ELLIS Delft Talk – Learning for Control: An Inverse Optimization Approach

Speaker: Peyman Mohajerin Esfahani

Abstract: In data-driven inverse optimization an observer aims to learn the preferences of an agent who optimizes an unknown objective function to make decisions. Inspired by recent developments in inverse optimization, we develop a tractable learning algorithm to replicate the behavior of a complex nonlinear feedback control law. The power of the proposed method is showcased in learning the popular model predictive control law. We further discuss how the control design can be robustified using the techniques from the emerging field of distributionally robust optimization.

To join this event, please contact Frans Oliehoek.

Networking session for Machine Learning @ TU Delft

As part of the AI Townhall series, the ELLIS Delft Talk by Peyman Mohajerin Esfahani will be followed by an online networking session for Machine Learning for TU Delft colleagues. To join this session, please sign up via here.