Test] - Starting... Running on... Device 0: GeForce GTX 1080 Ti Quick Mode Host to Device Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 12709.4 Device to Host Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 12893.4 Device to Device Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 371522.7 Result = PASS NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

18

null

2018-02-25

2018-02-25 01:35:39

2018-03-05

2018-03-05 00:02:37

1

false

en

2018-03-05

2018-03-05 04:16:10

44

17aed82ab3ce

14.169811

6

4

0

I’m a data science practitioner who has recently found it challenging to cultivate my practice. As Credijusto tech grows along with the…

5

Building a CentOS 7 Machine Learning Station with NVIDIA/CUDA integration: Everything I Learned I’m a data science practitioner who has recently found it challenging to cultivate my practice. As Credijusto tech grows along with the rest of the company, I find that I spent less time each day on data analytics and code, and more on the many non-technical tasks that are crucial for building a 10x team. I love turning raw data into insights and am fascinated with how emerging technologies today help us understand complex systems through data. To stay engaged with the field I decided I needed a home research station, and set off to create the most awesome one that I could afford. I’m delighted with what I ended up with, but it took 5 months of frustration, 2 busted motherboards, and about $750 of extra expenses, before I got the project off the ground. The goal here is to share lessons learned, both moral and technical, in the hopes that in doing so your process will be about 10x smoother than mine was. To get the basics read the tl;dr just below. If you read beyond that, here’s what I’ll talk about: Local hardware versus cloud — Why would you build your own machine when you can rent cloud capacity? I’ll list the components used to build ghost, and do a bit comparing the economics of cloud versus local. There are good reasons to go either way. A few high-level takeaways— Pedagogical notes for people about to embark on a similar project. If you’re ADD, like me, and lean a bit too much toward jumping into things before researching them, then this section is worth reading. Technical lessons — What most data scientists and technicians will actually want to read. If you’ve gotten your machine to post and now just want to know how to setup your environment then skip to here. tl;dr I built a workstation with an NVIDIA GeForce GTX 1080 Ti, CUDA-integrated, with an Intel i7–7700 CPU and 64GB’s of RAM. All-in cost was $2662. When you compare this to dedicated GPU instance plans on Paperspace the economics of cloud versus local are kind of a toss-up. What sold me on building a local station was convenience, but industry economics will increasingly move all but the most hard-core practitioners toward the cloud. A few high-level takeaways: Front-load planning and research before jumping in; Make a plan and stick to it — don’t react based on negative emotional energy; Handle your components with utter reverence; Don’t alternate graphics cables between your integrated and GPU ports; Experiments with software are cheap and fast. Experiments with hardware are expensive and slow. Exhaust potential software fixes before you mess around with the hardware; When stuck, invest in good help; Don’t let the perfect be the enemy of the great. Some technical lessons: Use CentOS with GNOME — unlike Ubuntu with Unity, everything worked fine I got the machine booting up; Configuring the NVIDIA drivers and installing CUDA is a detailed but straightforward process that’s pretty easy to navigate. See the Technical notes section to learn more. Local hardware versus the cloud Here’s the components I used for the base system (I only included the final motherboard). I bought all of this in Mexico, but converted the prices to USD at the time of purchase: NZXT Phantom 410 Mid Tower Case — $86 Cooler Master Hyper 212 EVO CPU Fan — $32 Corsair CMK32GX4M2A2666C16 RAM Memory Kit, 2x 16GB —$293 Kingston HX421C14FB/16 RAM Memory Kit, 2x 16GB — $287 Kingston SUV400S37/240G SSD — $88 Intel i7- 7700K Quad Core processor — $296 Gigabyte GV-N108TAORUS-11GD GeForce GTX 1080 Ti — $826 Corsair 850W Power Source — $97 NZXT FZ -140mm LED Case Fan —$20 Extra fan cords — $17 Gigabyte AORUS GA-Z270X-Gaming K7 Gaming Motherboard — $214 That’s the base hardware. Of course you also need to setup your desktop environment. Those components were: 3 foot DPI / HDMI cable — $9 15 meter Ethernet cable — $15 Multi-plug with 6 foot connector — $15 BenQ GL2760H 27" Monitor LED Full HD — $192 2 meter HDMI cable — $8 TP-Link Archer T9E AC1900 Wireless Adapter — $56 ASUS VS228H-P 21.4" Monitor — $86 Cooler Master Devastator II Keyboard/Mouse Combo — $25 Grand total for the base components: $2256. Grand total for the desktop environment: $406. All-in: $2662. Now, I did have two motherboard casualties that added around $500 of extra costs. I will likely be able to get $250 back on warranty. More below. A fair comparison of local versus cloud economics would account for the fact that, even when you go with the cloud, you still need a client machine to work on. I’m going to adjust those client machine costs out of my all-in costs, assume that you prefer working either in Unix or Linux, and that you’re willing to invest a little bit in user experience and hardware quality. Using those assumptions, a basic MacBook Air, which runs $999, seems like a reasonable choice. Subtracting the client machine costs I’m left with about a $1663 up-front investment in computing power. How much work do you have to do locally at that price point until you start getting a positive ROI? More than you might think. The costs of GPU cloud computing have really been coming down, making a locally-integrated cloud research environment a pretty good choice for many applications. The best cloud comp that I could find was the P5000 dedicated GPU instance on Paperspace. The P5000 has a 16GB memory speed compared to 11GB for the 1080. Both machines compute at a rate of 9 teraflops, meaning that for parallel computation they are going to have similar performance. Like the dedicated instance, my machine has 8 CPU cores (4 of which are virtual), but wins handily on RAM with 64GB versus 30GB. This instance costs either $0.65/hour or $290/month. If you go with hourly, you need to power down the machine whenever you’re not using it to avoid racking up heavy usage fees. If you use the hourly plan for only 10% of the hours in a month, accounting for the $5 storage fee, your monthly cost is going to be $44. We’ll call the $290/month price point the practitioner’s price point. This is someone who is training heavy models requiring a couple hours each, optimizing hyper-parameters across multiple training runs, and maybe mining some crypto. This person gets a positive ROI with my specs in less than 6 months. $44 a month is the hobbyist’s price point. This is someone who ducks into their environment here and there to train a couple models, maybe do the fast.ai course, and play around with the occasional Kaggle competition. This person gets a positive ROI in about 3 years, and should probably go cloud. I’m somewhere between the practitioner and hobbyist. On a strictly dollars and cents basis it’s a tossup whether this project is economically justified. For now, the convenience of being able to work locally was the final selling point. I don’t have to worry about turning the instance on and off. It’s easy to set up a multi-monitor working environment. I don’t need to port data back and forth between the client and the cloud server. If I take my WiFi adapter card out, which I’m currently not using anyway, I can add another powerful GPU. As the technology advances and the costs of high-end GPU computing continues to come down, it is likely that in a year or so I’ll be making similar arguments to fellow practitioners about the advantages of cloud that I currently make for companies. In the meantime local still felt like the right choice in terms of both the costs of computation and user experience. High-level takeaways Front-loading planning and research — I bought the wrong sized case at first and had to return it; Forgot to purchase the CPU fan; Needed extra case fans and connector cables; Didn’t think about the network card. Life keeps reminding me that putting in the time to make a solid plan, or at least to inform yourself, is typically worth it. Don’t get analysis paralysis, but with a highly technical project like this invest in informing yourself before jumping in. Treat your components with utter reverence — At some point in my first build attempt I bent the pins in the motherboard’s CPU slot. This is the ultimate avoidable problem and is easily solved by just being careful. Make a plan and stick to it/don’t be reactive— The first time I got ghost posting I installed Ubuntu and right away started having problems installing the NVIDIA drivers and the getting dual monitor display to work. Instead of attacking the problem in a structured way I started making a bunch of apt-get calls, switching back and forth on what problem I was trying to solve, and ultimately getting frustrated and yanking out the GPU to try and run both displays off of the CPU’s integrated graphics. This cost me the second of my two motherboards and leads me to the next point… When troubleshooting, rule out the low-cost, low-risk factors first — Experiments with software are cheap and fast. Experiments with hardware are expensive and slow. Exhaust potential software fixes before you mess around with the hardware. After a full day of trouble installing NVIDIA drivers and getting dual monitors running in Ubuntu I pulled out the GPU and connected my cables to the CPU’s integrated graphics ports. I got into the BIOS this way, but couldn’t log in to the operating system. I then replaced the GPU and hooked the graphics cables back to its ports. Apparently alternating cable connections between your integrated and GPU ports is a really bad idea: the machine didn’t even post after this. It would have been far cheaper to have tried out a different Linux distro, e.g. CentOS 7, before I start experimenting with hardware. This was what I went with, ultimately, and had I tried this first I would have saved a motherboard. When stuck, invest in good help — My friend José Carlos Nieto, Co-Founder of Mazing Studio, got me back on track after I killed my second motherboard. He’s also the one who turned me on to CentOS 7 instead of Ubuntu, which ended up being a great choice. His time wasn’t free, but it was totally worth it. Don’t let the perfect be the enemy of the great — I still haven’t gotten my WiFi adapter running. I froze the OS on José Carlos’ build when I dropped a bunch of .sofiles into the /lib/firmare directory. I had to rebuild the OS and configure all dependencies and drivers from scratch once more after this. After 5 months of hangups and procrastination I decided that I wasn’t going to make this a sticking point. I have a 15 meter Ethernet cable running through my living room and am leaving the WiFi for another day. Lemme know if you have any tips. Technical notes: getting it all running There are a lot of great guides out there for building a desktop PC. If you’re looking for a great machine learning rig then buy everything I’ve listed here and put it all together. You should consider buying a more powerful GPU, perhaps an Intel i9 CPU, and a faster SSD if you need to do a lot of I/O from the hard drive. In general, just go through the components list above and make sure you have all the key ingredients. Put it all together, get the BIOS posting, and the you’re ready to setup the environment: Installing the OS Once you’re in the motherboard’s BIOS you’ll need a boot image. I found a great guide for how to create this with macOS here. The steps were: Download the DVD ISO from the CentOS website. From the download directory run hdiutil convert -format UDRW -o centosdvd.img CentOS-7.0–1406-x86_64-DVD.iso to convert the iso file to a img file. Plug in an 8 GB USB and usediskutil list to find that disk. Unmount it with diskutil unmountDisk /dev/[disk_name]. The disk name will typically be something like disk1 or disk2. Copy the boot file img to the disk with sudo dd if=centosdvd.img.dmg of=/dev/[disk_name]. This will take a long time, in my case a couple hours. From here plug the USB into your rig while it’s turned completely off and turn on the power. From the BIOS menu you should have easy access via the UI to the boot settings, where you will select the USB drive you just inserted. For both the MSI motherboards that I destroyed and the Aorus motherboard that I finally used this was very straightforward. Select the boot disk as your first boot option, restart the computer, and go through the CentOS 7 install steps. I did the full install, but you could probably get away with the basic install, dropping in your key dependencies later. Configuring the NVIDIA drivers A quick note before I walk through how I did this with CentOS: after I crashed my first successful CentOS 7 install by adding some bad .so files to the /lib/firmware directory I tried to do this with Ubuntu. Surely this was going to be easier, right? After all, Ubuntu is know as the most user-friendly distro of Linux out there. You don’t have to spend much time googling for answers to realize that it’s very well-supported by the open source community. Without going into too much detail, I eventually wound up at this stack post as I tried to troubleshoot some issue related to X server/lightdm. I was able to finish the driver install but unable to to restart lightdm, unable to get back into the Unity GUI, and unable to install CUDA. After a full day of frustration which ended in me totally killing Unity and only being able to interact with the OS via the terminal in run level 3, I decided to give CentOS 7 another try. As a plus, I also liked the GNOME GUI a lot better, which came out-of-the-box with the CentOS 7 install. If you’re a Mac user, its multi-desktop environment will feel familiar. There was also zero fuss getting multiple monitors running. Once back in CentOS 7 territory these were the steps I followed to get the drivers running. I aggregated information and instructions from a few different sites. If I’ve missed any important steps here or you took a different route that worked well please let me know so that I can update. I started with this this guide. Several of the steps here are copied straight from that page: Run the following commands: 2. Download the appropriate NVIDIA driver. This will not necessarily be the “Latest Long Lived Branch version” that the install guide recommends. I recommend going to the NVIDIA Driver Downloads page and using the selectors to locate the appropriate driver file. For the 1080 Ti on a 64 bit Linux system that was this one. (If you have any doubts about whether you are running 32 or 64 bit you can confirm with uname -a in terminal) 3. Reboot your computer and then append rd.driver.blacklist=nouveau nouveau.modeset=0 to the GRUB_CMDLINE_LINUX section of /etc/default/grub. 4. Generate a new grub configuration to include the above changes: grub2-mkconfig -o /boot/grub2/grub.cfg. 5. Edit (or create if it doesn’t exist) the file /etc/modprobe.d/blacklist.conf and append blacklist nouveau. 6. Backup your old initramfs and then build a new one: 7. Reboot your machine. At this point I stopped using the first guide. The systemctl call it mentions wasn’t necessary. After the reboot your interface will likely look grainy and clunky. Don’t worry, you didn’t fry GNOME, you’re just not done with the install. Hit Cntl-Alt-F3 to enter your terminal prompt and enter root user with sudo su. 8. cd to the directory where you’ve installed your NVIDIA driver and execute the following: 9. Accept X override when asked. 10. User init 5 to return to desktop mode. If GNOME isn’t back to normal you may need to reboot the system once more. To do this enter terminal with init 3 and then sudo reboot. From there I was good to go with the NVIDIA drivers and GNOME was back to looking and functioning normally. The final river to cross on the way to enjoying machine learning paradise was the CUDA install. Again, the steps here were detailed but pretty easy to follow. I used the CUDA Toolkit Documentation from NVIDIA as my primary reference here. You can go through the preliminary steps to check the availability of a CUDA-enabled GPU, appropriate Linux distro, and your gcc installation if you like, but chances are that if you got this far you’ve already got these sorted. You should already have your kernel headers and development packages installed from the driver install, but run sudo yum install kernel-devel-$(uname -r) kernel-headers-$(uname -r) just in case. Download the appropriate NVIDIA CUDA toolkit runfile from here. If you have the same specs as me this should be Linux -> x86_64 -> CentOS -> 7 -> runfile (local). Not all the steps listed in the guide aren’t necessary if you’ve just gone through the driver install process. Enter the shell with sudo init 3 and cd to the directory containing your runfile download. Run the installer on silent mode to automatically accept the EULA and accept default parameters with sudo sh cuda_<version>_linux.run — silent. If you want finer control over the install remove the --silent flag. I did this, but I can’t recall if I changed any installation defaults. I’m not sure if this step was necessary, but you can also create an xorg.conf file from the NVIDIA GPU display with sudo nvidia-xconfig. After the install completes successfully sudo reboot to reboot the system and enter the GNOME GUI. Update your path variables with the CUDA binaries (change your version if not using 9.1): 8. A couple quick checks to verify both your NVIDIA driver version and CUDA install. I used this post-install guide as a reference for the following steps: 9. Install a writable copy of the CUDA samples and build them (again, pay attention to the 9.1 version number — depending on when you’re reading you may need to update it): 10. The runfile installer guide tells you to run the nbody example with ./nbody but at least when I went through these steps I did not have that in the root samples directory. Instead I followed the post-install guide I referenced above to run deviceQuery and bandwidthTest. Note that the paths mentioned in the guide have changed, or at least they were different for my setup. I entered the appropriate directory and executed the tests as follows: 10. That’s it. For good measure do one more sudo reboot, get back in to your terminal, and run nvcc --version and echo $PATH . The former should repeat the successful result, above, and the latter should include /usr/local/cuda/bin in the output. If that’s the case, congrats, you’re ready for training! If it’s not please let me know in the comments and let’s see if I can help you out with troubleshooting. Wrap-up This was a challenging project, but it feels great having the rig up and running. I only recommend this if your level of commitment is high and you have significant prior experience with machine learning and know that it’s something you really want to invest in. You could also mine some crypto that isn’t bitcoin or ethereum, but the economics of this won’t really add up for most folks. Otherwise start with an on-demand Paperspace GPU and get your feet wet with a few tutorials. I’m getting back into the flow with the fast.ai course, Deep Learning for Coders, and the Kaggle 2018 Data Science Bowl. Thinks I’d still like to solve on the hardware end include how to set up a reliable ssh/VPN tunnel (my router is an SAP, not a primary), and how to setup my WiFi adapter without crashing my build. Any tips would be welcome. Have fun out there. At the end of all this here’s how the workstation came out:

Building a CentOS 7 Machine Learning Station with NVIDIA/CUDA integration: Everything I Learned

58

building-a-centos-7-machine-learning-station-with-nvidia-cuda-integration-everything-i-learned-17aed82ab3ce

2018-05-20

2018-05-20 09:33:55

https://medium.com/s/story/building-a-centos-7-machine-learning-station-with-nvidia-cuda-integration-everything-i-learned-17aed82ab3ce

false

3,702

null

Tech

tech

Tech

142,368

Aaron Polhamus

Güero living in Mexico City, working at Credijusto.com, connecting with fellow humans, and taking in the scenery. Sometimes I take pictures: IG @aaronpolhamus

e42a6733c6f0

aaronpolhamus

80

64

20,181,104

null

0

null

0

435d0cd839de

2018-08-10

2018-08-10 02:06:20

2018-08-10

2018-08-10 02:13:20

3

false

en

2018-08-10

2018-08-10 02:13:20

3

17af55dce31b

8.187736

2

0

Note: This article is the second part of a series, in which Savvycom Team will discuss future of AI in healthcare — battling the world…

4

Future of AI in Healthcare (Part 2): Preventing Diabetes Note: This article is the second part of a series, in which Savvycom Team will discuss future of AI in healthcare — battling the world deadliest diseases. According to Wikipedia, the first clinical description of this illness was noted down by Aretaeus during the 1st century CE. At that time, Diabetes was only described as: A disease that caused an excessive amount of sweet urine which passed through the kidneys. Not much was known about this illness. In fact, diabetes was quite rare. No one would have image that one day, diabetes will become the biggest epidemic in human history — affecting 415 million people worldwide by 2018. Standing at number 4 In WHO’s TOP 10 deadliest disease worldwide, Diabetes is considered a progressive disease. As stated by the US National Library of Medicine, premature death caused by diabetes results in about 12 to 14 years of life lost. The patients and their family also incur medical costs that are 2 to 5 times higher than those without Diabetes. The annual direct health care costs of diabetes worldwide, for people in the 20–79 age groups, are estimated to be as much as 286 billion. The world seems daunting after all of those statistics. But, with the rise of Artificial Intelligence, advanced medical protocol seems to be on the horizon. Is there a future of AI in healthcare for Diabetes patients? 1. Diabetes: A world epidemic According to WHO, Diabetes is defined as ‘a chronic disease that occurs either when the pancreas does not produce enough insulin or when the body cannot effectively use the insulin it produces’ So, what is insulin? To answer that, let’s first start with another question: What is glucose? Glucose is a medical term for the sugar created by the digestive system as it consuming and breaking down foods. This sugar is the fuel for the cells — much like gasoline to cars. Without gasoline, cars cannot run. Without glucose, cells cannot continue living. However, an average human body contains more than 37.2 trillion cells, spread across a vast area. Therefore, turning food into glucose is not the final step. That glucose still has to get to every single cell in one’s body. And that’s where insulin came in. Insulin acts as the pump that transfers gasoline to the car’s fuel tank. Since cells have ‘locks’ that are called insulin receptors, insulin fits into these locks like a key. When insulin opens the locks, glucose is allowed to enter the house of cells. To be more specific, the pancreas is genetically coded to produced different amounts of insulin depending on how much glucose is in the bloodstream. When a person goes about their day, not eating much, it releases just a bit to keep things regulated. But during food consumption and processing, it generates a burst of insulin in response. Hence, insulin is vital factors to keep glucose at balance level and let one’s body operate at its optimal level. Having too much or too little glucose for a substantial period of time and one’s body starts running into complications — including the 2 types of diabetes. Remember the ‘lock-key’ metaphor earlier? Type 1 vs. Type 2 Diabetes | Healthstyle With type 1, there is no key. The patient’s body makes little to no insulin because the beta cells in the pancreas that make insulin are mistakenly destroyed by the body’s own immune system as it was fighting infection. This type is usually diagnosed in children and young adults. Type 2 is more common. With this type, there are faults in the key itself. The beta cells in the pancreas produce insulin, but not enough to keep blood sugar levels within a normal range or the body doesn’t respond properly to insulin. Without enough insulin to direct the flows of glucose, the glucose is left in the blood. This is what happens when someone is having “high blood sugar”. According to WebMD, early warning symptoms of Diabetes include small incidents such as increased thirst or hunger, frequent urination, unexplained weight loss, fatigue, blurred vision, and headaches. Leave it for a longer time, the patient will be exposed for 3 times more risk of experiencing heart attacks, strokes, nerve damages, infections that might lead to lower limb amputation, blindness, and kidney failure. As you can see, thanks to science progressions and enormous data of health records, we do have a certain level of knowledge in regard to how the disease progress. However, not much is known about the specific causes of diabetes. Scientists think that while type 1 diabetes is caused by sudden environmental factors (such as virus or infections), type 2 diabetes is created by a more collective group of triggers: obesity, lack of physical activities, genes and family history. 2. Is there a cure? This is going to sound very similar for every disease mentioned in this series: there is currently no cure for diabetes. Scientist and doctors in related fields have been trying for decades. With Type 1, clinical attempts focus on either replacing the damaged pancreas with a healthy one (through islet cell or pancreas transplant) or targeting the immune system in an effort to stave further damage to the pancreas. However, these efforts have experienced several shortcomings. Not only that donors are in very short supply, systematic reviews also find transplant results themselves tend to vary significantly. With regard to treatments targeting the immune system, the results remain blunt and non-specific. With type 2, it has been noticed that the number of patient rises along the global rate of obesity and metabolic syndrome — a cluster of conditions related to blood sugar, excess fat and abnormal cholesterol level. This has fueled an increase in weight loss surgical interventions,. However, depending on the country and insurance plans, such surgery can be costly. They’re also not risk-free with risks varying greatly depending on the person’s overall health profile and age as well as skill and experience of the surgeon. In the meantime, tremendous interest lies in the usage of different types of stem cell to regenerate the pancreas. This has been applied for both type 1 and type 2 diabetes in recent years with mixed results and limitations. For example, later stages of diabetes’ patients are not good candidates for stem cell therapy. 3. Future of AI in Healthcare: Diagnosis and Effective Control There is definitely hope since Diabetes can be controlled through effective medication and a healthier lifestyle. What’s vital here is the patient and doctors’ acknowledgment of the current situation. This means that early diagnosis, non-invasive test, and effective maintenance protocol are the key factors. With that being said, AI’s future in healthcare — particularly in Diabetes can be divided into three main categories: 3.1.Non-invasive early diagnosis: How: According to WHO, although detection is improving, the delay from disease onset to actual diagnosis may exceed to 10 years. Contributed reasons to this issue include the subtleness of early symptoms along with the complicated process of diagnosing — which involves a range of actors following the Finnish Diabetes Risk Score. As this method requires human intervention and expertise, it may be exposed to human errors. Highlighted projects: According to Reuter, one of the most influential complications of diabetes is diabetic retinopathy (DR) — damages in the eye blood vessels and vision loss. IDx-DR, a software produced by an Iowa-based company, utilizes AI software to self-assess the eye images taken by a retinal camera. After a series of comparison to a provided database, the software tells the doctor that the patient either has more than mild DR and should be referred to eye-care professionals or is “negative” and should be rescreened in 12 months. IDx-DR example | Source: Intro Wellness Result: In a clinical trial, IDx-DR was able to correctly identify the presence of more than mild diabetic retinopathy 87% of the time and identify those who did not have more than mild disease 89% of the time. It has now received the FDA’s authorization to provide screening decision without the need and assistant of a clinical. 3.2. Non-invasive Glucose Monitoring Systems: How: Once diagnosed, frequent adjustments of the insulin treatment plan are crucial for successfully achieving glucose controls goals. Not only is insulin optimization calculation is a time-consuming process, it also demands constant updating data from a board range of devices — glucose monitoring devices, insulin dose regimens, diet tracking calendar, exercise diary. Thus, traditional physicist only gets to see their patient once every few months. With an AI platform, machine learning algorithms can help automate the process of monitoring blood sugar levels and recommend adjustments in care. Highlighted projects: Founded in 2014, DreaMed Advisor cloud-based analytics platform uses machine learning to recommend optimal insulin dosages to maintain balanced glucose levels. For example, data from diabetes management systems are transmitted to the cloud. The patterns derived from analysis through its event detections and learning algorithm are referenced to provide automated recommendations for insulin dosing and treatment plan — in real time. Doctors can then access the cumulative data from the cloud and learn the patient’s unique habits and needs. DreaMed Advisor | DreaMed Advisor’s Youtube Result: The U.S. National Library of Medicine indicates that DreaMed began recruiting participants in December 2016 for an evaluation study in children and adolescents with type 1 diabetes. The result will be released in late 2018. 3.3. Nutrition Coaching: How: One of the biggest parts in taking control over this life-long illness is the patient’s diet. As one’s body experiencing internal chemical imbalance, that person needs to watch their intake in sugar, fat, protein and carb index. However, there isn’t one specific “diabetes diet”. Doctors need to work closely with their patient to customize their specific meal plan — which is of course, demand extensive knowledge in the nutrition. With machine learning, AI can help recommend meal options based on the specific diet criteria of the user. Highlighted projects: Founded in 2014, California-based Suggestic is taking a nutrition-focused approach to helping diabetics manage their health. The platform is built on an extensive database of over 1 million recipes and 500,000 restaurant menus. This data is used to train algorithms to recognize which food selections complement specific diets. The platform also uses an Adherence Score scale — ranging from green (optimal) to red (least optimal) to determine how well a meal option fits with a user’s diet. Suggestic Interface | Source: Suggestic Result: This app, even though only being made available for iOS devices, scores 4.8 stars out of over 150 reviews. They are also in partnership with The Institution for Functional medicine and Health Coach Institute. _____ On one hand, we have yet to understand the root causes of Diabetes. But on the other hand, we have come up with various protocol to ensure that the patients can still live a long and healthy life through disease-modifying treatments and lifestyle alterations. The most important thing at this point is to extensively understand the stages of diabetes and what implications they may have on the people’s life. Future of AI in Healthcare is bright. Hence, healthcare institutions need to catch up with this first wave of AI development, not only to remain sustainable and profitable but to ensure that they are doing their parts in the making of needed medical progression. It is not an easy task, and that’s why we are here help. Striving to be a key advocate for the future of AI in healthcare industry worldwide, Savvycom founded a new AI Lab back in March 2018. Leading by Dr. Long Tran, our team have developed three AI applications (Facial Recognition, Object Identification, and AskFred — an AI Chatbot) and are now in the process of commercialization. These technologies are developed with the hope that it will become the foundation of our future healthcare-related products. For example, AskFred can be used as a personal assistant to Diabetes patient — providing answers to the patient questions with the micro/macro nutrition components in each meal. If you have any request for further information regarding our service, please send us a quick email here.

Future of AI in Healthcare (Part 2): Preventing Diabetes

2

future-of-ai-in-healthcare-part-2-preventing-diabetes-17af55dce31b

2018-08-10

2018-08-10 02:13:20

https://medium.com/s/story/future-of-ai-in-healthcare-part-2-preventing-diabetes-17af55dce31b

false

2,024

Top 30 Mobile App Developers.

null

savvycom

null

Savvycom

contact@savvycomsoftware.com

savvycom-software

TECHNOLOGY,SOFTWARE DEVELOPMENT,BLOCKCHAIN TECHNOLOGY,ARTIFICIAL INTELLIGENCE,IOT

savvycom

Healthcare

healthcare

Healthcare

59,511

Savvycom Software

Top 25 IT Outsourcing Companies on Clutch

94c4025836fe

savvycom

3

2

20,181,104

null

0

null

0

d7813859318

2018-06-25

2018-06-25 22:59:11

2018-06-26

2018-06-26 14:10:19

2

false

en

2018-06-27

2018-06-27 07:12:37

3

17b09a40abb

6.345597

7

0

What the new European data protection regulation has in store for the future of AI and data-driven innovation.

5

The GDPR and the future of data-driven innovation What the new European data protection regulation has in store for the future of AI and data-driven innovation. John Bardeen, Walter Brattain, and William Shockley In 1947 the transistor was invented by John Bardeen, Walter Brattain, and William Shockley. Many scientists and partners contributed along the way, but in the end it was three men who really brought the transistor to life. Three men that brought vastly different talents and strengths to the table. John Bardeen was the thinker, a man who could look at an event no one else comprehended and go beyond common understanding to explain it. Walter Brattain was the tinkerer, a builder who could put together any contraption asked. William Shockley was the visionary, a seer who predicted how important the transistor would be long before anyone else. Their unique skills brought together in one laboratory enabled the perfect ground for a truly world-changing invention. While this invention is a perfect example of many different innovation paradigms, it is the fact that the collaborative expertise of all three men combined made the invention of the transistor possible. The future is personalized and collaborative If we fast forward to today, this holds more than ever. Nowadays, the knowledge possessed by an individual is rarely enough to spark groundbreaking inventions. In our growing customer-centric economy, innovation begins with understanding people. Companies partly do this by measuring, tracking, and storing personal data on individuals in order to quantify personal preferences and use this knowledge to tailor experiences and products towards each customer individually. Personal data is the core source of modern services and products and serves as the most important resource for the majority of modern technological advances and discoveries. This does not only hold for scientific settings but also for corporate R&D. So far companies have been heavily relying on data they collected themselves. Now, with our behaviour becoming more complex and digital services more diverse, companies are increasingly leveraging multi-faceted data from various outside sources in order to understand their consumers. This also holds for the supply side of data. According to a study by AIG, 75% of companies globally would share data if this provides them with “benefits” such as providing better customer products and services A recent example of this is an agreement between the city of Washington DC and Uber. Both entities collaborate on pooling ride data from numerous sources to improve the cities infrastructure, establish a curb data standar, and finally enable a better data-driven services. In the future we expect to see more and more companies engaging in sharing and leveraging multiple sources of relevant data for their product innovations. At least in theory. Enter GDPR Since the 25th of May, the GDPR has changed this notion. The General Data Protection Regulation was put into action with the main goal of giving consumers back the control over their data. This has been long overdue, with recent cases in the media, such as Facebook, showing the immense dangers of companies not taking user privacy seriously enough. On the other side, the GDPR simultaneously makes it increasingly difficult for companies to leverage personal customer data for secondary data use cases. Secondary data use cases include all cases where data is leveraged for cases outside the initial purpose the data was collected for in the first place. This often includes testing and developing new products driven by data such as AI- and ML-applications. One of the most important articles in the GDPR is article 6. It governs the legal reasons for compliant data processing. Next to processing data for compliance with a legal obligation, performance of a contract, the vital interest of a data subject (consumer), or a task carried out in the public interest, the reasons for collecting and processing data can be split into two big groups. On the one side, companies are allowed to collect customer data if they obtain explicit and informed consent from their customers for doing so. On the other side, companies may process customer data if it is necessary for the purposes of the legitimate interests pursued by the controller (the company). The big problem is that the concept “legitimate interest” is very broadly defined. Without clear indications by courts, it is currently up for interpretation what constitutes a legal legitimate interest of companies. This means that currently, the safest way for companies to use data for product innovations is to obtain consent from data subjects to do so. Next to obvious problems of free tracking and sharing of data of high-profile media cases such as Facebook, there has been a general resentment towards the quantification of our behavior. Especially the thirst for data of advertisement applications has lead general frustration with omnipresent data tracking. This resulted in the current negative stigma of other data-driven use cases such as for example AI. Thus, with a current lack of education around the importance of data in building personalized products, it will be highly unlikely that the majority of customers will consent to the use of their data for product tests and innovation — especially if this data will be shared with innovation partners by companies. Another big change comes with articles 13 to 17 of the GDPR. They require companies to provide a whole new degree of transparency when it comes to using personal data for example for machine learning models and AI products. In a nutshell, article 13 through 17 give the data subject, in this case the customer, complete transparency on where his or her data is stored, used, and the right to request his or her data to be permanently deleted. Article 15 is worth some extra attention. Article 15 of the GDPR mandates that personalization due to automated data processes requires explanation by the data controller. This does not necessarily mean that companies need to fear opening up their data-driven algorithms to external parties but they need to be able to explain the basic functionalities of their algorithms to a data subject. This might also hold for test purposes and internal product development purposes before such products are rolled out and commercialized. The ability for consumer to opt-out, require more granular data model management, in order to easily replace and delete personal data and easily retrain data-driven models on the updates dataset. What is more difficult though, is the apparent perception of lawmakers of how Machine Learning algorithms function. Algorithmic transparency means that a company can see how the decision is reached. With Machine Learning algorithms not being rule-based software, this becomes increasingly problematic. Ones again, the first court cases need to show, how detailed companies will be required to describe and show the use of data for Machine Learning. With a focus on leveraging data for innovation and product development, the GDPR in its current state results in two major problems: There is no legal ground for leveraging customer data for internal product innovation and testing (before commercialization). It does not lay out how companies ideally should process and work with data thus leaving a lot of room for uncertainty and interpretation. This means that leveraging customer data for data-driven innovation is becoming increasingly difficult. Often companies used to gather data with the idea to use it in future projects or developments without a clear idea of how these will look like. The GDPR brings a lot of uncertainty around collecting the necessary data for this in the first place. Since exploration of data and subsequently uncovering trends and insights in data lies at the heart of Data Science, this all raises two big question: How can we support the use of data for product innovation? How can we enable data-driven collaboration while protecting consumers? Franki Chamaki via Unsplash.com Synthetic data is the way of the future The answer to both is data anonymization — because truly anonymous data is not subject to data privacy regulations. The GDPR clearly states that the use of truly anonymous data is exempt from the regulation. Truly anonymous data is any data processed in a way that the privacy of individuals is preserved, in the sense that they do not incur any harm from the data being released. Therefore, the use of anonymous data renders the aforementioned problems unimportant and furthermore allows for companies to freely share data between and across each other — all while protecting their customers, as anonymous data per definition means that the re-identification of an individual data subject is impossible. And here lies the problem. Truly anonymizing data is difficult. Anonymizing data in a privacy-preserving manner takes time, resources, and significant domain expertise. Additionally, even if the generation of truly anonymous data has been successful, this often equals a significant loss in data utility, thus rendering an anonymous dataset useless. This is why we built Statice. Statice makes anonymizing data easy while maintaining data utility and data granularity. By leveraging the recent advances in machine learning and state-of-the-art privacy techniques, Statice enables companies to release highly granular datasets with no risk of identifying a single individual. We empower companies to open up their new synthetic data in a GDPR-compliant manner for product development, training new machine learning algorithms, and unlocking industry-wide insights — internally, or collaboratively with partners. Get in touch with us to learn how privacy-preserving synthetic data can be of help to you and the data-driven future of your business.

The GDPR and the future of data-driven innovation

80

the-gdpr-and-the-future-of-data-driven-innovation-17b09a40abb

2018-06-27

2018-06-27 22:10:59

https://medium.com/s/story/the-gdpr-and-the-future-of-data-driven-innovation-17b09a40abb

false

1,580

Statice is a tech startup based in Berlin. We allow you to freely share and collaborate over sensitive and private customer data without ever exposing it.

null

Statice-2020951004799331

null

Statice Stories

social@statice.io

statice-stories

null

StaticeBerlin

Privacy

privacy

Privacy

23,226

Sebastian Brent Weyer

Working on statice.ai - always happy to chat!

f4f2f223d94d

sebastianweyer

45

42

20,181,104

null

0

null

0

634d4b270054

2018-04-13

2018-04-13 12:17:17

2018-04-13

2018-04-13 12:18:19

1

false

en

2018-06-05

2018-06-05 08:50:53

3

17b122440282

1.188679

0

Lately, the DJI Drone Photography Award, make creative use of a drone to explore new photographic possibilities. The drone work is…

5

Drone Photography Balances The Tragedy With Beauty Lately, the DJI Drone Photography Award, make creative use of a drone to explore new photographic possibilities. The drone work is awe-inspiring, letting us consider the world from an alternative perspective. The Sand Castles (Part II) by Markel Redondo focuses on highlighting the Spain’s problem from a new perspective. “We live in a society with huge housing issues, where many cannot afford a place to live, yet Spain has more than three million empty homes,” said Redondo. The Salt Series by Tom Hegen documents salt production across Europe. Hegen loves exploring the relationship between man and nature and uses aerial photography via drone to document landscapes that have been transformed by human intervention. “The production of sea salt is one of the oldest forms of human intervention in natural spaces but we rarely ask where it actually comes from and how it is being produced,” said Hegen. Have a look at the winning photographs, by photographers Markel Redondo and Tom Hegen. Source: https://bit.ly/2qtu639 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.

Drone Photography Balances The Tragedy With Beauty

0

drone-photography-balances-the-tragedy-with-beauty-17b122440282

2018-06-05

2018-06-05 08:50:55

https://medium.com/s/story/drone-photography-balances-the-tragedy-with-beauty-17b122440282

false

262

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

2018-06-27

2018-06-27 04:41:46

2018-06-27

2018-06-27 04:53:52

1

false

en

2018-06-27

2018-06-27 04:53:52

1

17b13153fee2

3.041509

0

There have been numerous studies that show that AI contrary to conventional wisdom does not in fact reduce employment, rather it creates…

5

AI won’t steal your marketing job, here’s why There have been numerous studies that show that AI contrary to conventional wisdom does not in fact reduce employment, rather it creates jobs. Here are my own views and experience of how AI creates work. When people think about implementing AI to decision what offer or promotion to present to a customer (in any channel), this simultaneously starts the conversation about what offers they have. In many cases non-retailers will say that they don’t need AI, because they don’t have enough offers or things to say to customers. Conventional decisioning methods like next best offer models or business rules can do the job. On one hand this makes perfect sense, a limited amount of offers, the increase in level of accuracy provided by AI probably doesn’t add much value. The problem is this argument kinda misses the point. If you take an insurance company for example, they would reasonably argue that they don’t have many things to sell to a customer; a different insurance policy, an upgrade or some additional benefits. This can be managed by a simple rule or model that says if the customer doesn’t have it, offer it, which might come down to 10 different offers in total.This is what I call the Least Worst Offer (LWO). Even though we go to the trouble of building a model or rules, we approach the problem from the brand’s perspective not the customer’s perspective. It’s not the best offer we can make a customer, it’s the least worst of all of the things we could say. The concept of LWO is an important one, because many organisations fool themselves into thinking they are being customer-centric and granted they are better than asking everyone “Would you like fries with that?”. But is it really as good as it gets? This is where AI meets job creation in marketing. It starts with the idea that there are 10 different offers or interactions an insurance company can have with a customer. But let me ask you this, if a brand has 5 products, with 10 features each, 3 different pricing plans and 1,000,000 customers, how can they only have 10 different offers? Is it possible to distill that level of diversity and complexity into 10 offers? Of course it’s not. In the past we have been working with many constraints, which forced us to limit the number of customer offers, not least of which has been our ability to create and manage a larger range of offers. Then what does AI really do? AI releases resources from the operations of offer management, so that they can be redeployed into creating offers and messages that are more likely to resonate with your customers. We, at Digital Alchemy, have created a framework to help you with this. This framework develops interactions aimed at motivating customers based on easy-to-understand psychological techniques, in total there are more than 65 techniques in the framework. Here is an example of how it works. Imagine you have a retention offer for a home loan… A standard retention offer for a home loan might look something like this - “Your home loan is due for renewal soon, renew now for a 0.5% discount on the rate.” However, when using our framework, here are some additional propositions based on a handful in the overall framework; Social Proof — “90% of our customers renew their home loans with us, renew yours now and get a 0.5% discount.” Nostalgia effect — “We love that we have helped you buy your home 10 years ago. Don’t forget to rollover your loan, renew yours now and get a 0.5% discount for old times’ sake.” Sunk Cost Fallacy — “Avoid the hassle of reapplying for your loan, rollover now and get 0.5% discount.” Fear of Loss — “Rollover your home loan now or your special 0.5% discount will expire in 7 days.” Different customers will be responsive to different techniques. This is where AI can help decision and optimise which technique is right for each particular customer. Now imagine being able to connect with your customers at this deep psychological level rather than taking a punt on one message for all. Unless you are in an industry where all your customers are of a single psychological profile, you no longer need to be content with one offer, one proposition, one message. Build 1 million stories for 1 million customers, resulting in better customer engagement and more value created.

AI won’t steal your marketing job, here’s why

0

ai-wont-steal-your-marketing-job-here-s-why-17b13153fee2

2018-06-27

2018-06-27 04:53:52

https://medium.com/s/story/ai-wont-steal-your-marketing-job-here-s-why-17b13153fee2

false

753

null

Engagement Marketing

engagement-marketing

Engagement Marketing

124

Regan Yan

CEO of Digital Alchemy & Rabbit Rewards. Working alongside marketing leaders to drive performance and ROI through automation and a customer level orientation.

b21945482a2

reganyan

1

37

20,181,104

null

0

null

0

null

2018-09-15

2018-09-15 16:23:10

2018-09-15

2018-09-15 17:34:28

1

false

en

2018-09-15

2018-09-15 17:44:48

3

17b370e33cd9

3.033962

2

0

Everyone has a different opinion and definition of success.

4

The success of AI is the key Everyone has a different opinion and definition of success. Building blocks My definition of success Success begins in the silent chambers of our brains. If you are delusional enough to believe that you are a winner, it will just be a matter of time until you materialize it. A wise man once said: “Great battles are won in the tent” Having the courage and perseverance to fail multiple times and not give up, is one of the arsenals that highly successful use, only because of their fear giving up is greater than the fear of failing. Of course, they win but they fail more than everyone else because they know something most people don't know, which is: “Success is a journey, not a destination. “— Ben Sweetland This is not The Secret or think positive, believing is one of the secret ingredients for success but that’s not all, in order for all to work you have to add a lot of work into the recipe. “Being successful means that you’re working hard and walking your walk every day. You can only live your dream by working hard towards it. That’s living your dream.” — Marlon Wayans Why I believe that AI is the key For starters, I want to clear out a big misconception people have about what AI is, sci-fi gave the masses an idea that AI is here to steal our jobs and eventually exterminate us. AI can, will and is here to augment us, the earlier we join forces towards the rapid development of this field, valuing the different point of views everyone has, we can make AI great. With the tireless power of AI on our side, we can solve the biggest problems humanity is facing currently. This way giving us more time to be what we truly are which is being human and be able to focus on what really matters in life which is also being human, loving and be loved, caring and being taken care of. If you are a soldier that wants to join the army of AI scientists or enthusiasts that want to contribute to empowering machines to distil information and give us breakthrough discoveries that would have taken us a few millennia to a make, I want you to know that you are not alone. Your contribution will empower AI to give us a boost that in a natural order of things it wouldn't be possible for us to live long enough to see, one example is, AI can help us become an interplanetary species if not intergalactic. Who knows the secrets that lie beyond this planet of ours that are out there just waiting for us to uncover them. “If we are left to our own experiences, we suffer from a shortage of data”- Stephen Covey AI is still a baby Of course, we are still in the early stage but if we start from now developing autonomous systems that besides thinking on their own have as a priority our safety and possesses human experience so it can understand us, I think with this the future of AI will be brilliant. If we consider the human safety factor as one of the pillars in which we will build AI systems, we will see the biggest advancements in the tech and other fields of all times. “I rather live in infinite delusion rather than limited realism” — Ravi Dubey Like raising a child, they are a mirror reflection of what we are, teach and what they experience. “An apple doesn't fall far from the tree” If you are just getting started or are already professional in the area imagine a world in which with your contribution we have more time to love and be loved above all else, time to be Human which is the greatest gift that has been given to us. Continue to learn, the road to success is not and will never be linear. Failures are the best experiences and practically everything that comes before success. Success is not a destination is an endless line of beneficial and achievable goals. There is always something else that we want after attaining a dream or goal. There are no limits to imagination, dream big and fight for what is yours. Thank you for reading. If you have any thoughts, comments or critics please comment down below. If you like it please give me a round of applause👏👏 👏(+50) and share it with your friends.

The success of AI is the key

11

the-success-of-ai-is-the-key-17b370e33cd9

2018-09-17

2018-09-17 15:50:06

https://medium.com/s/story/the-success-of-ai-is-the-key-17b370e33cd9

false

751

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Prince Canuma

Computer Engineer Student, Web Dev. & ML /AI dev

d21c64e9e7ac

prince.canuma

31

73

20,181,104

null

0

null

0

null

2018-05-23

2018-05-23 08:15:44

2018-05-23

2018-05-23 09:56:50

0

false

en

2018-05-23

2018-05-23 10:04:26

1

17b3772cfc3

1.807547

1

0

Phase I : Detection of automated scans (Kinda Anomaly Detection)

5

Machine learning and security(I) Phase I : Detection of automated scans (Kinda Anomaly Detection) Machine learning is eating the world. From communication and finance to transportation, manufacturing, and even agriculture, nearly every technology field has been transformed by machine learning and artificial intelligence, or will soon be. With machine learning offering (potential) solutions to everything under the sun, it is only natural that it be applied to computer security, web-application security and cyber security, fields which intrinsically provides the robust data sets on which machine learning thrives. Automated scanners are automated tools thats the scan web application, normally from the outside, to look for security vulnerabilities such as Cross-site scripting, SQL Injection, Command Injection, Path Traversal and insecure server configuration. More details about automated scanning tools can be found here.. https://www.owasp.org/index.php/Category:Vulnerability_Scanning_Tools I tried to detect automated scans that are running in my web-application using ML and found some really promising results. Takeaways : Before diving deeply into which algorithm to use we have to analyze the use case first. There will be two conclusions that i made 1) We have very few automated user agents/points which are anomaly and 2) Distance between anomaly user agent and normal user agent are far. Challenges while having ML in Security: More false positives and detecting anomaly user agent(automated scans) is very difficult Dataset : 4 Dimensional data => Avg. request param count, Avg. response time, Avg. response status, Total number of request. .. 1)After collecting my data i tried to apply clustering algorithm DBScan to get all the odd mans out by setting average distance between the points as epsilon and minimum points in that cluster as 10. So the groups which have lesser number of points(10 points) are detected. And those points are mainly the automated user agents. Above approach will look like semi supervised learning as my data is labeled and also i am tuning my epsilon and minimum number of points with that data. .. 2)Another approach i tried is by having one class SVM. This is a bit easy, i did a trick by giving training data as day-1 data for day0 and this looks unfair but i am impressed by the results ;). This approach is purely learning from past days and if there’s any anomaly user agent i will mark it and remove that from next day training. (Need to have automated one to verify the automated user agent) .. 3)Also i clubbed k-means and db-scan/kNN. First i will run k-means with k as 3,5,7,9 and 11. Then take the points which are far away from the cluster centers(say top 50 points in each cluster, this will result in 250 points) and run db-scan/kNN on the resultant data set. References : 1) Machine Learning and Security by David Freeman; Clarence Chio 2) Darktrace : An enterprise immune system

Machine learning and security(I)

35

machine-learning-and-security-i-17b3772cfc3

2018-05-24

2018-05-24 06:27:45

https://medium.com/s/story/machine-learning-and-security-i-17b3772cfc3

false

479

null

Machine Learning

machine-learning

Machine Learning

51,320

harry

There is no sudden penetration of knowledge without gradual practice and training

dd24dd14424

harry_nallasamy

22

39

20,181,104

null

0

null

0

null

2018-04-19

2018-04-19 12:43:13

2018-04-19

2018-04-19 12:43:05

1

false

tr

2018-04-19

2018-04-19 12:45:45

2

17b49da26d62

2.10566

0

null

5

Veri Bilimi ve 2018'de Veri Bilimci Olmak 21.yüzyıl mesleklerinden biri olarak lanse edilen veri bilimi, her geçen gün biraz daha popüler hale geliyor. Özellikle teknoloji çağında büyüyen neslin ilgi duyduğu bu meslek, az bilinen ve fazlasıyla ihtiyaç duyulan dallarından biri. Big data teknolojisiyle birlikte ortaya çıkan veri bilimci, herhangi manuel veya yargısal bir hataya fırsat veremeden, geniş çaplı ve karmaşık verileri incelemekle sorumlu kişidir. Sorulara ve sorunlara veri analizine dayalı cevaplar bularak veri odaklı karar vermeye ve iş stratejileri oluşturmaya yardımcı olur. Veri bilimi hemen her alanda ihtiyaç dahilinde olduğu için, veri bilimcileri günümüz şartlarında ciddi anlamda birer kurtarıcıdır. “Nasıl veri bilimci olabilirim?” son dönemde Quora’da en çok cevaplanan sorulardan biri oldu. Bu nedenle biz de veri bilimi ile ilgilenmek, veri bilimci olmak isteyenler için bilgilendirici bir içeriğe yer vermek istedik. Eğer bir veri bilimcisi olmakla ilgileniyorsanız; Size vereceğimiz en iyi tavsiyemiz, şimdiden yolculuğunuza hazırlanmaya başlamak olacaktır. Temel kavramları anlamak için zaman ayırmak, bu alanla gerçekten ilgilenip ilgilenmediğinize de karar vermenize yardımcı olacaktır. Bir veri bilimcisi olma yoluna başlamadan önce, bunu neden yapmak istediğiniz konusunda kendinize karşı dürüst olmanız önemlidir. Bu konuda kendinize sormanız gereken bazı sorular olabilir: İstatistiklerden ve programlardan hoşlanıyor musunuz ya da en azından şu ana kadar öğrendiklerinizden hoşlandınız mı? En yeni teknikler ve teknolojiler hakkında sürekli olarak öğrenmeniz gereken bir alanda çalışmaktan keyif alır mısınız? Ortalama bir maaşa sahip olacağınızı bilseniz bile bir veri bilimcisi olmak ister misiniz? Diğer iş konularında da (örneğin veri analizi vb.) iyi misiniz? Kendinize bu soruları sorun ve kendinize karşı dürüst olun. Evet cevabını verdiyseniz, o zaman bir veri bilimcisi olmak için yola çıktınız demektir. Bir veri bilimcisi olma yolu, muhtemelen önceki deneyiminize ve ağınıza bağlı olarak biraz zaman alacak. Buna da hazır olmalısınız. Biraz somut unsurlara değinmemiz gerekirse eğer, her şeyden önce matematik ilgi duyduğunuz alanlardan biri olmalı. Eğer matematik tutkunuysanız ve olasılık hesaplamak size keyifli bir bulmaca çözmek gibi geliyorsa, iyi veri bilimci olma yolunda daha kolay ilerleyebileceğinizi söyleyebiliriz. Tabi ki matematik bilmek tek başına yeterli değil. İstatistik ve programlama alanında da gerekli becerilere sahip olmak gerekiyor. Bir liste yapmak gerekirse eğer; Veri bilimcilerde bulunması gereken becerileri: Veri tabanı bilgisi Tahmini analitik ve ML Büyük veri bilgisi Sunum becerileri Bir veri bilimcisi, her zaman istatistikleri anlama ve yorumlamada iyi olan kişidir. Zaten iş dünyasında bu kadar talep yoğunluğunun oluşması da bu becerilerine bağlı. Şu anda Birçok şirket hem geleceğe yönelik stratejiler yapmak hem de çalışanların perspektifini oluşturmak için veri bilimi çalışanlarından yararlanıyor. Bir anlamda şirketteki rolleri oldukça büyük. Bu nedenle şirketlerde en yüksek ücreti alan çalışanlardan olmaları da şaşırtıcı değil. Bizler belki de her gün pazarlama görüşmeleri yapıyoruz, ama lüks bir otomobili satın alması için insanları nasıl çağıracağımızı bilmiyoruz. Alıcısının çok az olduğu bir aracı pazarlamanın da arkasında çok fazla analitik öğrenme var, alıcısı çok olan aracın arkasında da. Pazarlama yapmaya başlamadan önce hangi ürün için müşteri nasıl çağırılır, o ürün için kaç adet arama yapılmıştır, ne kadar veri harcanmıştır ve günün hangi bölümünde harcanmıştır gibi bilgileri bile bilmek gerekir. İşte dünyanın en iyi pazarlama stratejilerinin arkasında da veri bilimi ve veri bilimcileri vardır.

Veri Bilimi ve 2018'de Veri Bilimci Olmak

0

veri-bilimi-ve-2018de-veri-bilimci-olmak-17b49da26d62

2018-04-19

2018-04-19 12:45:46

https://medium.com/s/story/veri-bilimi-ve-2018de-veri-bilimci-olmak-17b49da26d62

false

505

null

Data Science

data-science

Data Science

33,617

Datateam Bilgi Teknolojileri

http://www.datateam.com.tr/

48081bdbc544

socialdatateam

9

11

20,181,104

null

0

pip3 install boto3 #!/usr/bin/env python3 import boto3 #Function for starting athena query def run_query(query, database, s3_output): client = boto3.client('athena') response = client.start_query_execution( QueryString=query, QueryExecutionContext={ 'Database': database }, ResultConfiguration={ 'OutputLocation': s3_output, } ) print('Execution ID: ' + response['QueryExecutionId']) return response {"name":"Alice","sex":"F","city":"Seattle","country":"USA","age":25,"job":"Professional Zombie Killer"} {"name":"Bob","sex":"M","city":"Los Angeles","country":"USA","age":40,"job":"Actor Extraordinaire"} {"name":"Joe","sex":"M","city":"New York","country":"USA","age":35,"job":"Policeman"} {"name":"Amanda","sex":"F","city":"Los Angeles","country":"USA","age":29,"job":"Ex Child Star"} s3_input = 's3://athena-how-to/data' database = 'test_database' table = 'persons' create_database = "CREATE DATABASE IF NOT EXISTS %s;" % (database) create_table = \ """CREATE EXTERNAL TABLE IF NOT EXISTS %s.%s ( `name` string, `sex`string, `city` string, `country` string, `age` int, `job` string ) ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe' WITH SERDEPROPERTIES ( 'serialization.format' = '1' ) LOCATION '%s' TBLPROPERTIES ('has_encrypted_data'='false');""" % ( database, table, s3_input ) #Query definitions query_1 = "SELECT * FROM %s.%s where sex = 'F';" % (database, table) query_2 = "SELECT * FROM %s.%s where age > 30;" % (database, table) #!/usr/bin/env python3 import boto3 #Function for executing athena queries def run_query(query, database, s3_output): client = boto3.client('athena') response = client.start_query_execution( QueryString=query, QueryExecutionContext={ 'Database': database }, ResultConfiguration={ 'OutputLocation': s3_output, } ) print('Execution ID: ' + response['QueryExecutionId']) return response #Athena configuration s3_input = 's3://athena-how-to/data' s3_ouput = 's3://athena-how-to/results/' database = 'test_database' table = 'persons' #Athena database and table definition create_database = "CREATE DATABASE IF NOT EXISTS %s;" % (database) create_table = \ """CREATE EXTERNAL TABLE IF NOT EXISTS %s.%s ( `name` string, `sex`string, `city` string, `country` string, `age` int, `job` string ) ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe' WITH SERDEPROPERTIES ( 'serialization.format' = '1' ) LOCATION '%s' TBLPROPERTIES ('has_encrypted_data'='false');""" % ( database, table, s3_input ) #Query definitions query_1 = "SELECT * FROM %s.%s where sex = 'F';" % (database, table) query_2 = "SELECT * FROM %s.%s where age > 30;" % (database, table) #Execute all queries queries = [ create_database, create_table, query_1, query_2 ] for q in queries: print("Executing query: %s" % (q)) res = run_query(q, database, s3_ouput) me@linuxbox$ ./athena-script.py Executing query: CREATE DATABASE IF NOT EXISTS test_database; Execution ID: a7754952-****-****-****-65970b60e580 Executing query: CREATE EXTERNAL TABLE IF NOT EXISTS test_database.persons ( `name` string, `sex`string, `city` string, `country` string, `age` int, `job` string ) ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe' WITH SERDEPROPERTIES ( 'serialization.format' = '1' ) LOCATION 's3://athena-how-to/data' TBLPROPERTIES ('has_encrypted_data'='false'); Execution ID: ed98c98f-****-****-****-56b0ae69cca9 Executing query: SELECT * FROM test_database.persons where sex = 'F'; Execution ID: f68e622a-****-****-****-fd5aea939b9a Executing query: SELECT * FROM test_database.persons where age > 30; Execution ID: aa047728-****-****-****-a414963f1c98

15

null

2017-08-01

2017-08-01 03:52:46

2017-08-01

2017-08-01 03:53:59

1

false

en

2017-08-01

2017-08-01 03:54:53

4

17b4d0c592b6

3.6

31

6

0

Introduction

5

Automating AWS Athena batch jobs with Python 3 Introduction AWS Athena is certainly a powerful tool for all those people that need to analyze vast amounts of data in S3. It’s as simple as dumping your data in S3, define the database and data format and voila! You can now analyze your data using standard SQL queries. Since AWS service logs from Cloudwatch, RDS, ELB, and IAM are all in JSON and dumped in S3, it’ll be all queryable making security or operational audits less tedious for all AWS systems engineers/administrators out there. It is really easy, amazingly fast and cost effective at $5 per TB compared running custom EMR jobs which require huge costly short lived machines that take forever to run and a big headache if it fails mid process. Just don’t forget to put an object lifecycle rules on your output S3 bucket to avoid extra storage costs as all query results get dumped to S3. I’m not gonna dive on the inner workings of Athena but if you want to learn more here’s the link to the documentation page. Without further ado, here’s a short how-to to automate Athena batch jobs using a simple python3 script to get you started. Installing and configuring the Boto3 SDK Install the SDK to make API calls to AWS. To configure the credentials please refer to the link below and setup the authentication method best suited to your situation. http://boto3.readthedocs.io/en/latest/guide/quickstart.html#configuration Function to query Athena Here we define a function that we can reuse for all our queries and accept three basic parameters — the query, database and a custom s3 output path. Sample S3 data Athena supports a variety of data formats and compression formats. For this tutorial, we will just use a plain old JSON file uploaded to S3. Creating the database and table Much like a normal SQL server, a database must be created first to house all the tables which direct to the s3 data based on the ‘LOCATION’ attribute defined during table creation as seen below. Select queries Standard S3 queries that we’re all familiar with. Pasting it all together API calls on Athena are asynchronous so the script will exit immediately after executing the last query. As of this writing, boto3 still doesn’t provide a waiter. You will just have to write your own waiter based on the execution ID returned. Now time to execute the script. Since this will just query one JSON file it’ll finish instantaneously and immediately show up in S3. It’s up to your next script or job to process this further. From my production experience, our 800GB($4) Athena batch job finishes around 15 minutes down from 1–2 hours from our previous EMR based solution that costs around 20–30$ per run. It was a win in all aspects when we moved to Athena, less cost, less time less complicated, and easy to learn, manage and automate. That’s it, hope this helps! Happy automating!

Automating AWS Athena batch jobs with Python 3

136

introduction-17b4d0c592b6

2018-06-09

2018-06-09 23:21:53

https://medium.com/s/story/introduction-17b4d0c592b6

false

901

null

Aws Athena

aws-athena

Aws Athena

12

DevOps Global Elite

null

636dd153dee1

devopsglobaleli

65

57

20,181,104

null

0

null

0

null

2018-08-30

2018-08-30 15:07:23

2018-08-31

2018-08-31 14:13:41

3

false

en

2018-08-31

2018-08-31 14:22:43

6

17b4ea94a27e

2.591509

1

0

While being transformative as a concept, we’re still not fully aware of how revolutionary will artificial intelligence (AI), and machine…

5

What if ads are controlled by a machine? While being transformative as a concept, we’re still not fully aware of how revolutionary will artificial intelligence (AI), and machine learning (ML) truly become. We are just beginning to find out different applications of these technologies in the marketing and advertising industry. Combining human efforts with algorithms is the way forward. What can a machine do without a human? At Adssets we always strive to automate and simplify creative work. The goal is to use as much relevant content (decided by a human) as input, and match that we conditional data to increase relevance and engagement. Algorithms all the way The machine lets us write algorithms that select different sources of data based on predefined scenarios. In this case the defined scenarios are basically twofold, (1) machine controlled outcomes and (2) predicted models. We use these both to apply control points that match our clients predictions. Algorithms help our clients optimize and make informed decisions based on real time data and ultimately improve their campaign results. Let me explain with an example both of these. Machine controlled outcome. At Adssets we use Ad Desicioners to manage and control the outcomes of our customers campaigns. Our vision for the Ad Decisioners is to help clients get real time insights about their campaign performance and user behavior. For example adding 10 different ads into A/B Simple could result in four different optimizations; (I) score based on viewability, (II) score based on Engagement/Interactions, (III) score based on clicks and (IV) combined scores from any of the above. The algorithm is easy to understand, you rank the ad and then you can let the machine combine different scoring from different parameters to get a combined score. Naturally you automate so that only the top ranking ad will get traffic…but for how long? Another job for the AI or not? Predictive model In this scenario you have insights of a desired outcome with a set of control points. You know that if the following parameters are met your registrations will be maximized. Parameters would look like this (i) You reach above 0.5% in clicks, (ii) View time is above 2secs and (iii) the add is shown with a audience specific message. This is easy to automate and optimize for, but you need more data then just ten ads. Preferably you need 100’s if not 1000’s, hence ad automation is needed. What if we want to feed learning into the Predictive Model? Ex. when the skies are blue which content works best? Is that AI? “person holding black tablet computer” by Brooke Cagle on Unsplash To keep it simple we need to know what we want to achieve. To get the right outcome therefore is a combination of human and machine better and cheaper then only AI. With only AI you have unknown elements to be able to calculate outcomes, what happens when it is working, you’re unsure. It will be extremely hard to visualize WHY the AI made these decisions and HOW it’s better than other scenarios. You‘re basically handing over the trust to a Machine with a undefined algorithm if you automate it. Then again, AI may be the right way to visualize scenarios that you then stick into your Machine Learning tools…like we do at Adssets!

What if ads are controlled by a machine?

11

what-if-ads-are-controlled-by-a-machine-17b4ea94a27e

2018-08-31

2018-08-31 14:22:43

https://medium.com/s/story/what-if-ads-are-controlled-by-a-machine-17b4ea94a27e

false

541

null

Machine Learning

machine-learning

Machine Learning

51,320

Adssets AB

Adssets supplies leading brands with Rich Media mobile solutions. Our mission is to maximise time spent between brand and consumer while delivering world-class

3143d11796df

adssets

4

5

20,181,104

null

0

null

0

d800127b34b8

2018-06-11

2018-06-11 21:16:01

2018-06-11

2018-06-11 21:02:58

1

false

en

2018-08-22

2018-08-22 18:56:46

8

17b5be723cd3

3.879245

1

0

The Age of Big Data is upon us! As companies compete on a global scale, they are amassing ever more information to give themselves a…

4

Information Intel: The Promises & Pitfalls Of Data For Companies Of All Sizes The Age of Big Data is upon us! As companies compete on a global scale, they are amassing ever more information to give themselves a strategic advantage. This deluge of data opens up enormous opportunities for firms to more effectively target customers, bid for resources, and organize their operations. But it also creates serious risks, which any organization that attempts to use Big Data must prepare for. Many people assume that Big Data is only valuable for large companies, and thus that small firms have no need to learn about it or prepare for its risks. In fact, firms of all sizes and industries can benefit immensely from the rise of data. Thus no matter what your company does or what scale you operate on, you need to prepare for all the effects of Big Data, including: Positive Promises Big Data has no shortage of benefits for businesses. By amassing data from every possible source and analyzing it effectively, firms of all sizes have the opportunity to: Market More Effectively– Modern marketing is entirely dependent on data. Companies need to know what customers’ preferences are, down to subtle details like what time of day they like to shop or what types of items they tend to buy in tandem. The more data that a company has, the easier it is to identify strong correlations between customers’ search and shopping habits. Firms can then target each customer with the specific marketing content they are likely to respond to, exactly at the time of day they’ll be on the Web. This dramatically raises the chance that a customer will respond positively and make a purchase. Develop New Products– Besides affecting the way that companies present their products, Big Data affects what products those companies offer in the first place, notably by helping them develop new items to meet existing needs. This is particularly true for health researchers and medical technology firms, which often struggle to develop drugs and equipment that can address common diseases. By gathering a cornucopia of data and interpreting it effectively, these firms and researchers can more quickly and accurately determine the causes of even the most mysterious diseases. From here, it’s just a short step to figuring out what methods can treat or cure those illnesses. As a result, medical companies can quickly bring new treatments to the market, allowing them to earn steady profits while improving public health. Organize Operations– If a firm wants to stay profitable for the long haul, it must regularly streamline its operations and cut unnecessary costs. But this is easier said than done, given that many sources of waste are subtle and difficult to detect. But Big Data offers the opportunity for companies to assess and improve efficiency like never before. The more data they gather on their operations and the more effectively they interpret it, the easier it is for a firm to find redundant activities, inefficient production processes, and other sources of waste. This lets them cut costs, boost profits, and stay successful. These Big Data benefits are valuable to all companies, but they’re particularly likely to help the smallest ones. Small firms often struggle to cut costs because they can’t achieve economies of scale, and they have trouble appealing to customers without a widely-recognized brand. But with Big Data, they can identify all opportunities for savings while determining and offering exactly what it is their customers want. In this way, they may be able to compete on a level playing field with even the largest brands. Big Data Disadvantages As with many new developments and economic trends, the pitfalls of big data are every bit as potent as its advantages. Companies can’t just begin gathering information willy-nilly. Not only is a deluge of data not particularly useful by itself, but it can actually harm firms’ operations if they aren’t careful. The problem is principally one of processing. Companies are gathering far more data than it has ever been possible for human beings to interpret on their own. They thus have become dependent on computers to analyze it for them. But there are limits to how well current processing technology can understand data, particularly visual images and other items that computers have always had trouble recognizing. This means human operators must still do a lot of the work of labeling data, as well as determining what specific correlations that computers should search for in it. With so much data to sort through, human operators often feel stuck and confused. This leads to lower levels of productivity, combined with an inability to find much useful information in the data provided. Such problems are particularly serious for small companies, which have fewer employees to handle all this information. As a result, they end up losing money on Big Data and come to view it as a poor investment. The good news is that processing technology is becoming more efficient and sophisticated each year. Of particular note are recent advances in artificial intelligence, which are allowing us to create computers that mimic the form and function of the human mind. AI applications have the ability to process much larger amounts of data than previous generations of computers. They can also understand it in more complex ways, notably by recognizing sounds and visual images in addition to words. At this rate, it won’t be long before firms can handle even the largest loads of data with ease. Imaginea Ai is amassing the applications, data, and expertise to make artificial intelligence a reality. For more information on our efforts and all that big data can do for the world, visit imaginea.ai today. Originally published at imaginea.ai on June 11, 2018. Want to hear more from Imaginea Ai? Follow us on Instagram, Facebook, LinkedIn and Twitter.

Information Intel: The Promises & Pitfalls Of Data For Companies Of All Sizes

1

information-intel-the-promises-pitfalls-of-data-for-companies-of-all-sizes-17b5be723cd3

2018-08-22

2018-08-22 18:56:46

https://medium.com/s/story/information-intel-the-promises-pitfalls-of-data-for-companies-of-all-sizes-17b5be723cd3

false

975

Build better AI. Faster. Together.

null

imagineaai

null

Imaginea Ai

marketing@imaginea.ai

imaginea-ai

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

imagineaai

Big Data

big-data

Big Data

24,602

Imaginea.Ai

We are an #AI company driven to help industries work better, faster and smarter.

37340374daff

imaginea

10

13

20,181,104

null

0

null

0

null

2018-09-21

2018-09-21 17:53:35

2018-09-21

2018-09-21 17:56:20

0

false

en

2018-09-21

2018-09-21 17:57:09

0

17b6c92f7b9d

2.69434

0

If you’ve worked in a customer support role, then you’re well aware of how many inquiries could be settled by a simple Google search…

1

4. 24/7 Customer Support If you’ve worked in a customer support role, then you’re well aware of how many inquiries could be settled by a simple Google search. Rather than bleed payroll, AI chatbots are now able to respond to simple requests to save both time and money. First touch points are handled by AI and then either settled or pushed to an actual representative. This drastically drops off a large portion of inbound requests that are handled by the AI, freeing up the time of your actual employees. Google recently showed off their Duplex AI system that can replicate human speech with startling accuracy to the effect of even being able to incorporate vocal subtleties like ‘umms’ and ‘ahhs’. The application this could have for not only customer support but also for sales would be unprecedented. The machine learning component for customer support resides in testing interactions and building more successful responses and conversation pathways. Each company would have an AI chat system that would improve over time based on successful outcomes. The chat system is deployed with a generalized flow at the beginning for measuring and testing purposes before narrowing down on the best ones which are then again tested and refined. The final product is a customer support team that can now focus on broader issues rather than be inundated with trivial matters. 5. Dynamic Pricing Dynamic pricing incorporates machine learning in order to deliver varying pricing for every potential customer. What this would look like is a user being served the price of $300 due to being identified as from a high income bracket, whereas another user being identified from a low income bracket would be served with the price of $200. In contrast to A/B testing that requires someone to run it, dynamic pricing is controlled by an algorithm that works in the background. As with any machine learning model, the prolonged use of it will only serve to bolster its efficacy. Algorithms can even go so far as to look at what competitors are pricing products at, and factor that into the model as well. The data that the algorithms can process ranges from previous purchasing behaviour, financial status, nearby events, location, seasonality, time of day, and more. Ultimately if relevant data on something can be acquired, then it can be integrated into the AI engine. Demand forecasting can also be a component of what goes into the algorithm and can be used to anticipate heightened or lowered demand in order to reflect the best price. 6. KPI Analysis and Monitoring Salespeople are losing out on important work time to record and track their KPIs in part due to the widespread presence of customizable performance dashboards for almost all CRMs. When given countless options for dashboard customization, salespeople will divert from their work to select and arrange dashboards to highlight their strengths and bury their weaknesses. Machine learning models will be able to not only be able to uncover deep insights about KPIs but also record and log them automatically without the need to take salespeople out of their workflows. With AI, there won’t be a need to set predefined KPIs and track for them. Because AI engines log vast amounts of data on every individual employee, they will be able to identify customized KPIs that are unique to the individual in order to objectively highlight their best and worst metrics. In order to match an AI’s capabilities, an organization would need to employ a full-time team of data scientists — and even then they would get obliterated on a time, cost, and efficiency comparison, all while the AI engine is working in the background. Basix has two AI engines that work in isolation yet also feed off one another. The Smart Engine is the master tactician that tees up the next piece of work that’s been predictively labeled as most urgent and impactful. The Insight Engine is the unrelenting bookworm that tirelessly sifts through data looking to find game-changing insights that can be both tested by the Smart Engine and visualized for every team member. Signing up for a demo will allow you to test drive the AI engines and see for yourself what it means to truly supercharge your team. Book with us now!

4. 24/7 Customer Support

0

4-24-7-customer-support-17b6c92f7b9d

2018-09-21

2018-09-21 17:57:09

https://medium.com/s/story/4-24-7-customer-support-17b6c92f7b9d

false

714

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Basix Concepts

Technology startup building the next generation of sales growth products.

dce88cfeb48f

basixconcepts

0

4

20,181,104

null

0

null

0

245daf6aeccd

2017-09-15

2017-09-15 01:58:09

2017-09-15

2017-09-15 05:34:21

1

false

en

2017-09-15

2017-09-15 11:34:25

0

17b6d061a08e

4.516981

1

0

When dealing with data, companies need to understand their path from data collection to selling valuable products and services, or they are…

4

From collection to cash, how to deal with data When dealing with data, companies need to understand their path from data collection to selling valuable products and services, or they are just going to waste their time and money on consultants. I’m going to talk about the relationship between two mythical companies, a large fertiliser supply and service company called Ferty and a large agricultural company called Farmo. We will follow Ferty as it collects and uses data to increase the value of its service and then builds new product opportunities. Suppliers in traditional industries like agriculture can create additional value for customers by collecting, sharing and interpreting data. Image from philipwareing.co.nz, a service provider that advertises GPS control and mapping capability Data needs to be clean Data is more usable when it is relatively free of contamination. In the same way that cities don’t allow dirty heavy industry in their drinking water catchments, you should work hard to improve the cleanliness of data flowing into your system. You should be doing the difficult work of ensuring the data is captured as cleanly as possible and flows through to storage reliably, this is more sustainable than trying to clean contaminated or incomplete data sets later. Example Ferty manufactures and spreads fertiliser from its trucks directly onto farm fields. Ferty believes that collecting data is important to its business, so it instruments its trucks with a control system that allows them to report the type, quantity and location of fertiliser spread. If they invest heavily in this project, do a good job of building this control system, physically re-engineer their trucks to allow accurate measurement, have a robust communications network to bring the data back and then store it in a way that is reliable and accessible, they will have clean, usable data. Ferty could use this data for invoicing, analysing fertiliser application rates, predicting supply chains and providing insights to customers. However, if they didn’t re-engineer their trucks to measure accurately, their network was unreliable or they dump their data into a place that is not easily accessible, Ferty will never get a return on their investment. Data needs to flow quickly to your customer If you are collecting large amounts of data from doing business with your customers, you should probably prioritise building the pipes to get that data into their systems as quickly as possible. As you connect your systems to their systems, you will need to have a reliable, easy to consume source of usable data available. Also, it is important to understand that data is often most valuable to customers immediately after it is collected, because it can inform near real-time improvement and optimisation. This is why modern companies don’t just email reports, they build APIs. Example Continuing to think about our friends at Ferty, we already know they are collecting lots of high quality, detailed data about fertiliser application on farm fields. Now think about them delivering over thousands of acres of fields owned by Farmo. Farmo uses the latest agricultural management software, recording application of chemicals, growth and yield rate of crops and scheduling production to meet the needs of their customers. Because Ferty stored their data in an accessible place, they are now well placed to offer API access to the fertiliser application data to Farmo. Farmo can take the data flowing in about fertiliser application and combine it with chemical application, growth and yield rate data and further optimise their farming operations. Not only that, because the data is near-real time they can re-schedule their equipment if the data shows fertiliser application is running earlier or later than expected. Selling Bottled Data In California bottled water is mostly sourced from public water sources (it is just tap water), but it is also about 560 times more expensive than tap water. It is important to understand that bottled water isn’t just another source of potable water like the tap is. Instead Coca-Cola Amatil supplies customers with a solution to their thirst, 300ml of cold, clean drinking water in a convenient to carry package right when and where they need it, whether it is by the beach, at the petrol station or in a cooler at a party. If you are trying to create greater value from data, you need to understand how to package a low value, hard to consume commodity and deliver it to a customer as a solution to their problem, right when they need it most. Example Ferty is now in the comfortable position of collecting large amounts of raw data and supplying it directly to Farmo. Farmo combines their live feed of fertiliser data with other agricultural data to optimise their production. But Ferty realises they can also combine it with other data they have in order to provide some discreet, highly valuable insights. So Ferty takes the experimental test data about soil types, rainfall and leach rates that they collected when they created each fertiliser and use it to build a model that predicts the optimum time to apply more fertiliser. Ferty builds this algorithm into a cloud service and is now ready to sell their new optimisation service as a product. With this optimisation service, Farmo can pay a monthly fee to automatically submit their soil information and rainfall data and receive automated recommendations on the optimum date, rate and blend for fertiliser re-application. Farmo chooses to pay Ferty for this optimisation because applying the wrong blend, too early or too late wastes money, reduces yield or slows crop delivery. Don’t just collect data, create value Ferty was in a weak position prior to starting out on its data journey. Farmo was paying for Ferty to drive a truck and spread fertiliser, Farmo was investigating just buying their own truck and buying cheaper fertiliser. Ferty didn’t keep dropping its price, instead it found a way to provide a more valuable service. It first focused on collecting clean, usable data about its operations for its customer, then it provided this data directly and allowed the customer to optimise their operations. After they did this, Farmo is not thinking about buying a truck and cheap fertiliser, because they are too busy thinking about how they can combine this new data with other sources to optimise their end-to-end farming operation. Ferty didn’t charge additional money for this, but they did manage to maintain their product and service premium. After stabilising their market position and focusing on providing a valuable service to their customers, Ferty had the time to explore new product opportunities. In this case, Ferty had the capability to build an optimisation service product that could combine the research and testing they had done for their fertilisers with data their customers already collect. Farmo pays separately for the value this new product provides. Ferty doesn’t mind that using their optimisation service might reduce Farmo’s fertiliser consumption, because they understand that in the long run their success is only possible if they help their customers succeed.

From collection to cash, how to deal with data

1

from-collection-to-cash-how-to-deal-with-data-17b6d061a08e

2017-12-23

2017-12-23 12:27:09

https://medium.com/s/story/from-collection-to-cash-how-to-deal-with-data-17b6d061a08e

false

1,144

The job of building and leading software product teams. Flying Spaghetti Monster by Guillaume Febvrel from Noun Project.

null

Delivering Software

null

delivering-software

LEAN SOFTWARE DEVELOPMENT,SOFTWARE DEVELOPMENT,TEAM COLLABORATION,AGILE,SOFTWARE

null

Data

data

Data

20,245

Austin Turner

Builder and leader of software teams, lover of great food and herder of dogs.

8ae939f38355

austinturner01

5

7

20,181,104

null

0

null

0

null

2017-11-01

2017-11-01 16:53:01

2017-11-01

2017-11-01 15:00:00

3

true

en

2017-11-02

2017-11-02 20:11:00

7

17b7e59ae983

2.519811

8

0

By Silla Brush

4

Robots in Finance Bring New Risks to Stability, Regulators Warn Photo by Tolga Akmen/AFP/Getty Images. By Silla Brush Banks and hedge funds that rely on artificial intelligence threaten to inject risks into the financial system that could exacerbate a future crisis, according to global regulators. The financial industry’s rush to adopt AI raises the potential that firms will become overly dependent on technologies that herd them toward the same view of risks and could “amplify financial shocks,” according to a study published on Wednesday by the Financial Stability Board, a panel of regulators that includes the U.S. Federal Reserve and European Central Bank. “AI and machine learning applications show substantial promise if their specific risks are properly managed,” the FSB said in a report that called for additional monitoring and testing of robotic technologies designed to lessen human involvement. “Taken as a group, universal banks’ vulnerability to systemic shocks may grow if they increasingly depend on similar algorithms or data streams.” The FSB, headed by Bank of England Governor Mark Carney, said that many of the technologies are being designed and tested in a period of low volatility in financial markets, and, as a result, “may not suggest optimal actions in a significant economic downturn or in a financial crisis.” Artificial intelligence is a branch of computer science that aims to imbue machines with aspects of reasoning. The term now includes machine learning, which is the ability for computers to learn by ingesting data, and natural language processing — the ability to read or produce text. The world’s biggest banks and hedge funds are embracing the tools, driven by the availability of major new sources of data that can be analyzed quickly with computer power and at the same time a desire to cut costs and employment levels. Management consultant Opimas LLC estimated in March that AI would result in a cut of 230,000 workers at financial firms worldwide by 2025, with the hardest hit being 90,000 people in asset management. Firms are using AI and machine learning to assess the credit quality of borrowers, price insurance contracts, automate interactions with clients and estimate the risk of trading positions, the FSB said. Hedge funds relying purely on AI and machine learning technologies are growing rapidly and have about $10 billion in assets under management, the FSB said, citing an estimate from a unnamed financial firm. The FSB said technology’s potential to cut costs and drive new profits is even creating an “arms race” among firms to demonstrate their use of AI. In the process, firms may be relying on a small number of third-party technological developers and services. If those were to fail, the effect would ripple across the wider financial system and contribute to major disruptions at large financial firms at the same time. “These risks may become more important in the future if AI and machine learning are used for ‘mission-critical’ applications of financial institutions,” the FSB said. “Moreover, advanced optimization techniques and predictable patterns in the behavior of automated trading strategies could be used by insiders or by cyber-criminals to manipulate market prices.” Originally published at www.bloomberg.com on November 1, 2017.

Robots in Finance Bring New Risks to Stability, Regulators Warn

21

robots-in-finance-bring-new-risks-to-stability-regulators-warn-17b7e59ae983

2018-08-25

2018-08-25 01:42:10

https://medium.com/s/story/robots-in-finance-bring-new-risks-to-stability-regulators-warn-17b7e59ae983

false

522

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Bloomberg

Find more like this at bloomberg.com

3d76181076e6

bloomberg

133,717

225

20,181,104

null

0

null

0

null

2017-12-02

2017-12-02 01:08:08

2017-12-02

2017-12-02 02:16:24

0

false

en

2017-12-02

2017-12-02 02:26:36

6

17b80b88f1dd

3.14717

11

2

0

A month ago, I embarked upon an odd, but ultimately fruitful quest. In prior years, I’d participated in the internet project NaNoWriMo…

5

30 Days, 30 Papers: Machine Learning Writing Month A month ago, I embarked upon an odd, but ultimately fruitful quest. In prior years, I’d participated in the internet project NaNoWriMo (National Novel Writing Month), during which you commit to write 1666 words a day for all 30 days of November. I like the mechanics of this challenge: the way it gets you over the hump of perfectionism and laziness into Actually Doing The Thing, but I’m not really much of a fiction writer. But, I realized there was something I was equally motivated to give myself motivation to do: read and summarize new machine learning literature. And so, Machine Learning Writing Month was born. The terms of the challenge: write at least 500 words, summarizing a ML paper or concept, on each day in November. The reduction in actual written length was to account for the time necessarily spent in reading and comprehending the actual subject matter. The acronym was slightly ill-chosen (MaLeWrimo is a bit uncomfortably on the nose, when it comes to ML-community gender dynamics), and the posts definitely vary a bit in quality (I feel some desire to defensively label the ones written in their entirety after 1 am), but I found the challenge to myself really powerful, and I’m very proud to say I’ve kept up with it to completion. I’ve collected a moderately-organized list of posts at the end of this document, but first, I want to share some overall reflections and strategies I came to during this process. I found it absolutely necessary to put notes in your own words, instead of copying a phrase used by the paper writer. If you haven’t done the work of actually mapping the words to concepts, you haven’t internalized enough to be trusted to explain the idea. One workaround I found, for cases where I wanted to keep track of what was said, but not deceive myself into believing I understood more than I did, was to always use quotes when I pulled sentences direct from the paper. Quotes were my signal to myself that “this is a thing being claimed, but I haven’t yet understood what it means, or why it’s true” Along similar lines, deploying humor in my notes was surprisingly helpful. While I don’t have a great theory on this one, my best guess is that injecting silliness into notes made me feel more like I was talking to a friend, which put me more in the frame of mind of explaining (to myself) rather than simply transcribing Admitting a lack of understanding is enormously freeing. If you force yourself to stay in the register of “fully informed expert”, you push yourself to the extremes of either not sharing any understanding of an idea until you’ve understood it fully, or else pretending you understand something by simply repeating an assertion from a paper, which doesn’t do your readers any favors. Allowing myself to say, “yeah, I didn’t really grok this bit” was both a functionally necessary thing when I’d given myself a time limit and needed to sleep eventually, and also, I think, a useful thing to have gotten in the habit of doing. Repetition is a gift to your reader. One frequently frustrating pitfall of papers was to explain their architecture only once, in one set of words, and leave you to hunt around for clues to build yourself a full picture. When you’re writing a thing down, you have the full concept in your mind, and it can often feel sufficient to simply give one framing of it; one lower dimensional projection, as it were. But, for someone who is entirely new to your idea, one framing might not be enough; it might not tell them what’s novel, what’s relevant; you might just have used a confusing term in your first explanation, and manage to remedy it in a second. Ensembling slightly orthogonal versions of a thing together is a foundational strategy of machine learning practice; it should be foundational to machine learning explanations as well. Summaries List If any of the ideas I’ve covered in these summaries represent things you want to learn, I’d love for you to take a read through and let me know what was useful, what was confusing, and, in general, what feedback you have on my approach. Keeping with the tradition of a community obsessed with gradient descent, the best way to learn is with clear feedback! ImageNet Hall of Fame [3 posts] (ResNet, Inception, & DenseNet) Machine Translation [4 posts] (Translation using monolingual corpuses, and the role of attention in state of the art translation models) Generative Models [5 posts] (Variational AutoEncoders, GANs, PixelCNN, WaveNet) Theory [4 posts] (Capsule Networks, Generalization Theory, Bayesian DL) Reinforcement Learning [8 posts] (My attempt to get up to speed on RL, from basics, through Prioritized Replay and Double Q Learning, up to Alpha Go Zero) Safety & Semi-Supervision [4 posts] (Ladder Networks, Few-Shot Learning, Adversarial Examples, AI Safety)

30 Days, 30 Papers: Machine Learning Writing Month

34

machine-learning-writing-month-17b80b88f1dd

2018-04-17

2018-04-17 20:13:31

https://medium.com/s/story/machine-learning-writing-month-17b80b88f1dd

false

834

null

Writing

writing

Writing

167,305

Cody Marie Wild

machine learning data scientist; lover of cats, languages, and elegant systems; professional curious person.

b6da92126145

cody.marie.wild

1,405

2

20,181,104

null

0

null

0

null

2018-01-07

2018-01-07 11:02:16

2018-01-07

2018-01-07 11:05:31

0

false

en

2018-01-07

2018-01-07 11:05:31

4

17ba50a408d8

0.875472

1

0

Using environments

5

AI and Machine learning tools: lastest Using environments One thing that’s helped me tremendously is having separate environments for Python 2 and Python 3. I used conda create -n py2 python=2 and conda create -n py3 python=3 to create two separate environments, py2 and py3. Now I have a general use environment for each Python version. In each of those environments, I've installed most of the standard data science packages (numpy, scipy, pandas, etc.). Remember that when you set up an environment initially, you'll only start with the standard packages and whatever packages you specify in your conda create statement. I’ve also found it useful to create environments for each project I’m working on. It works great for non-data related projects too like web apps with Flask. For example, I have an environment for my personal blog using Pelican. Sharing environments When sharing your code on GitHub, it’s good practice to make an environment file and include it in the repository. This will make it easier for people to install all the dependencies for your code. I also usually include a pip requirements.txt file using pip freeze (learn more here) for people not using conda. More to learn To learn more about conda and how it fits in the Python ecosystem, check out this article by Jake Vanderplas: Conda myths and misconceptions. And here’s the conda documentation you can reference later.

AI and Machine learning tools: lastest

1

ai-and-machine-learning-tools-lastest-17ba50a408d8

2018-01-07

2018-01-07 13:23:25

https://medium.com/s/story/ai-and-machine-learning-tools-lastest-17ba50a408d8

false

232

null

Python

python

Python

20,142

Pawat Chomphoosang

optimist, passionate technologist, lifehacker, doer

f8c7a1060975

pchomphoosang

33

57

20,181,104

null

0

null

0

f0d3356b0ce0

2018-03-07

2018-03-07 02:46:12

2018-03-07

2018-03-07 05:52:31

3

false

en

2018-03-09

2018-03-09 00:29:09

10

17c0348a568d

2.995283

9

0

Spirit Character Engine is an Authoring Tool and SDK for creating expressive, AI-driven characters within a narrative context. They can…

5

Tutorial: Expressive AI-Driven Conversational Characters in AR: Spirit Character Engine in Unity with ARCore + SALSA LipSync Spirit Character Engine is an Authoring Tool and SDK for creating expressive, AI-driven characters within a narrative context. They can exhibit personality, emotional expressivity, and conversational richness. The way the technology works is this: you pass in user input (whether that's typed or spoken natural language and even gestures) to the API, and Character Engine will select what the most salient response might be — from the Character’s knowledge model and plot information. It then sends this selection in a callback to the game engine — giving you information on the character’s chosen response (in text), their emotional state, and more. Authors Note: this is a demo of the tech stack, and thus definitely not a fully-realized Character Engine character! Given this, it’s easy to integrate Character Engine with a number of options for audiovisual representation using third party tools. A quick, easy option if you want to make an AR character with some animation and lip sync ability tied to the character’s speech output is to use Unity, with the following tech stack: Spirit Character Engine + SALSA + RTVoice + ARCore + Unity Getting Started To get started with Unity & ARCore, you need the latest version of Unity (2017.2+) and the Android 7+ SDK (API level 24 or higher). For more information on this, you can read this official Google quickstart for Unity. Also, download the ARCore SDK for Unity. Create a new Unity 3D project and import the ARCore SDK for Unity .unitypackage file by clicking the menu Assets > Import package > Custom Package and import everything in the Unity package. Once you know the location of your project on disk, export your project from your Character Engine Authoring Tool, by pressing the ‘Export’ button in the menu. First, make sure your Project Settings are set to export to a StreamingAssets folder in your Unity project. Set up your project for Character Engine according to the instructions in your Spirit documentation — this is currently limited access to existing beta clients, but you can also sign up for Beta access now. Character Engine Authoring Tool 5. Next, the rest of the ARCore setup can be found here: https://developers.google.com/ar/reference/unity/ 6. Finally, the SALSA integration is quick and easy — you can download the SALSA with RandomEyes with LipSync from the Unity Asset Store. 7. Install SALSA with RandomEyes into your project by importing from the asset store straight into your project. 8. Install RT-Voice. You can download it from the Unity Asset Store here: https://www.assetstore.unity3d.com/en/#!/content/41068 9. Import the SALSA with RandomEyes RT-Voice support package (Salsa_RTVoice). Select [Assets] -> [Import Package] -> [Custom Package…] and Browse to the [Salsa-RTVoice_1.0.0.unitypackage] file and [Open]. 10. Set up a SALSA 2D or SALSA 3D enabled visual character. The SALSA integration is quick and simple — the SALSA 2D or SALSA 3D AudioSource just needs to be set to the RTVoice (TTS) plugin GameObject/AudioSource. SALSA takes care of the rest. That’s it! Below is a demonstration of this tech stack at work with a very quickly put-together Character Engine project. ! This example demonstrates setting up a character that is rendered in AR, and provides robust yet lower-fidelity animations and lip sync as its a 2D, cartoon-ish character. However, we also support many types of other audiovisual representation for Characters too — such as high-resolution lip-sync and facial animation. In fact, our upcoming demos to be debuted at Game Developers Conference 2018 address these and provides great examples. If you want to get in touch to discuss these, let us know: hello@spiritai.com. Spirit AI builds tools to make the future of digital interactions better: both with virtual humans, and real humans. We make Character Engine, for authoring dynamic improvisational AI characters, and Ally, a tool for detecting and intervening in the social landscape of online communities — to curtail online harassment, or to promote positive behaviour.

Tutorial: Expressive AI-Driven Conversational Characters in AR: Spirit Character Engine in Unity…

22

expressive-ai-driven-conversational-characters-in-ar-spirit-character-engine-in-unity-with-arcore-17c0348a568d

2018-04-24

2018-04-24 04:14:38

https://medium.com/s/story/expressive-ai-driven-conversational-characters-in-ar-spirit-character-engine-in-unity-with-arcore-17c0348a568d

false

648

AI for humans

null

spiritai

null

Spirit AI

hello@spiritai.com

spirit-ai

ARTIFICIAL INTELLIGENCE,MACHINE LEARNING,ONLINE HARASSMENT,PROCEDURAL GENERATION,ONLINE SAFETY

theSpiritAI

Virtual Reality

virtual-reality

Virtual Reality

30,193

Mitu Khandaker

Game designer • Founding team & Creative Partnerships @ http://SpiritAI.com • AI-driven tools to improve the future of games, entertainment & beyond. 👸🏾

2f92df2f11e2

mituk

1,331

415

20,181,104

null

0

null

0

null

2018-05-23

2018-05-23 18:00:34

2018-05-23

2018-05-23 18:34:10

0

false

en

2018-05-23

2018-05-23 18:36:45

2

17c04f88a88d

9.622642

2

0

A few months ago, I quit my job as data scientist at a management consulting firm for reasons outlined in the letter below (originally…

5

Open Letter to Shail Jain, Knowledgent CEO A few months ago, I quit my job as data scientist at a management consulting firm for reasons outlined in the letter below (originally published here and posted on LinkedIn). Board members Manny Arturi and Peter Gibson, Shail Jain, and other senior leadership soon blocked me on LinkedIn while the post racked up over 4,000 views. Interestingly, tens of current and former employees — mostly male, and including the former Chief Data Scientist — added me on LinkedIn and sent messages of appreciation, shock, and even offers of assistance in finding a next position. In the months since I quit I have gathered that there have been no significant changes to leadership at the company, which, while disappointing, is not surprising. If you can identify with the letter, I’m sorry (and we’re in this together!); if you’re surprised by the letter, please think about what you can start doing to discourage these kinds of workplace cultures. Thanks for reading. February 2018 Shail Jain, Knowledgent CEO, I’ve just gotten off the phone with Kerry Ann, your head of HR; given that she only said the investigation into my complaints is closed and that, “appropriate action has been taken,” I wanted to take the time to write a few notes. All of this was previously discussed with you in your office on 6 Feb 2018, with Tom Johnstone listening in on the phone for the better part of the conversation. Quotations are paraphrased unless noted as direct quote. I gave Gregory Banacki, Knowledgent General Counsel, advance notice of my intention to write this. During the meeting in your office, you readily agreed that Ari Yacobi, Knowledgent Chief Data Scientist, is not in fact a data scientist (AKA Adeel Arif, UPenn M.S. Tech Management 2014). You stated that you placed him in that role so that he could “put together a good team.” You did not respond when I asked why he has the title, “Chief Data Scientist.” You also suggested that he did not want me to present to you a Keras TensorFlow neural net image classifier I built before he could see it so that he did not “look dumb” (your words). As we discussed, he has been overseeing your “marquee” client project (your word) for six months with very poor results and a “data scientist” in charge who does not program. You were surprised to hear about the non-programming when I told you, talked to Ari immediately following our meeting, and then he came to me and asked me to take over the project because it was “getting nowhere.” I said no because they’ve already made a mess of it, and he deflated his shoulders and whined, “But (my name), what am I going to dooooo?” When he had to present at a conference in Boston (tweeted from Knowledgent account: “Our Chief Data Scientist Ari Yacobi presented our unique AI-based approach to molecular similarity search at the PRISME Forum’s semi-annual meeting in Cambridge, MA!”), he asked me the week before to do research for him and said that Knowledgent had never actually completed the project he was planning to present; when I asked which similarity algorithm he would tell conference attendees he used, he said he did not know. I asked him point blank how he was going to manage questions from people who actually know what they’re talking about, and he responded: “I don’t know, I’m a little nervous because I’ve never lied this much in a presentation before.” When I texted him the day after my meeting with you to say that I had just given you two weeks’ notice unless I don’t report to him anymore, he responded (direct quote, all [sic]): “I am little shocked and surprised — i took a lot of pride in having a person like you on my team and for finding you. Let’s talk tomorrow” I responded that I had nothing to say until I heard from you first. I have never seen any of his code, nor has he demonstrated he’s capable of giving any meaningful feedback on mine; based upon your admission that he’s “not a data scientist,” my question stands as to why he deserves the title. Tom Johnstone stopped responding to my e-mails when I expressed these concerns plainly to him. Why do you trust him after apparently misleading you for at least six months on your “marquee project”? Why is your senior leadership unwilling to communicate with employees who are actually capable? Tom, why do you talk to other people about what great work I was doing and then not respond to me? Sharad Kumar, Chief Technology Officer (CTO), drove a group of four additional colleagues (including Ari Yacobi, myself, another female coworker, and business analyst Prateek Peres-da-Silva) to an internal happy hour for a client project at a country club in NJ. On the way there, he said, “OK guys, we all know sexual harassment is wrong…. Me Too…. ok, don’t harass people…. but I mean really guys, if you’re not getting sexually harassed, there’s probably something wrong with you, right?” Ari said nothing. At the country club, Sharad came up to me at the bar and asked if there was “anyone special” in my life. I made up an answer. On the way back to Knowledgent HQ with the same people in the car, he repeated, “if you’re not getting sexually harassed, there’s probably something wrong with you.” I asked in a clear, loud voice if he says that to his daughter (whom I believe is 12 based on what he said in the interim). No one spoke. You brought up the fact that both you and Tom Johnstone have daughters; given that you currently oversee a company where this type of incident apparently does not faze senior management, my only response is sympathy to each of your daughters. As you know, your sales men John Rubino and Manus Gallagher sent me to a client site before the Scope of Work (SOW) had been signed. Rubino asked me after my first day, in writing, to provide a list of duties I thought I might be performing there which could be added on the SOW. As you also know, Rubino and Manus also agreed to put my laptop specs in the SOW as I had previously had a very negative experience with 4G RAM on a different client machine. When the client admin contacted me and mentioned the laptop, I forwarded her message to the sales guys to confirm they had included specs in SOW. They said they left out specs because they don’t go in “this kind of SOW” and asked me to deal with client admin directly. Any company that makes data scientists talk to client admin about hardware without putting specs in contracts does not take data science seriously. I played you their apology voicemails which were left within 15 minutes of my Cc:ing Tom Johnstone in a response. Further, Rubino immediately began sending naïve-sounding e-mails to the client admin explaining machine spec requirements seemingly out of the blue. This is a multi-million dollar project. What do these guys get paid for? Another one of your sales guys, Matt Arellano, didn’t know I was being allowed to listen in on the call at the time and said, “look, we all know the model is a piece of shit,” referring to a statistical model I was building. On subsequent calls, after repeated attempts to explain to him how to interpret classification reports, he continued to bungle accuracy and recall and complained that confusion matrices are “too confusing.” It is outside of his capacity to determine whether a model with 85.1% accuracy and 9% recall is a “piece of shit.” I was told by the project manager that Matt had asked me not to speak during the final client presentation. In a dry run with the client project lead, the project manager and client partner contradicted each other. The client turned to me; I had been silent. I calmly made eye contact with the two who had contradicted each other and then explained the results of my work. It was clear and made sense. He told our client partner afterward that he could tell something was not right because I hadn’t been reacting to anything anyone was saying. Fast forward: I met Matt Arellano for the first time in person on a trip to client site several months later to present the results to the client CIO. I answered many technical questions on the fly, including several about slides which he had requested be removed and against which I strongly cautioned. The Knowledgent client partner said he had never heard Matt Arellano so quiet during a meeting. I approached Matt Arellano afterward while he was saying good bye to the client, said good bye to him myself, and he turned and walked away without saying anything to me. This is in front of the client. The project manager from the paragraph above, in my first week at Knowledgent, said to me, “Even more people would listen to you if you wore high heels.” I responded, “I’m a programmer. I wear all black and flats. You wear high heels.” Several months later, Prateek Peres-da-Silva gave me blatant elevator eyes and asked if I had ever dressed for a client meeting before, in anticipation of the first one for my second project at Knowledgent. I knew he knew I had just been on client site for an 8-week POC, and said, “Yes, but go ahead.” He proceeded to point at my pants and tell me to wear dress pants (I was wearing dress pants); to point at my shirt and tell me to wear a button-up (I was wearing a 100% silk button-up); to point at my shoes and tell me to “wear shoes like (woman at the company who’s petite and nontrivially above me in age) wears…. basically, dress like she dresses.” I was wearing all black and black flats (leather, several hundred dollars), which I wear specifically because 1) we’re in NYC, Shail, and 2) I don’t want to deal with garbage like this trying to tell me what to wear when I’m a data scientist with serious work to do. After said meeting, he had car trouble and pulled over on the side of the highway. I asked him if he had a car jack; he said he thought so, took it out, and stared at it. I asked if he knew how to jack a car. I jacked his car for him. At some point he told me that Ari Yacobi had told him that I was the “only real data scientist at the company.” Additionally, as I mentioned to you, he said that in your office he announced to a group of people that the “marquee” client’s daughter is “hot.” He expressed concern that you would fire him afterward; however, it is my understanding that you took no action and continuously exhibit tacit approval of behavior like this. At a business lunch with Ari Yacobi and a “data scientist” (in quotes because, as you confirmed with Ari Yacobi after my meeting with you, this particular person does not program and would better serve in a business analyst role), those two were discussing their wives at home with their newborns. After a very long while of not participating in the conversation, I apologized for not having much to contribute as a means to move the conversation along. The “data scientist” eventually proceeded to ask me 1) how old i am, 2) if i want children, 3) if I’ve ever considered freezing my eggs. Ari Yacobi sat silently. I put my fork down and said, sternly, that I would like children and that is not something I would like to discuss at lunch. I have screenshots of glowing praise from this “data scientist” to me on Slack, both before and after this incident. When I reported these things to your head of HR, Kerry Ann MacIsaac, after you left the phone call, she asked me, 1) why I did not report these things to her sooner; 2) in which month Prateek Peres-da-Silva scrutinized my outfit, since she had apparently done a dress code work shop a few months earlier (there is no dress code my outfit would have broken unless it were, “Do not wear all black”); 3) could not answer whether, on average, she thinks that men or women are sexually harassed more often in the United States, 4) why, if I thought Knowledgent did not take data science seriously, did I come and work there. I explained that a very competent female data scientist had interviewed me (whom, I have heard from Ari Yacobi, you lost under duress), and that Ari Yacobi willfully misrepresents his capacity as a data scientist. In the phone conversation — in which I quit effective immediately because you insisted on keeping Ari Yacobi in his current position even though you did not argue when I said he’s a fraud — you even stooped to argue that, “‘data science’ means something different at Knowledgent.” You should probably share this alternative definition on the company website because right now Knowledgent makes pretty bold claims about how seriously it takes data science under the standard definition. Lastly, and this was not discussed in your office, I wanted to bring up your offer of having dinner at my choice of Italian restaurant in Manhattan (your terms) with you, Tom Johnstone, and Ari Yacobi if I could pull together a D3.js visualization within a set period of time. All four of us were present in the room for that offer. I’m curious why you thought I would somehow be interested in spending time with you and two married men outside of working hours instead of, for example, getting a bonus or a promotion or perhaps even a raise to meaningfully acknowledge that and all the other work I was doing for you. Would you have made the same offer to a hard-working male? Something tells me you would have thought a bonus were more appropriate. Maybe I’m wrong. Anyone at Ari Yacobi’s level or under would have throttled me in the upcoming performance review — what’s that? The Tableau guy who can’t program? The one with whom you said another data scientist also cancelled all monthly calls because they’re pointless and another demonstration of an ineffective and tortuous reporting system? The one who said if he needs “serious data science” done, he goes to the one you and I discussed can’t program? Please. — so I came straight to you with the concerns bulleted above. We know how that turned out. I think enough of my sentiment is implicit in what’s written above. I’ll end by saying that this industry, or any industry, does not have space for the kind of behavior you’ve been enabling. Best, REV PS — Several people reached out to me since I quit, and it would be remiss of me not to say how much I appreciate them and miss them. You guys are great. Thank you. Written on February 26, 2018

Open Letter to Shail Jain, Knowledgent CEO

26

open-letter-to-shail-jain-knowledgent-ceo-17c04f88a88d

2018-05-23

2018-05-23 18:36:46

https://medium.com/s/story/open-letter-to-shail-jain-knowledgent-ceo-17c04f88a88d

false

2,550

null

Data Science

data-science

Data Science

33,617

Rebecca V

null

98cae8a8e0ca

rebecca.e.vitale

2

1

20,181,104

null

0

null

0

null

2018-03-01

2018-03-01 17:12:22

2018-03-01

2018-03-01 17:41:26

3

false

en

2018-03-01

2018-03-01 17:41:26

2

17c126e7f46d

3.516038

10

3

0

We marketers have short memories.

4

Your Chatbot is Making Me Bounce! We marketers have short memories. Back in 2004, pop-up ads were the most hated website user experience. Then, along came ad blockers to the rescue. Through the years since, clever marketers have come up with novel ways to thwart ad-blocking technology, like native advertising that “pretends” to be helpful content or pop-up lead conversion forms that masquerade as “valuable offers”. The idea is that being helpful trumps being intrusive every time and that people will voluntarily turn off their ad blockers because your content is so valuable. Blocking ads isn’t just a desktop or browser thing. Facebook has had a long and complex history of allowing users to block ads, even though advertising is what fills the company’s coffers. Philosophically, even though Facebook and other media sites are technically “spying on you”, they counter that by serving up relevant ads, they’re making life easier for you, and, oh yeah, it’s in Facebook’s Terms and Conditions. Raise your hand if you enjoy pop-ups and obviously sales-related native ads in the middle of your website, mobile and social media user experience. That’s what I thought. Nada. Zero hands. So why do we continue this barrage of pop-up offers, ads and other annoying tactics? Because that’s our job, many will argue. We get paid to generate sales leads, and if we have to bombard people with our stuff (and call it “inbound”) to get lead conversions, so be it. If there’s a new trick that helps us double or triple our lead conversion stats, deal me in, right? Which leads me to chatbots Chatbots are cool, because they are AI-powered (sort of). Most use some form of NLP (neuro-linguistic programming) and/or NLG (natural language generation) to have a machine-driven conversation with a website or social site visitor. Some are truly “intelligent” in the sense that they can listen to our questions and responses and formulate a reasonable answer based on machine learning. Most are really just automated responses based on multiple choice questions or data that’s been collected via marketing automation. We’ve all used them, and now practically every site has one. But do they work? Yes and no Let’s start with “no”. Chatbots are the new pop-up. When I visit your website for the first time, I’m looking to explore — to see for myself what you offer and why it might be of value to me. Or maybe I’m reading a blog post I saw in a search or a Facebook post. As marketers would say, I’m at the top of the sales funnel and I’m at Awareness Stage 0 in the buyer journey. Am I ready for a sales call? Hell no! Do I want your chatbot assistant asking me annoying questions before I even have a chance to explore? What do you think? You have caused me to bounce before I read the first word of your page or post. We both lose because you lost a lead conversion opportunity, and I wasted my time visiting your website. So, where’s the “yes” in the chatbot experience? Let’s take the bot off the home page and put it where it belongs, on mid-to-bottom-funnel pages. Product and Service pages, pricing pages, FAQs, Customer Support and Contact Us pages come to mind. These are places where visitors might very well be looking for answers to specific questions, and they might well be willing to give up that all-important email or, better yet, have a conversation with a real human being. I actually really love it if I can get my billing question answered via chat. No, I’m not putting my CC or SSN information in the chat box though! Maybe I have a question about product size or colors, or maybe I’m wondering about compatibility with other products. All fair game! So, what’s in it for marketers? Good question. Since we’re so obsessed with lead generation, why not do it the old-fashioned way? Let’s be the best product-service-company for the needs of the customer and make sure the user/visitor experience is the best it can possibly be. Let’s stop asking them a bunch of sales questions as they peek into the top of the sales funnel and wait until they’re ready for that conversation. Then, let’s make it as easy as possible using a variety of options including chatbots, forms, emails and Heaven-forbid, real conversation with one of our experts. But wait, what about “conversational marketing”? Isn’t that the latest trend and most effective way to build sales pipeline? How about dropping the “marketing” part of that buzz phrase and just go with “conversation”?

Your Chatbot is Making Me Bounce!

124

your-chatbot-is-making-me-bounce-17c126e7f46d

2018-05-15

2018-05-15 13:45:20

https://medium.com/s/story/your-chatbot-is-making-me-bounce-17c126e7f46d

false

786

null

Website

website

Website

11,610

John McTigue

I'm a semi-retired marketing agency owner with a side of biohacker. I'm exploring longevity, fighting disease naturally and passive income these days.

6b6707459426

john_94869

42

63

20,181,104

null

0

null

0

null

2017-12-14

2017-12-14 09:37:34

2017-12-14

2017-12-14 09:38:26

0

false

en

2017-12-14

2017-12-14 09:38:26

2

17c188c163e

1.954717

0

Big Data Analytics adopts high volume of industrial data and helps business organizations to examine the complex volumes of corporate data…

3

How Big Data Analytics is Improving Indian Companies Big Data Analytics adopts high volume of industrial data and helps business organizations to examine the complex volumes of corporate data to obtain precise and actionable data results and insights regarding the latest market predictions and trends, customer and industrial preferences for products and services, market correlations, analysis of revenue and growth data, and many more important business factors. There are a lot of effective solutions which Big Data Technologies is providing to the Indian Business Market and helping organizations leverage the large amount of data to explore new business corners, identify new customers and business deals, streamlined all the business operations, improved the customer experiences and bottom-line operations and hence the customer satisfaction at the end result. To know more about us: http://canopusdatainsights.com/ Better Decision Making: As the Big Data Analytics involves in-memory operations and the speed of processing the data by Hadoop Technology offers companies a rare combination to analyze new data sources from the raw data, analyze information immediately from the data sets and then helping companies to make the better decisions much faster and in an efficient manner. Reduced Data Cost: Big Data Analytics mainly uses Cloud-based technologies and analytics strategies to provide large storage for data volumes and to use the data segments in identifying and targetting the new business in many efficient ways. New Design and Product Services: Big Data Analytics helps organizations to find new product trends to increase the business and create the development roadmap for designing new targets and trends for the business to empower customer needs and products. The Big Data Analytics Techniques enhances the decision-making ability for a company to analyze the company’s data thoroughly and then process it on the basis of the evidence results rather than the market business intuition predictions. These new evidence-based research and results increase the areas for a business in terms of betterment of the revenue and growth. And, it opens the new opportunities and business identifications of many industries. Canopus Data Insights, an Indian Data Analytics Company is a rising tech-leader in Outsourcing the Data Analytics Services, Big Data Services, Data Science-based Products, to the customers and the clients, from small organizations to big enterprises. The company is having an experienced team of data scientists to get the result-oriented and productive insights from the raw and distributed data and transform the data into better business results. The company provides expert services in Data Analytics, Data Gathering, Data Driven Model Creation, Data Quality Management, Data Science, Big Data Services and Data Visualization to many business industries. It also gives quality service for all the mainstreams of Market Analytics Services such as Customer Segmentation, Buzz Monitoring, Churn Analysis, Upsell, and Cross-Sell. About the Company: Canopus Data Insights is an Indian Data Science Company that is skilled in providing Data Science, Big Data, Data Analytics, Predictive Analytics and Machine Learning Services to its customers worldwide. Visit their website www.canopusdatainsights.com or call +91 731–2551963 to know more about their extensive work in Data Analytics and Data Services.

How Big Data Analytics is Improving Indian Companies

0

how-big-data-analytics-is-improving-indian-companies-17c188c163e

2017-12-14

2017-12-14 09:38:27

https://medium.com/s/story/how-big-data-analytics-is-improving-indian-companies-17c188c163e

false

518

null

Big Data

big-data

Big Data

24,602

Canopus Infosystems

Canopus Infosystems is an ISO 9001:2008 Certified Company. Canopus Infosystems is a trusted and best mobile app and website development company in India.

9112323b59f8

ankit.jain_86719

7

1

20,181,104

null

0

null

0

aae30e806935

2018-08-04

2018-08-04 16:22:25

2018-08-11

2018-08-11 20:10:02

1

false

pt

2018-08-28

2018-08-28 17:36:39

3

17c2da304708

1.822642

0

Quando fazemos alguns gráficos podemos deparar com um (ou alguns) pontos bem afastados do geral, os famosos outliers. Esses pontos podem…

5

Detectando outliers nos modelos estatísticos Quando fazemos alguns gráficos podemos deparar com um (ou alguns) pontos bem afastados do geral, os famosos outliers. Esses pontos podem interferir nos valores da sua análise… e aí vem a pergunta: devo ou não retirar este(s) ponto(s)? O debate sobre a remoção (ou não) de outliers é amplo. Então fica a seu critério, pois não existe resposta certa ou errada. Métodos mais “modernos” de análises, como o machine learning, entretanto, precisam passar por análises de outliers e pela remoção desses pontos obrigatoriamente. Nós veremos dois diferentes métodos para detectar os outliers no R. Ambas as funções utilizadas aqui para estes testes são distribuídas por padrão junto com o R, então não será necessário instalar nenhum tipo de pacote. Cook’s distance A Distância de Cook verifica a influência de uma observação na análise de regressão utilizando mínimos quadrados. Após os cálculo dos valores de influencia de cada um dos pontos, aqueles que apresentarem um maior valor poderão ser removidos. Vamos criar algumas varáveis para a construção do modelo e testar a Distância de Cook com o comando cooks.distance: De acordo com as distâncias calculadas, o ponto 10 poderá ser removido para a construção de um novo modelo, pois uma das recomendações é remover os pontos com a distância > 1 ou >4*n. Podemos colocar esses valores em um gráfico para visualizarmos melhor: Plotando os valores da Distância de Cook. Bonferroni Outlier Test No caso do Teste de Bonferroni, ele apresenta um p-valor para a tomada de decisão. No caso dos modelos lineares generalizados (GLM), o maior resíduo sob a distribuição t é utilizado. Nesse caso, deveremos utilizar o argumento cars::outlierTest, indicando que a função para o teste deve ser retirada do pacote cars: O resultado deste teste também apontou o ponto 10 como um outlier e poderá ser removido. Ambos os testes utilizados apontaram para uma mesma observação como outlier, mas isso pode variar com os seus dados. Mas qual teste é melhor utilizar? Você deve analisar como são realizados os cálculos e os pressupostos dos testes para escolher o que acha melhor. Uma vez escolhido o teste, continue com ele até o final. É possível acontecer de você retirar um outlier do modelo e, após rodar o teste novamente, ele apontar outro ponto que não apareceu em um primeiro momento. Isso acontece porque o cálculo é realizado de acordo com o modelo construído. Um novo modelo (sem um ponto) implicará em novos valores de influência. Visite nosso site ou mande um e-mail para viniciusbrbio@gmail.com. Você também pode me encontrar no Twitter. Se preferir, poderá adicionar o feed do blog.

Detectando outliers nos modelos estatísticos

0

detectando-outliers-nos-modelos-17c2da304708

2018-08-28

2018-08-28 17:36:39

https://medium.com/s/story/detectando-outliers-nos-modelos-17c2da304708

false

430

Bioestatística e Data Science

null

bio-data-blog

bio-d@protonmail.com

bio-data-blog

R,ESTATÍSTICAS,DATA SCIENCE,CIENCIA DE DADOS,ECOLOGIA

viniciusbrbio

R

r

R

1,558

Vinícius Rodrigues

null

cf78c91e2cb

viniciusbrbio

15

34

20,181,104

null

0

null

0

null

2018-03-24

2018-03-24 22:51:06

2018-03-25

2018-03-25 16:11:13

0

false

en

2018-03-30

2018-03-30 17:52:05

1

17c4264349f1

1.833962

1

0

There are already many camera phone applications on the market that utilize AI to do things, such as, image classification, object…

2

PhotoCoco — Automating the Smart Phone Camera There are already many camera phone applications on the market that utilize AI to do things, such as, image classification, object tracking, and filters. AI has very powerful “shaping finding” capabilities given an objective function to minimize. Sometimes, the problem is not fully defined or there is not enough historical data to capture the process or target. It can be difficult to define an objective function in these situations. For example, AI might need sample images of an object to track or an inital starting point of the object. The question that I am trying to answer with PhotoCoco is “When is it the best time to take a photo and can it be automated?” Obviously, this question is broad and needs to be drilled down into at least a particular situation, i.e., object motion detection. PhotoCoco, a camera phone app, utilizes a motion detection system that separates camera and object motion using machine learning and computer vision. Camera motion can be corrected for, maybe due to a shaky hand at high zoom levels, by shifting the frame to maintain a “steady perspective.” The number of pictures taken depends on thresholds for object detection and motion blur. An artificial control system is used to automatically tune these thresholds in real-time to take an average of 1 photo per N seconds. There are benefits to using PhotoCoco over the regular camera phone app. In addition to being able to take photos normally, by pressing the Trigger button located on the UI, the app has an option to take photos at regular intervals. The app guarantees that a picture will not be taken while the camera is focusing. The camera can zoom much farther than the regular camera app. The regular camera app limits the zoom capability of the camera, PhotoCoco does not. PhotoCoco can create clear, stop-frame videos from the captured photos. PhotoCoco’s adjustable features: 1. Focusing 1.1 Adjustable camera focusing intervals 1.2 PhotoCoco automatically focuses on sections of the camera that experience the most change. 2. Video Creation 2.1 Video creation from saved photos 2.2 Adjustable frame rate of saved video 3. Object and Camera motion 3.1 Object motion detection 3.2 Adjustable camera motion blur tolerance 3.3 Adjustable object motion blur tolerance 3.5 Automatically take photos based on object motion detection 4. Artificial Control System 4.1 Set the average photo rate 4.1 Set how aggressively the system will adjust parameters (Photo Charge Speed) 4. Other Features 4.1 Automatic brightness adjustment 4.2 Real-time motion correction 4.3 Trigger photo capture at specified time intervals PhotoCoco works with iPad but with limited functionality. This is mainly due to the low performance of iPad hardware. System Requirements PhotoCoco requires a iOS or iPad device with 10.0 or later installed. Rohan Kotwani — Data Scientist B.S. in Electrical Engineering & M.S. in Analytics For More Information Please email me at rohankotwani@icloud.com.

PhotoCoco — Automating the Smart Phone Camera

1

photococo-automating-the-smart-phone-camera-17c4264349f1

2018-03-30

2018-03-30 17:52:07

https://medium.com/s/story/photococo-automating-the-smart-phone-camera-17c4264349f1

false

486

null

Photography

photography

Photography

98,594

Rohan Kotwani

null

7db535a7549e

rohankotwani

55

3

20,181,104

null

0

null

0

f702855ffe47

2017-10-17

2017-10-17 00:00:21

2017-10-17

2017-10-17 00:00:22

11

false

en

2017-10-17

2017-10-17 00:00:22

12

17c449723639

2.545283

0

null

3

Coleção de Ferramentas da Apache para Big Data e Machine Learning # medium.com Apache Sqoop Apache Hive Apache Airflow Apache Prediction I/O Apache Hadoop Apache Spark Apache Oozie Apache… My Xoogler Startup: Tensorflight # medium.com Image courtesy of Samson Duborg-Rankin Artificial Intelligence For Property Insurance Editor’s note: Fresh o… WTF? Programming the master algorithm for the future we want (podcast) # news.impactalpha.com Good Business In `What’s the Future,’ entrepreneur Tim O’Reilly provides tools for redrawing the maps of our… WHY DO WE NEED ARTIFICIAL INTELLIGENCE IN TALENT ACQUISITION? # medium.com Ahead of the HR Technology Conference, we are seeing a bunch of new releases about artificial intelligence p… AI Biweekly: 10 Bits from Oct (Pt 1) # medium.com October 2nd — Instagram Releases Shoppable Tags Feature in US US businesses using BigCommerce now have acces… Epic adding AI, other capabilities to support clinicians # medium.com Artificial intelligence and improved capabilities designed to incorporate other sources of patient data were… Why does your company need a Facebook bot? # chatbotsmagazine.com Traditional apps are slowly dying. It’s a bold statement, but with the increased size of apps and the hassle… Artificial intelligence: Hype, hope, fear # legalainews.com If my email inbox is anything to go by, a technology revolution is under way that is going to transform all … AI Goes To Law School # legalainews.com Some law schools are beginning to consider what their relationships with AI could look like. Over the last y… Communication is hard (part 1) # medium.com Why it’s so tough to deliver a natural experience on voice assistant platforms He may have lacked emotion an… AI can revolutionize HR, but deploy with care # venturebeat.com GUEST: Artificial intelligence is transforming HR. We’ve already seen a marked rise in the number of HR tool… AI innovation could cause an ethical conundrum for organized religion # venturebeat.com GUEST: Over the past decade or so, artificial intelligence has gone from being a plot point in movies to bei…

12 new things to read in AI

0

12-new-things-to-read-in-ai-17c449723639

2017-10-17

2017-10-17 00:00:23

https://medium.com/s/story/12-new-things-to-read-in-ai-17c449723639

false

330

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