Title: Sampling-based Dynamic Optimization: Theory, Analysis and Applications

Date: Dec 5th , 2022

Time: 4 - 5 PM EST

Meeting Link:

https://gatech.zoom.us/j/98702113478?pwd=cGlpVGwwMnlnQ284TUFGMysrUnRLUT09

Ziyi Wang

Machine Learning PhD Student

School of Aerospace Engineering
Georgia Institute of Technology

Committee

1 Dr. Evangelos Theodorou (Advisor)

2 Dr. Arkadi Nemirovski

3 Dr. Enlu Zhou

4 Dr. Justin Romberg

5 Dr. Ye Zhao

Abstract

This thesis focuses on sampling-based optimization for dynamical systems. We systematically investigate three main perspectives on sampling-based dynamic optimization, namely Stochastic Search, Variational Inference and Variational Optimization. We compare between the perspectives and against state-of-the-art sampling-based dynamic optimizers. A unified analysis on the convergence and sampling complexity of the perspectives is then provided along with numerical examples. We also apply these perspectives in different scenarios. Starting with standard stochastic MPC and risk sensitive optimization with CVaR, we then move on to more complex dynamical systems like the jump diffusion process and opinion dynamics. Finally, a general distributed optimization framework is provided for scaling sampling-based dynamic optimizers for multi-agent control using the consensus ADMM algorithm.