Playing Atari with few neurons: Improving the efficacy of reinforcement learning by decoupling feature extraction and decision making.

We propose a new method for learning compact state representations and policies separately but simultaneously for policy approximation in vision-based applications such as Atari games. Approaches based on deep reinforcement learning typically map pixels directly to actions to enable end-to-end training. Internally, however, the deep neural network bears the responsibility of both extracting useful information and making decisions based on it, two objectives which can be addressed independently. Separating the image processing from the action selection allows for a better understanding of either task individually, as well as potentially finding smaller policy representations which is inherently interesting. Our approach learns state representations using a compact encoder based on two novel algorithms: (i) Increasing Dictionary Vector Quantization builds a dictionary of state representations which grows in size over time, allowing our method to address new observations as they appear in an open-ended online-learning context; and (ii) Direct Residuals Sparse Coding encodes observations in function of the dictionary, aiming for highest information inclusion by disregarding reconstruction error and maximizing code sparsity. As the dictionary size increases, however, the encoder produces increasingly larger inputs for the neural network; this issue is addressed with a new variant of the Exponential Natural Evolution Strategies algorithm which adapts the dimensionality of its probability distribution along the run. We test our system on a selection of Atari games using tiny neural networks of only 6 to 18 neurons (depending on each game's controls). These are still capable of achieving results that are not much worse, and occasionally superior, to the state-of-the-art in direct policy search which uses two orders of magnitude more neurons.

A low-latency, big database system and browser for storage, querying and visualization of 3D genomic data.

Recent releases of genome three-dimensional (3D) structures have the potential to transform our understanding of genomes. Nonetheless, the storage technology and visualization tools need to evolve to offer to the scientific community fast and convenient access to these data. We introduce simultaneously a database system to store and query 3D genomic data (3DBG), and a 3D genome browser to visualize and explore 3D genome structures (3DGB). We benchmark 3DBG against state-of-the-art systems and demonstrate that it is faster than previous solutions, and importantly gracefully scales with the size of data. We also illustrate the usefulness of our 3D genome Web browser to explore human genome structures. The 3D genome browser is available at http://3dgb.cs.mcgill.ca/.

Experience using web services for biological sequence analysis.

Programmatic access to data and tools through the web using so-called web services has an important role to play in bioinformatics. In this article, we discuss the most popular approaches based on SOAP/WS-I and REST and describe our, a cross section of the community, experiences with providing and using web services in the context of biological sequence analysis. We briefly review main technological approaches as well as best practice hints that are useful for both users and developers. Finally, syntactic and semantic data integration issues with multiple web services are discussed.