This resource is part of a series on specific topics related to data science: regression, clustering, neural networks, deep learning, Hadoop, decision trees, ensembles, correlation, outliers, regression Python, R, Tensorflow, SVM, data reduction, feature selection, experimental design, time series, cross-validation, model fitting, dataviz, and many more. To keep receiving these articles, sign up on DSC. Previous entries are listed below the picture.

For the month of April, we also recognize the most popular posts and blogger based on unique page views (UPV) and social shares. Most Viewed and Most Shared - Platinum Badge ( 20,000 UPV AND 2,000 shares) 10 Free Must-Read Books for Machine Learning and Data Science, by Matthew Mayo Most Viewed - Gold Badges ( 10,000 UPV) Forrester vs Gartner on Data Science Platforms and Machine Learning Solutions, by Gregory Piatetsky Top 20 Recent Research Papers on Machine Learning and Deep Learning, by Thuy Pham Most Viewed - Silver Badges ( 5,000 unique PV) Awesome Deep Learning: Most Cited Deep Learning Papers, by Terry Taewoong Um 5 Machine Learning Projects You Can No Longer Overlook, April, by Matthew Mayo Keep it simple! How to understand Gradient Descent algorithm, by Jahnavi Mahanta Top mistakes data scientists make when dealing with business people, by Karolis Urbonas (new) New Online Data Science Tracks for 2017, by Brendan Martin (new) Cartoon: Machine Learning - What They Think I Do, by Harrison Kinsley Data Science for the Layman (No Math Added), Annalyn Ng and Kenneth Soo Most Shared - Gold Badges ( 1,000 shares) Forrester vs Gartner on Data Science Platforms and Machine Learning Solutions, by Gregory Piatetsky Top 20 Recent Research Papers on Machine Learning and Deep Learning, by Thuy Pham Top mistakes data scientists make when dealing with business people, by Karolis Urbonas (new) Awesome Deep Learning: Most Cited Deep Learning Papers, by Terry Taewoong Um Most Shared - Silver Gold Badges ( 500 shares) Keep it simple! How to understand Gradient Descent algorithm A Brief History of Artificial Intelligence, By Francesco Corea The 42 V's of Big Data and Data Science, by Tom Shafer (new) Deep Stubborn Networks - A Breakthrough Advance Towards Adversarial Machine Intelligence, by Matthew Mayo (new) Awesome Deep Learning: Most Cited Deep Learning Papers, by Terry Taewoong Um 5 Machine Learning Projects You Can No Longer Overlook, April, by Matthew Mayo Keep it simple! How to understand Gradient Descent algorithm A Brief History of Artificial Intelligence, By Francesco Corea The 42 V's of Big Data and Data Science, by Tom Shafer (new) Deep Stubborn Networks - A Breakthrough Advance Towards Adversarial Machine Intelligence, by Matthew Mayo (new)

Many machine learning systems are built to solve the hardest examples of a particular task, which often makes them large and expensive to run--especially with respect to the easier examples, which might require much less computation. For an agent with a limited computational budget, this "one-size-fits-all" approach may result in the agent wasting valuable computation on easy examples, while not spending enough on hard examples. Rather than learning a single, fixed policy for solving all instances of a task, we introduce a metacontroller which learns to optimize a sequence of "imagined" internal simulations over predictive models of the world in order to construct a more informed, and more economical, solution. The metacontroller component is a model-free reinforcement learning agent, which decides both how many iterations of the optimization procedure to run, as well as which model to consult on each iteration. The models (which we call "experts") can be state transition models, action-value functions, or any other mechanism that provides information useful for solving the task, and can be learned on-policy or off-policy in parallel with the metacontroller. When the metacontroller, controller, and experts were trained with "interaction networks" (Battaglia et al., 2016) as expert models, our approach was able to solve a challenging decision-making problem under complex non-linear dynamics. The metacontroller learned to adapt the amount of computation it performed to the difficulty of the task, and learned how to choose which experts to consult by factoring in both their reliability and individual computational resource costs. This allowed the metacontroller to achieve a lower overall cost (task loss plus computational cost) than more traditional fixed policy approaches. These results demonstrate that our approach is a powerful framework for using rich forward models for efficient model-based reinforcement learning. While there have been significant recent advances in deep reinforcement learning (Mnih et al., 2015; Silver et al., 2016) and control (Lillicrap et al., 2015; Levine et al., 2016), most efforts train a network that performs a fixed sequence of computations. Here we introduce an alternative in which an agent uses a metacontroller to choose which, and how many, computations to perform.

The paper presents a novel, principled approach to train recurrent neural networks from the Reservoir Computing family that are robust to missing part of the input features at prediction time. By building on the ensembling properties of Dropout regularization, we propose a methodology, named DropIn, which efficiently trains a neural model as a committee machine of subnetworks, each capable of predicting with a subset of the original input features. We discuss the application of the DropIn methodology in the context of Reservoir Computing models and targeting applications characterized by input sources that are unreliable or prone to be disconnected, such as in pervasive wireless sensor networks and ambient intelligence. We provide an experimental assessment using real-world data from such application domains, showing how the Dropin methodology allows to maintain predictive performances comparable to those of a model without missing features, even when 20\%-50\% of the inputs are not available.

Google has challenged China's top Go player to a series of games against its artificial intelligence technology. It said the software would play a best-of-three match against Ke Jie, among other games against humans in the eastern Chinese city of Wuzhen from 23-27 May. Last year, the Google program recorded a 4-1 victory against one of South Korea's top Go players. One expert said that result had come as a surprise. "A lot of AI researchers have been working on Go because it's the most challenging board game we have," said Calum Chace, author of Surviving AI. "The conventional wisdom was that machines would ultimately triumph but it would take 10 years or so. "The win was a big wake-up call for a lot of people, including many outside the AI community." Google's AlphaGo software was developed by British computer company DeepMind, which was bought by the US search firm in 2014. Its defeat of Lee Se-dol in March 2016 is seen as a landmark moment, similar to that of IBM's Deep Blue AI beating Garry Kasparov at chess in 1997. Several of the moves AlphaGo made defied conventional wisdom but ended up paying off. However, many Go aficionados did not recognise Mr Lee as the world's top player at the time of the contest. So, the new competition against 19-year-old Mr Ke - who is the current number one according to a popular but unofficial player-ranking system - has the potential to bring additional prestige to Google. "We've been hard at work improving AlphaGo to become even more creative, and since playing Lee Se-dol, the program has continued to learn through self-play training," a spokeswoman for DeepMind told the BBC. "We intend to publish more scientific papers in the future, which will include further details of AlphaGo's progress." Google added that Mr Lee would also be invited, but was not sure if he would attend. Over the past year, DeepMind's technology has also been used to find ways to reduce energy bills at Google's data centres as well as to try to improve care in British hospitals. A fresh wave of positive publicity could help Google find further uses for its tech. "If it loses this match, a lot of people will be delighted to claim that Google and DeepMind has overpromised and that this is the kind of hype we always get with AI," commented Mr Chace. "But I wouldn't have thought Google is taking a huge risk.

One of our core aspirations at OpenAI is to develop algorithms and techniques that endow computers with an understanding of our world. It's easy to forget just how much you know about the world: you understand that it is made up of 3D environments, objects that move, collide, interact; people who walk, talk, and think; animals who graze, fly, run, or bark; monitors that display information encoded in language about the weather, who won a basketball game, or what happened in 1970. This tremendous amount of information is out there and to a large extent easily accessible -- either in the physical world of atoms or the digital world of bits. The only tricky part is to develop models and algorithms that can analyze and understand this treasure trove of data. Generative models are one of the most promising approaches towards this goal.

These predictive capabilities of AI (existent across applications involving more than just digital assistants including those for healthcare and autonomous vehicle deployments) are widely facilitated by deep learning algorithms. The general nature of machine learning algorithms is to increase their effectiveness with the more data processed. Deep learning, which also includes neural network algorithms specifically designed to mimic the human brain, encompasses an assortment of data to achieve this functionality; depending on the use case autonomous vehicles and virtual assistants can process data regarding the time, the weather, geography, and personal information about the user.

As artificial intelligence changes the financial services industry, should advisors feel threatened? They see the current wave of technology as an opportunity to provide better service. Much has been made of the research that finds automation will steal work from millions of Canadians, including a University of Toronto study from last year. But it's hard even for experts in the field of artificial intelligence (AI) to see wealth managers entirely replaced by chatbots, or apps for financial planning and investment. Still those involved in the rapidly changing discipline of computer science say the financial industry is fertile ground for increasingly intelligent automation.

One of the biggest confusions about "Artificial Intelligence" is that it is a very vague term. That's because Artificial Intelligence or AI is a term that was coined way back in 1955 with extreme hubris: We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions, and concepts, solve kinds of problems now reserved for humans, and improve themselves. AI is over half a century old and carries with it too much baggage.

ParlAI (pronounced "par-lay") is a framework for dialog AI research, implemented in Python. Over 20 tasks are supported in the first release, including popular datasets such as SQuAD, bAbI tasks, MCTest, WikiQA, WebQuestions, SimpleQuestions, WikiMovies, QACNN & QADailyMail, CBT, BookTest, bAbI Dialog tasks, Ubuntu Dialog, OpenSubtitles, Cornell Movie and VQA-COCO2014. Included are examples of training neural models with PyTorch and Lua Torch, with batch training on GPU or hogwild training on CPUs. Using Theano or Tensorflow instead is also straightforward. Our aim is for the number of tasks and agents that train on them to grow in a community-based way.