BIMSA >
BIMSA-Artificial Intelligence and Machine Learning Group Seminar
Matrix and Tensor Computation at the Core of Modern AI: From Deep learning to Adversarial Robustness and Explainability
Matrix and Tensor Computation at the Core of Modern AI: From Deep learning to Adversarial Robustness and Explainability
Organizer
Speaker
Mansoor Rezghi
Time
Friday, May 16, 2025 2:30 PM - 4:00 PM
Venue
A3-1-301
Online
Zoom 537 192 5549
(BIMSA)
Abstract
Matrix computations have traditionally played a central role in solving physics and engineering problems. In recent years, however, this mathematical foundation has shifted toward artificial intelligence, forming the mathematical backbone of many modern machine learning techniques. This talk explores how matrices and tensor methods—higher-order extensions of matrices—serve as the mathematical backbone of AI. Specifically, we examine their applications in building efficient neural networks, such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), enabling significant parameter reduction, and examine how subspace-based representations can design new and efficient adversarial attacks on neural networks and enhance model explainability.
Speaker Intro
Prof. Mansoor Rezghi received his Ph.D. in Applied Mathematics from Tarbiat Modares University in 2009, including a one-year research visit to Linköping University under Prof. Lars Eldén. He served as an Assistant Professor in the Department of Computer Science at Tarbiat Modares University from 2012 to 2018. Since 2018, he has been an Associate Professor at the same institution. His research interests include generative AI, deep learning, tensor and matrix computation, and inverse problems in AI.