Presented on Thursday, June 27th, 2024, 10:30 AM, room B220
Speaker
Ido Greenberg (Technion)
Title
Real-World AI: Risk and Robustness in Reinforcement Learning and Kalman Filtering
Abstract:
Real-world applications of reinforcement learning (RL) are often sensitive to risks and uncertainties. Robustness to risk and uncertainty is often addressed by optimization of a risk measure of the returns, instead of their expectation. This mere change of objective raises several surprising challenges: data efficiency is gravely compromised; the optimizer may become entirely blind to successful experience; and the policy gradients become biased. In the first part of the talk, we will discuss these challenges and propose solutions.
Another type of uncertainty that requires robustness is unrealistic modeling assumptions. In the second part of the talk, we will discuss the implications of such model-misspecification in the field of Kalman filtering (KF). In particular, we will demonstrate that common KF methods are highly sensitive to assumption-violations, and will propose a simple, straight-forward solution to enhance the robustness to such violations.
Bio: Ido Greenberg is a PhD candidate at the Technion and a research intern at Nvidia. His PhD focuses on making the extraordinary achievements in the RL literature more applicable to real-world problems. His research includes fundamental works about risk-aversion and awareness in RL, as well as practical applications for conversational planning (at Google) and NP-hard routing problems (at Nvidia). His research interests also include medical and biological forecasting and filtering.
Relevant links:
You can watch previous talks in Panopto and on our RU-vid channel.
Future lectures are published on our website and google calendar.
If you want to subscribe for email updates about the next talks or unsubscribe, please use our mailing list.
26 июн 2024
