Data-driven Sequential Decision Making: Reinforcement Learning and Optimization

Friday, March 29, 2019, 11:00 am - 12:00 pm PDTiCal
1016 (10th floor classroom on eastside)
This event is open to the public.
AI Seminar
Yuandong Tian, Facebook
Video Recording:

Modeling sequential decision making in complicated environment is the underlying thread for many research directions. For example, along a decision trajectory, Reinforcement Learning focuses on the cumulative rewards, while optimization in non-convex landscape focuses on the quality of its final destination, and how fast the destination is achieved. More importantly, such a decision making process can be learned and improved from the past experience, with the help of powerful pattern-matching deep neural network models. This talk introduces multiple works along this broad direction: the OpenGo project that reproduces AlphaZero using 2000 GPUs for 9 days training and beats 4 top-30 professional players with 20-0, RL agents for real-time strategy games learned with the help of natural language, and the data-driven optimization procedures that achieve theoretical guarantees in non-convex and good performance in combinatorial problems.

Yuandong Tian is a Research Scientist and Manager in Facebook AI Research, working on deep reinforcement learning and its applications in games, and theoretical analysis of deep models. He is the lead scientist and engineer for ELF OpenGo and DarkForest Go project. Prior to that, he was a researcher and engineer in Google Self-driving Car team in 2013-2014. He received Ph.D in Robotics Institute, Carnegie Mellon University on 2013, Bachelor and Master degree of Computer Science in Shanghai Jiao Tong University. He is the recipient of 2013 ICCV Marr Prize Honorable Mentions.

« Return to Upcoming Events