Farzan Farnia

Assistant Professor of Computer Science and Engineering

University of Hong Kong

Title:

Reliable and Concise Interpretation of Deep Neural Networks

Time: Wednesday, May 3rd 2023, 15:15 – 17:00

Abstract:

Explaining the predictions of deep neural networks has been a topic of great interest to the machine learning community. This talk will focus on gradient-based saliency maps for interpreting neural network models. We will discuss the fidelity and robustness of standard gradient-based interpretation schemes. Also, we explore interpretation schemes that result in sparse and structured saliency maps. As our contribution, we introduce MoreauGrad as an interpretation scheme based on a neural network’s Moreau envelope. We prove the certifiable robustness of MoreauGrad to norm-bounded input perturbations and propose a sparse version of MoreauGrad by applying L1-norm regularization to its dual formulation. We numerically examine the robustness properties of gradient-based saliency maps and display the visual results of their application to standard image datasets.

Biography:

Farzan Farnia is an Assistant Professor of Computer Science and Engineering at The Chinese University of Hong Kong. Prior to joining CUHK, he was a postdoctoral research associate at the Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, from 2019-2021. He received his master’s and Ph.D. in electrical engineering from Stanford University and his bachelor’s degrees in electrical engineering and mathematics from the Sharif University of Technology. At Stanford, he was a graduate research assistant at the Information Systems Laboratory advised by David Tse. Farzan’s research interests span statistical learning theory, information theory, and convex optimization.