Workshop proposal: Deep Learning in Computational Cognitive Science

A new generation of deep neural network architectures has driven rapid advances in AI over the last ten years. These architectures include convolutional neural networks (CNNs), recurrent neural networks (RNNs), and many variants and extensions. Computational cognitive scientists and neuroscientists have now begun to explore these techniques, and how they might combine with other computational tools such as Bayesian models, symbolic grammars and rule-systems, probabilistic programs, and reinforcement learning. The goal of this workshop is to bring together some of the leading researchers working at this interface, for short talks and an integrative discussion of open questions and promising directions. Talks will cover many areas of cognition including perception, problem-solving and planning, decision-making, language and social cognition. The focus will be on models of human behavior, but the potential bridge to neural studies in humans (via fMRI) and animals (via physiology) will also be explored. Most talks will assume only a basic familiarity with neural networks, and so should be accessible to all CogSci attendees. We hope to be scheduled for an afternoon slot, and have coordinated our plans with the DeepMind’s Deep Learning tutorial proposed for the morning which could serve as an introduction to more advanced methods that several talks will build on.