Seminar: Bridging Physical Simulation and Deep Learning

Zico Kolter, PhD, Carnegie Mellon University

All dates for this event occur in the past.

Scott Laboratory
Scott Laboratory
Room E525
201 W 19th Ave
Columbus, OH 43210
United States

Abstract

Machine learning techniques, notably those based upon deep learning, have achieved breakthrough success in a number of fields spanning computer vision, speech recognition and natural language processing. However, these methods are typically viewed as "black boxes," oftentimes making it challenging to model phenomena that are subject to more formal constraints, such as those that arise in physical systems.

In this talk, I will highlight some of recent work in bridging the gap between physical simulation and deep learning. Specifically, will include recent work in building a differentiable physics engine, which simulates traditional rigid body dynamics, but does so in a way that allows for exact integration within deep learning pipelines. Highlighted will be the performance of this system on domains such as system identification, physical simulation from visual data and model-based reinforcement learning. Some extensions of this work will also be discussed, including methods for similarly integrating ODE simulation into deep networks. As a whole, this line of work provides an important avenue to connecting "traditional" simulation tools with data-driven learning.

About the speaker

Zico Kolter is an assistant professor in the School of Computer Science at Carnegie Mellon University, and also serves as chief scientist of AI research for the Bosch Center for Artificial Intelligence. His work focuses on the intersection of machine learning and optimization, with a large focus on developing more robust, explainable and rigorous methods in deep learning. In addition, he has worked on a number of application areas, highlighted by work on sustainability and smart energy systems. He is the recipient of the DARPA Young Faculty Award, and best paper awards at KDD, IJCAI and PESGM.

 

Hosted by Professor Jack McNamara.