Seminar: The Jacob Ziv Communication and Information Theory seminar
Trust-Based Resilience in Distributed Optimization: Conquering Adversarial Agents
Lecturer:
Dr. Michal Yemini
Research Areas:
| How can multi-agent systems learn who to trust—and still reach the right decision—when under attack? The growing deployment of cyber-physical multi-agent systems calls for distributed optimization algorithms that are not only efficient but also resilient to adversarial behavior. This talk presents a suite of recent advances in trust-based resilient distributed optimization and introduces a rigorous framework for robust, scalable learning and optimization in adversarial multi-agent systems. At the heart of this work is the integration of stochastic trust assessments into algorithmic design—empowering agents to learn which peers to trust and adaptively mitigate the impact of malicious agents. We begin with the constrained distributed optimization problem in the presence of adversarial agents and prove convergence to the optimal nominal point for undirected communication graphs, in contrast to state-of-the-art results in resilient consensus. Building on this, we introduce algorithms for unconstrained distributed optimization over directed network topologies, culminating in the Resilient Projected Push-Pull (RP3) algorithm. RP3 provably achieves geometric convergence (in expectation)—even under adversarial interference—through a synergy of gradient tracking, growing constraint sets, and trust-weighted updates. Finally, if time permits, we will discuss the Friedkin-Johnsen model under diminishing stubbornness to understand dynamic consensus behavior in networks, and explore how the introduction of decaying competition reveals fundamental trade-offs in convergence rate and equilibrium behavior. |
| Michal Yemini is an assistant professor (senior lecturer) at Bar-Ilan University, Ramat-Gan, Israel. Prior to that, she was an associate research scholar at Princeton University, a postdoctoral researcher at Stanford University, Stanford, USA, and a visiting postdoctoral researcher at Princeton University. Her main research interests include distributed optimization, sequential decision-making, learning theory, information theory, and percolation theory. She is a co-winner of the Zuckerman Travel and Research STEM Fund at Harvard University. She additionally received the Eric and Wendy Schmidt Postdoctoral Award for Women in Mathematical and Computing Sciences, the Council of Higher Education’s Postdoctoral Fellowships Program for Outstanding Women in Science, and the Bar-Ilan University’s Postdoctoral Fellowship for Women. She obtained her BSc in computer engineering from the Technion-Israel Institute of Technology, Haifa, Israel, in 2011. In 2017 she received her PhD degree in the joint MSc-PhD program from the Faculty of Engineering, Bar-Ilan University, Ramat-Gan, Israel.
|