Lunch will be served at 11:45 AM.
Algorithms are increasingly deployed in large-scale, real-world systems composed of decentralized agents, posing fundamental questions about their behavior. On the one hand, agents may employ sophisticated learning algorithms to interact with each other for their own individual aims, which induce complex long-run dynamics. On the other, algorithms and systems must also be designed to operate well in the presence of these highly strategic and dependent interactions.
In this talk, I will explain how we can develop new mathematical frameworks towards understanding algorithm design and performance in these settings by leveraging their rich structure. In the first part of the talk, I will discuss my recent results on long-run outcomes of decentralized learning dynamics in strategic queuing systems. I will then discuss new insights and results on learning network structure in multi-agent systems from dynamical models of data, which yield realism and computational efficiency beyond the classical statistical framework.
Jason Gaitonde is a Postdoctoral Associate at Massachusetts Institute of Technology, mentored by Elchanan Mossel. He received his Ph.D. in Computer Science from Cornell University in 2023, where he was advised by Éva Tardos. His research interests lie in theoretical computer science, with an emphasis on the interplay between algorithms, game theory, learning theory, and probability.