Title: Stochastic Optimization for Dynamical Systems

Date: 1/11/2022

Time: 8:00 AM EST

Location: Montgomery Knight Building, MK 317

BlueJeans: https://bluejeans.com/291682357/6219

Ziyi Wang

Machine Learning PhD Student

School of Aerospace Engineering
Georgia Institute of Technology

Committee

1. Evangelos Theodorou (Advisor)

2. Arkadi Nemirovski

3. Enlu Zhou

Abstract

Sampling-based dynamic optimization methods have seen many different applications in recent years. Algorithms of this type relies on a control policy distribution to generate control trajectory realizations, which are used to propagate the system states. The cost of each state-control trajectory is then evaluated and used to update the control policy distribution. This proposal presents three different perspectives for deriving sampling-based dynamic optimizers, namely stochastic search, variational inference and variational optimization. Each perspective provides its unique benefits in algorithmic development. Three different future research directions are proposed to improve the applicability, performance, and robustness of the framework. On the applicability side, future research involves application of sampling-based algorithms to quantum systems and social models described by Hawkes processes. On the performance and robustness side, the proposed research is on incorporating the robust covariance steering framework and fractional Gaussian noise for improved exploration.