Recently Google DeepMind program AlphaGo Zero achieved superhuman level without any help - entirely by self-play! Here is the Nature paper explaining technical details (also PDF version: Mastering the Game of Go without Human Knowledge) One of the main reasons for success was the use of a novel form of Reinforcement learning in which AlphaGo learned by playing itself. The system starts with a neural net that does not know anything about Go. It plays millions of games against itself and tuned the neural network to predict next move and the eventual winner of the games. The updated neural network was merged with the Monte Carlo Tree Search algorithm to create a new and stronger version of AlphaGo Zero, and the process resumed.

Recently Google DeepMind program AlphaGo Zero achieved superhuman level without any help - entirely by self-play! Here is the Nature paper explaining technical details (also PDF version: Mastering the Game of Go without Human Knowledge) One of the main reasons for success was the use of a novel form of Reinforcement learning in which AlphaGo learned by playing itself. The system starts with a neural net that does not know anything about Go. It plays millions of games against itself and tuned the neural network to predict next move and the eventual winner of the games. The updated neural network was merged with the Monte Carlo Tree Search algorithm to create a new and stronger version of AlphaGo Zero, and the process resumed.

Hard to believe that it's only been a year since I was doing the previous end-of-year round up. So much has happened in the world of AI that it is hard to fit in a couple of paragraphs. Don't expect too many details, but do expect a lot of links to follow up on them. If I have to pick my main highlight of the year, that has to go to AlphaGo Zero (paper). Not only does this new approach improve in some of the most promising directions (e.g.

Finally for the past few months I have been working on AI for medicine and healthcare. I am also happy to see that the rate of innovation in less "traditional" domains like healthcare is quickly picking up. AI and ML have been applied to medicine with years, starting with expert and Bayesian systems in the 60s and 70s. However, I often find myself citing papers that are only a few months old. Some of the recent innovations presented this year include the use of Deep RL, GANs, or Autoencoders to represent patient phenotypes.

Recently Google DeepMind program AlphaGo Zero achieved superhuman level without any help - entirely by self-play! Here is the Nature paper explaining technical details (also PDF version: Mastering the Game of Go without Human Knowledge) One of the main reasons for success was the use of a novel form of Reinforcement learning in which AlphaGo learned by playing itself. The system starts with a neural net that does not know anything about Go. It plays millions of games against itself and tuned the neural network to predict next move and the eventual winner of the games. The updated neural network was merged with the Monte Carlo Tree Search algorithm to create a new and stronger version of AlphaGo Zero, and the process resumed.

It is hard to answer this question without a proper definition of what "trending" means. Also "deep learning" is being used as a very broad category that even includes methods that are not per se "deep". For example, Adversarial Methods[1] are not necessarily connected to Deep Learning despite the fact that they have been popularized in that context. In any case, I will try to answer the question by looking at recent conferences (e.g.

"In machine learning, is more data always better than better algorithms?" No. There are times when more data helps, there are times when it doesn't. Probably one of the most famous quotes defending the power of data is that of Google's Research Director Peter Norvig claiming that "We don't have better algorithms. We just have more data.". This quote is usually linked to the article on "The Unreasonable Effectiveness of Data", co-authored by Norvig himself (you should probably be able to find the pdf on the web although the original is behind the IEEE paywall).

I won't go into details of how to do this, but in a few words all you need to do is to train your independent models and then use their predictions as features of another ML model that puts them together. This ensemble layer can be as simple as a logistic regression or as complex as a deep Neural Network, but given your question I would definitely encourage you to start with the simplest model possible.

For a product like Quora to succeed, it needs to manage three different aspects well: user engagement, quality of the content, and monetization. The first one should be obvious: a product cannot grow unless users find the content they read and the product itself engaging. The second one, quality, connects directly to our mission to share and grow knowledge. Content needs to have a certain degree of quality to even qualify as knowledge. Not only that, we actually want users to engage in reading and writing good quality content, because that is the only way to scale our knowledge platform.

At Quora, our mission is to "share and grow the world's knowledge". We do this by getting the right questions to the right people, and the existing answers to people who are interested in reading them. We need to build a complex ecosystem of algorithms where we value issues such as content quality, engagement, demand, interests, or reputation. Fortunately, we have lots of very good quality data on which to build machine learning solutions that can help address the previous requirements. In this talk, VP of Engineering Xavier Amatriain will describe some interesting uses of machine learning at Quora that range from different recommendation systems such as personalized ranking of the home feed, to classifiers built to detect duplicate questions or spam.