Back-propagation is the method by which almost all neural networks are trained today. In a recent interview, Hinton, one of the fathers of Deep Learning, mentioned that he would “throw it all away and start again”. To push materially ahead, entirely new methods will probably have to be invented.

As part of Facebook AI Research (FAIR), this new team will join more than 100 scientists across Menlo Park, New York, and Paris in working to advance the field of artificial intelligence. The Montreal lab will house research scientists and engineers working on a wide range of AI research projects, but it will also have a special focus on reinforcement learning and dialog systems.

Sophia Genetics, a big data analytics company that’s using AI to help medical professionals diagnose and treat patients through genomic analysis, has raised $30 million in a round of funding led by Balderton Capital, with participation from 360 Capital Partners, Invoke Capital, and Alychlo.

An algorithm which accounts for the fact that other agents are learning too, and discovers self-interested yet collaborative strategies like tit-for-tat in the iterated prisoner’s dilemma.

An easy-to-understand introduction to Proximal Policy Optimization (PPO) and extensions to solve MuJoCo and RoboSchool environments. All code is on Github.

The algorithms that power machine learning are still designed manually. This raises a natural question: can we learn these algorithms instead?

To allow the creation of new and useful visualizations, Google announced the release of a set of APIs that allows developers to add custom visualization plugins to TensorBoard. 

Official PyTorch implementations of the models described in Multi-scale context aggregation by dilated convolutions and Dilated Residual Networks. 

TensorFlow 1.3 introduces two important features: Datasets and Estimators. This post shows how they fit in the TensorFlow architecture.

This dataset contains six million ratings for ten thousand most popular (with the most ratings) books from goodreads.com.

A “brief” 200-page introduction. This work aims at providing an introduction to key concepts, algorithms, and theoretical frameworks in machine learning, including supervised and unsupervised learning, statistical learning theory, probabilistic graphical models and approximate inference.

A general-purpose neural embedding model that can solve a wide variety of problems: labeling tasks such as text classification, ranking tasks such as information retrieval/web search, collaborative filtering-based or content-based recommendation, embedding of multi-relational graphs, and learning word, sentence or document level embeddings. In each case the model works by embedding those entities comprised of discrete features and comparing them against each other – learning similarities dependent on the task.

Expected policy gradients (EPG), which unify stochastic policy gradients (SPG) and deterministic policy gradients (DPG) for reinforcement learning. Inspired by expected SARSA, EPG integrates across the action when estimating the gradient, instead of relying only on the action in the sampled trajectory.