Learning and approximation algorithms for data-driven multi-stage robust decision-making

Yassine Nemmour (1), Diego A. Agudelo-España (1), Jia-Jie Zhu (2)

(1) Max Planck Institute for Intelligent Systems, Tübingen; (2) Weierstrass Institute, Berlin

Tuesday, November 16, 2021, 11:00

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany

In this talk, we will first present two recent works in 1) constraint tightening for stochastic MPC, where we robustify against distribution shifts using the maximum mean discrepancy; and 2) learning uncertainty-aware dynamics using large-scale kernel machines, and computing worst-case costs via Pontryagin’s minimum principle and the Frank-Wolfe algorithm. Finally, the talk will end with a discussion on multi-stage data-driven optimization and topics beyond the classical notion of robustness, highlighting what we already know, and what we do not yet know.

Meeting-ID: 627 9173 7415
Password: syscop2021