Mahyar Fazlyab: Control‐Theoretic Tools in Analysis and Synthesis of Neural Network Driven Systems with Performance Guarantees
Abstract: Neural Networks (NNs) have become increasingly effective at many difficult machine-learning and control tasks. However, the nonlinear and large-scale nature of neural networks makes them hard to analyze and, therefore, they are mostly used as black-box models without formal guarantees. This issue becomes even more complicated when NNs are used in learning-enabled closed-loop systems, where a small perturbation can substantially impact the system being controlled. In the first part of this talk, we present a convex optimization framework, inspired by robust control, that can provide useful certificates of stability, safety, and robustness for NN-driven systems. In the second part of the talk, we address the problem of incorporating the safety guarantees of robust control into NN‐driven uncertain dynamical systems. Specifically, we will develop scalable methods that integrate custom projection layers into a neural network‐based policy that enforce robust control specifications (stability, safety, etc) during both training and runtime.
Bio: Mahyar Fazlyab is an Assistant Professor in the Department of Electrical and Computer Engineering at Johns Hopkins University (JHU) since July 2021. Dr. Fazlyab received his Ph.D. in Electrical and Systems Engineering (ESE) from the University of Pennsylvania (UPenn) in 2018. Dr. Fazylab’s research interests are at the intersection of optimization, control, and machine learning. His current research focus is on the safety and stability of learning-enabled autonomous systems.