Timezone: »

Infer to Control: Probabilistic Reinforcement Learning and Structured Control
Leslie Kaelbling · Martin Riedmiller · Marc Toussaint · Igor Mordatch · Roy Fox · Tuomas Haarnoja

Sat Dec 08 05:00 AM -- 03:30 PM (PST) @ Room 516 CDE
Event URL: https://sites.google.com/view/infer2control-nips2018 »

Reinforcement learning and imitation learning are effective paradigms for learning controllers of dynamical systems from experience. These fields have been empowered by recent success in deep learning of differentiable parametric models, allowing end-to-end training of highly nonlinear controllers that encompass perception, memory, prediction, and decision making. The aptitude of these models to represent latent dynamics, high-level goals, and long-term outcomes is unfortunately curbed by the poor sample complexity of many current algorithms for learning these models from experience.

Probabilistic reinforcement learning and inference of control structure are emerging as promising approaches for avoiding prohibitive amounts of controller–system interactions. These methods leverage informative priors on useful behavior, as well as controller structure such as hierarchy and modularity, as useful inductive biases that reduce the effective size of policy search space and shape the optimization landscape. Intrinsic and self-supervised signals can further guide the training process of distinct internal components — such as perceptual embeddings, predictive models, exploration policies, and inter-agent communication — to break down the hard holistic problem of control into more efficiently learnable parts.

Effective inference methods are crucial for probabilistic approaches to reinforcement learning and structured control. Approximate control and model-free reinforcement learning exploit latent system structure and priors on policy structure, that are not directly evident in the controller–system interactions, and must be inferred by the learning algorithm. The growing interest of the reinforcement learning and optimal control community in the application of inference methods is synchronized well with the development by the probabilistic learning community of powerful inference techniques, such as probabilistic programming, variational inference, Gaussian processes, and nonparametric regression.

This workshop is a venue for the inference and reinforcement learning communities to come together in discussing recent advances, developing insights, and future potential in inference methods and their application to probabilistic reinforcement learning and structured control. The goal of this workshop is to catalyze tighter collaboration within and between the communities, that will be leveraged in upcoming years to rise to the challenges of real-world control problems.

=== Intel AI is proud to sponsor Infer2Control @ NeurIPS 2018 ===
Early detection of tumors. Predicting equipment failures before they happen. Having a natural conversation with your home or car. Making retail more personal than ever. This is Artificial Intelligence powered by Intel, and companies around the globe are using it to make money, save money, and advance the future of their industry. At Intel, we’re using decades of expertise in silicon, software, communications, memory and storage to create the new technologies that AI demands. Technologies that break barriers between data center and edge, server and network, training and inference, model and reality – maximizing the economics of AI to take data from theory to real-world success. Learn more: ai.intel.com

Author Information

Leslie Kaelbling (MIT)
Martin Riedmiller (DeepMind)
Marc Toussaint (Universty Stuttgart)
Igor Mordatch (University of Washington)
Roy Fox (UC Berkeley)
Roy Fox

[Roy Fox](royf.org) is an Assistant Professor and director of the Intelligent Dynamics Lab at the Department of Computer Science at UCI. His research interests include theory and applications of reinforcement learning, algorithmic game theory, information theory, and robotics. His current research focuses on structure, exploration, and optimization in deep reinforcement learning and imitation learning of virtual and physical agents and multi-agent systems. He was previously a postdoc at UC Berkeley, where he developed algorithms and systems that interact with humans to learn structured control policies for robotics and program synthesis.

Tuomas Haarnoja (UC Berkeley)

More from the Same Authors