Direct contextual policy search methods learn to improve policy parameters and simultaneously generalize these parameters to different context or task variables. However, learning from high-dimensional context variables, such as camera images, is still a prominent problem in many real-world tasks. A naive application of unsupervised dimensionality reduction methods to the context variables, such as principal component analysis, is insufficient as task-relevant input may be ignored. In this paper, we propose a contextual policy search method in the model-based relative entropy stochastic search framework with integrated dimensionality reduction. We learn a model of the reward that is locally quadratic in both the policy parameters and the context variables. Furthermore, we perform supervised linear dimensionality reduction on the context variables by nuclear norm regularization. The experimental results show that the proposed method outperforms naive dimensionality reduction via principal component analysis and a state-of-the-art contextual policy search method.

Recent advances in machine learning have made significant contributions to drug discovery. Deep neural networks in particular have been demonstrated to provide significant boosts in predictive power when inferring the properties and activities of small-molecule compounds. However, the applicability of these techniques has been limited by the requirement for large amounts of training data. In this work, we demonstrate how one-shot learning can be used to significantly lower the amounts of data required to make meaningful predictions in drug discovery applications. We introduce a new architecture, the residual LSTM embedding, that, when combined with graph convolutional neural networks, significantly improves the ability to learn meaningful distance metrics over small-molecules. We open source all models introduced in this work as part of DeepChem, an open-source framework for deep-learning in drug discovery.

Multi-hop inference is necessary for machine learning systems to successfully solve tasks such as Recognising Textual Entailment and Machine Reading. In this work, we demonstrate the effectiveness of adaptive computation for learning the number of inference steps required for examples of different complexity and that learning the correct number of inference steps is difficult. We introduce the first model involving Adaptive Computation Time which provides a small performance benefit on top of a similar model without an adaptive component as well as enabling considerable insight into the reasoning process of the model.

The O'Reilly Artificial Intelligence Conference call for speakers is open Underneath all the AI hype, real breakthroughs are happening--and obstacles to applied AI are being overcome--allowing AI developers to create software that doesn't just do what it's told, but has the ability to anticipate the needs of its users through a combination of pattern recognition, knowledge, planning, and reasoning. Enterprise bots are emerging to participate in conversations and carry out repetitive tasks. Deep learning toolkits are becoming essential tools for software engineers and data scientists. Frameworks are being developed that promise point-and-click development of intelligent conversational interfaces to relatively unsophisticated developers. There is a growing--and urgent--need for information on applied AI, as opposed to the kind of research presented at academic conferences.

It's all well and good to ask if androids dream of electric sheep, but science fact has evolved to a point where it's beginning to coincide with science fiction. No, we don't have autonomous androids struggling with existential crises -- yet -- but we are getting ever closer to what people tend to call "artificial intelligence." Machine Learning is a sub-set of artificial intelligence where computer algorithms are used to autonomously learn from data and information. In machine learning computers don't have to be explicitly programmed but can change and improve their algorithms by themselves. Today, machine learning algorithms enable computers to communicate with humans, autonomously drive cars, write and publish sport match reports, and find terrorist suspects.

Artificial Intelligence (AI) is making progress in great strides, or at least it appears so! Almost no week passes by without some major announcements of new challenges solved by AI technology or new products powered by AI. Indeed many quantifiable factors attest an unprecedented level of activity: capital investments, number of academic papers, number of products involving AI technology, they all are on a steep rise in the past five years. Computers are already very capable at some specialized tasks that require reasoning and other abilities that we typically associate with intelligence. For example, computers can play a decent game of chess or can help us order our holiday photos. Despite this genuine progress, we are still a long way from human level intelligence because our best artificial intelligence systems are not general purpose. They cannot quickly adapt to novel tasks the way most humans can do.

For a while (at least several months since many people began to implement it with Python and/or Theano, PyLearn2 or something like that), nearly I've given up practicing Deep Learning with R and I've felt I was left alone much further away from advanced technology… But now we have a great masterpiece: {h2o}, an implementation of H2O framework in R. I believe {h2o} is the easiest way of applying Deep Learning technique to our own datasets because we don't have to even write any code scripts but only to specify some of its parameters. That is, using {h2o} we are free from complicated codes; we can only focus on its underlying essences and theories. With using {h2o} on R, in principle we can implement "Deep Belief Net", that is the original version of Deep Learning*1. I know it's already not the state-of-the-art style of Deep Learning, but it must be helpful for understanding how Deep Learning works on actual datasets. Please remember a previous post of this blog that argues about how decision boundaries tell us how each classifier works in terms of overfitting or generalization, if you already read this blog.

Some of the smartest minds in the UK are being lured away from their research positions at Oxford and Cambridge by DeepMind -- a London-based AI lab that was acquired by Google for £400 million in 2014. More than a dozen AI researchers have left the academic powerhouses over the last couple of years for what are likely to be better-paid roles at DeepMind, according to LinkedIn. Steven Cave, the director of Cambridge University's new Centre for the Future of Intelligence, believes that the exodus of talent from academia to corporates is something of a problem. "The best people are being offered huge sums of money to go and work at these tech companies," Cave told Business Insider in Cambridge last week. "You find that you're talking to someone and they're expressing a great deal of interest in a research project and then they're snapped up. We understand that ambitious young people want to work at these big name companies and earn lots of money and that's fine. But at the same time we hope that there will be enough bright young things who are motivated by the intellectual challenge of the issues we're working on and by the sense of wanting to do something good that makes a difference for the world."

We recently launched SpaceNet on AWS, an open corpus of training data established with the goal of enabling advancements in machine learning using satellite imagery. To accelerate this initiative, we're thrilled to announce The SpaceNet Challenge in collaboration with CosmiQ Works and NVIDIA, which is being facilitated by Topcoder. This is the first in a series of recurring open innovation competitions focused on developing next generation computer vision algorithms for automated mapping. With $34,500 of prizes, the first challenge is to tackle the automated extraction of 2D building footprints from imagery. The competition officially starts on 11/14, but you can pre-register today.

When it comes to artificial intelligence (AI), perhaps very few people can claim they fathered a huge part of it. One such man is Jürgen Schmidhuber. Schmidhuber is considered the"father of very deep learning," and the pioneer of deep learning neural networks. In fact, he built the foundations for many of the AI systems we find in our smartphones today. If anyone can predict how far AI will go in the next couple years, it's him. During a talk at WIRED2016, Schmidhuber presented the future of AI as something beyond just taking over jobs.