- This event has passed.
ECE RISE SEMINAR SERIES: Trustworthy AI – Dr. Mahdi Haghifam
March 5 @ 12:00 PM – 1:00 PM
4610 Engineering Hall
The Interplay of Generalization, Memorization, and Privacy in Trustworthy Machine Learning
Abstract:
Machine learning is transforming numerous aspects of modern society, and its increasing use in high-stakes applications necessitates responsible development. In this talk, I will present my research on the foundations and methodologies for building trustworthy ML, focusing on three interconnected challenges: generalization, memorization, and privacy. First, I will explore generalization: how can we ensure that ML models reliably predict on unseen data? I will discuss my work on developing novel information-theoretic measures to characterize and reason about generalization. Next, I will examine data memorization, showing how it can coexist with generalization and may even be necessary for accurate learning. Finally, I will focus on differential privacy, a rigorous framework for mitigating data memorization, and present my work on designing differentially private optimization algorithms. I will conclude by discussing key open questions in the area of trustworthy ML.
Bio:
Mahdi Haghifam is a Distinguished Postdoctoral Researcher at Khoury College of Computer Sciences, Northeastern University, hosted by Jonathan Ullman. He received his PhD from the University of Toronto and the Vector Institute, where he was advised by Daniel M. Roy. Mahdi’s research focuses on the foundations and algorithms of trustworthy machine learning, particularly in the areas of privacy, generalization, and memorization. During his PhD, he worked as a research intern at Google Brain and ServiceNow Research. His contributions have been recognized with a Best Paper Award at ICML 2024 and several fellowships from the University of Toronto.
