CSE DSI Machine Learning Seminar with Mohammad Ali Maddah-Ali (ECE, UMN)

Coding Theory Helps Reliability, Robustness, and Generalization in Machine Learning

In this talk, we introduce the concept of general coded computing, i.e., processing combinations of data instead of raw data, and discuss how to design optimal codes using principles from learning theory. We then demonstrate how this framework can be applied to tackle three key challenges in large-scale machine learning and advance the state of the art: (i) mitigating the impact of stragglers and adversarial servers in distributed learning, (2) improving both in-distribution and out-of-distribution generalization, and (3) enhancing robustness against adversarial perturbations at the inference time.

Mohammad Ali Maddah-Ali is an Associate Professor at the University of Minnesota Twin Cities. He received his Ph.D. from the University of Waterloo, Canada. He then held a postdoctoral fellowship at the University of California, Berkeley, from 2008 to 2010. From September 2010 to September 2020, he worked as Research Scientist at Nokia Bell Labs in New Jersey. His research interests include information theory, machine learning, and blockchain networks.

Dr. Maddah-Ali has received several prestigious honors, including the IEEE Communications Society and IEEE Information Theory Society Joint Paper Award (2015), and the IEEE Information Theory Society Paper Award (2016). He served as an Associate Editor for the IEEE Transactions on Information Theory (2019–2022) and as Lead Editor for the IEEE Journal on Selected Areas in Information Theory. He is currently a Distinguished Lecturer of the IEEE Information Theory Society and a Fellow of the IEEE.