| Time: | November 28, 2024 |
|---|---|
| Download as iCal: |
|
Patricia Pauli, M.Sc.
Institute for Systems Theory and Automatic Control
University of Stuttgart
Stuttgart, Germany
Thursday 2024-11-28 5 p.m.
IST Seminar Room 2.255 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen
Abstract
Despite the numerous successful applications of convolutional neural networks across various fields including image and natural language processing, they are black-box models that are not fully understood and lack robustness guarantees. This limits their use in safety-critical applications such as autonomous driving. In this talk, I address this problem by leveraging well-established concepts and principles from control theory to derive and enforce robustness guarantees for convolutional neural networks. One key measure for robustness of a neural network is its Lipschitz constant, yet finding the Lipschitz constant of a neural network is an NP-hard problem. Therefore, we are interested in (i) obtaining accurate upper bounds on the Lipschitz constant for convolutional neural networks and in (ii) designing convolutional neural networks with user enforced Lipschitz bounds, i.e., with robustness guarantees. In doing so, we exploit properties of the underlying building blocks of neural networks, for instance the fact that nonlinear activation functions are slope-restricted and that convolutions can be represented in state space, to develop semidefinite programming-based methods for the robustness analysis and robust design of convolutional neural networks.
Biographical Information
Patricia Pauli received master’s degrees in Mechanical Engineering and Computational Engineering from the Technical University of Darmstadt, Germany, in 2019. During her M.Sc., she spent one year at the University of California, Berkeley and she carried out her research for her master’s thesis at Massachusetts Institute of Technology under the supervision of Dr. Anuradha Annaswamy. Since 2019, Patricia has been a Ph.D. student with the Institute for Systems Theory and Automatic Control under supervision of Prof. Frank Allgöwer and a member of the International Max-Planck Research School for Intelligent Systems (IMPRS-IS). In 2022, she spent a research stay at the University of Sydney hosted by Ian Manchester. Her research interests are at the intersection of control theory and machine learning. She is especially interested in robust neural networks and learning-based control.