Forget skinny soya milk lattes, the best way to get ahead at work is to grab yourself a full-fat mocha. That's according to scientists who studied the behaviour of people drinking coffee and hot chocolate over the course of a year. They found that while coffee boosts your energy levels, hot chocolate helps to relieve anxiety - and combining the two is the best way to improve your attention span. Subjects were given cocoa, cocoa with caffeine, caffeine without cocoa, and a placebo drink with neither caffeine nor cocoa. The study was double blind, meaning both the experimenter and the people taking part in the study were not aware of what drink was given to each participant.

Fujitsu today announced that it has received RIKEN's order for the "Deep learning system," which in terms of operations will be one of the largest-scale supercomputers in Japan specializing in AI research. The RIKEN Center for Advanced Intelligence Project will use the new system, scheduled to go online in April 2017, as a platform to accelerate R&D into AI technology. The system's total theoretical processing performance will reach 4 petaflops(1). The system will be comprised of two server architectures, with 24 of NVIDIA DGX-1 servers and 32 FUJITSU Server PRIMERGY RX2530 M2 servers, along with a high-reliability, high-performance storage system. Fujitsu is leveraging the extensive know-how that it and Fujitsu Laboratories Ltd. have in high-performance computing development and AI research to build and operate one of Japan's most advanced AI research systems.

A team of scientists has developed an artificial intelligence system called DeepStack that recently defeated professional poker players. The team of computing scientists from University of Alberta's Computer Poker Research Group, including researchers from Charles University in Prague and Czech Technical University, said DeepStack bridges the gap between approaches used for games of perfect information with those used for imperfect information games. "Poker has been a longstanding challenge in artificial intelligence," said Michael Bowling from the University of Alberta, Canada, in the paper published in the journal Science. It is the quintessential game of imperfect information in the sense that the players don't have the same information or share the same perspective while they are playing," Bowling added. Imperfect information games are a general mathematical model that describes how decision-makers interact. Artificial intelligence research has a storied history of using parlour games to study these models, but attention has been focused primarily on perfect information games. "We need new AI techniques that can handle cases where decision-makers have different perspectives," Bowling noted. DeepStack extends the ability to think about each situation during play -- which has been famously successful in games like checkers, chess, and Go -- to imperfect information games using a technique called continual re-solving. This allows DeepStack to determine the correct strategy for a particular poker situation without thinking about the entire game by using its "intuition" to evaluate how the game might play out in the near future.We train our system to learn the value of situations," Bowling said.

When should a criminal defendant be required to await trial in jail rather than at home? Software could significantly improve judges' ability to make that call--reducing crime or the number of people stuck waiting in jail. In a new study from the National Bureau of Economic Research, economists and computer scientists trained an algorithm to predict whether defendants were a flight risk from their rap sheet and court records using data from hundreds of thousands of cases in New York City. When tested on over a hundred thousand more cases that it hadn't seen before, the algorithm proved better at predicting what defendants will do after release than judges. Jon Kleinberg, a computer science professor at Cornell involved in the research, says one goal of the project was to show policymakers the potential benefits to society of using machine learning in the criminal justice system.

Peter Shulman, an associate history professor at Case Western Reserve University in Ohio, was lecturing on the reemergence of the Ku Klux Klan in the 1920s when a student asked an odd question: Was President Warren Harding a member of the KKK? He confessed that he was not aware of that allegation, but that Harding had been in favor of anti-lynching legislation, so it seemed unlikely. But then a second student pulled out his phone and announced that yes, Harding had been a Klan member, and so had four other presidents. For most of its history, Google did not answer questions. Users typed in what they were looking for and got a list of web pages that might contain the desired information.

Deep reinforcement learning methods attain super-human performance in a wide range of environments. Such methods are grossly inefficient, often taking orders of magnitudes more data than humans to achieve reasonable performance. We propose Neural Episodic Control: a deep reinforcement learning agent that is able to rapidly assimilate new experiences and act upon them. Our agent uses a semi-tabular representation of the value function: a buffer of past experience containing slowly changing state representations and rapidly updated estimates of the value function. We show across a wide range of environments that our agent learns significantly faster than other state-of-the-art, general purpose deep reinforcement learning agents.

We discuss a Bayesian formulation to coarse-graining (CG) of PDEs where the coefficients (e.g. material parameters) exhibit random, fine scale variability. The direct solution to such problems requires grids that are small enough to resolve this fine scale variability which unavoidably requires the repeated solution of very large systems of algebraic equations. We establish a physically inspired, data-driven coarse-grained model which learns a low- dimensional set of microstructural features that are predictive of the fine-grained model (FG) response. Once learned, those features provide a sharp distribution over the coarse scale effec- tive coefficients of the PDE that are most suitable for prediction of the fine scale model output. This ultimately allows to replace the computationally expensive FG by a generative proba- bilistic model based on evaluating the much cheaper CG several times. Sparsity enforcing pri- ors further increase predictive efficiency and reveal microstructural features that are important in predicting the FG response. Moreover, the model yields probabilistic rather than single-point predictions, which enables the quantification of the unavoidable epistemic uncertainty that is present due to the information loss that occurs during the coarse-graining process.

Deep generative models have been wildly successful at learning coherent latent representations for continuous data such as video and audio. However, generative modeling of discrete data such as arithmetic expressions and molecular structures still poses significant challenges. Crucially, state-of-the-art methods often produce outputs that are not valid. We make the key observation that frequently, discrete data can be represented as a parse tree from a context-free grammar. We propose a variational autoencoder which encodes and decodes directly to and from these parse trees, ensuring the generated outputs are always valid. Surprisingly, we show that not only does our model more often generate valid outputs, it also learns a more coherent latent space in which nearby points decode to similar discrete outputs. We demonstrate the effectiveness of our learned models by showing their improved performance in Bayesian optimization for symbolic regression and molecular synthesis.

In this paper we propose a synergistic melting of neural networks and decision trees (DT) we call neural decision trees (NDT). NDT is an architecture a la decision tree where each splitting node is an independent multilayer perceptron allowing oblique decision functions or arbritrary nonlinear decision function if more than one layer is used. This way, each MLP can be seen as a node of the tree. We then show that with the weight sharing asumption among those units, we end up with a Hashing Neural Network (HNN) which is a multilayer perceptron with sigmoid activation function for the last layer as opposed to the standard softmax. The output units then jointly represent the probability to be in a particular region. The proposed framework allows for global optimization as opposed to greedy in DT and differentiability w.r.t. all parameters and the input, allowing easy integration in any learnable pipeline, for example after CNNs for computer vision tasks. We also demonstrate the modeling power of HNN allowing to learn union of disjoint regions for final clustering or classification making it more general and powerful than standard softmax MLP requiring linear separability thus reducing the need on the inner layer to perform complex data transformations. We finally show experiments for supervised, semi-suppervised and unsupervised tasks and compare results with standard DTs and MLPs.

To improve the efficiency of Monte Carlo estimation, practitioners are turning to biased Markov chain Monte Carlo procedures that trade off asymptotic exactness for computational speed. The reasoning is sound: a reduction in variance due to more rapid sampling can outweigh the bias introduced. However, the inexactness creates new challenges for sampler and parameter selection, since standard measures of sample quality like effective sample size do not account for asymptotic bias. To address these challenges, we introduce a new computable quality measure based on Stein's method that quantifies the maximum discrepancy between sample and target expectations over a large class of test functions. We use our tool to compare exact, biased, and deterministic sample sequences and illustrate applications to hyperparameter selection, convergence rate assessment, and quantifying bias-variance tradeoffs in posterior inference.