Yang Cai speaking at theory lunch this week
11 views
Skip to first unread message
Haipeng Luo
May 5, 2025, 12:34:02 PMMay 5
to usc-theo...@googlegroups.com, usc-t...@googlegroups.com
Hi all,
Yang Cai from Yale is visiting us on Thursday and will be speaking at an additional theory lunch (see talk info below). His schedule is very flexible currently. If you'd like to meet with him, please send me your availability.
Thanks,
Haipeng
=======
Title: Learning and Games: A New Frontier in Algorithmic Game Theory
Abstract: Over the past decade, machine learning has achieved remarkable progress across a wide range of applications. This success is largely attributable to the paradigm of training machine learning systems by minimizing a single loss function using gradient-descent-based optimization algorithms. However, the landscape is shifting, and many of the outstanding challenges in machine learning now lie at its intersection with game theory. These challenges arise either from explicitly strategic environments—such as multi-agent reinforcement learning systems, autobidding in auctions, and algorithmic pricing—or from problems that can be implicitly modeled as games, such as robustifying models against adversarial attacks, training generative models, and alignment.
These emerging problems present new challenges for both machine learning and game theory. On the one hand, it remains unclear what types of optimization algorithms are effective in these game-theoretic settings. On the other hand, the games arising from machine learning problems often involve non-concave utilities, whereas traditional game theory has largely focused on games with (quasi)-concave utilities. In this talk, we will discuss recent progress in tackling these challenges and explore the obstacles and opportunities that lie ahead for both machine learning and game theory.
Abstract: Over the past decade, machine learning has achieved remarkable progress across a wide range of applications. This success is largely attributable to the paradigm of training machine learning systems by minimizing a single loss function using gradient-descent-based optimization algorithms. However, the landscape is shifting, and many of the outstanding challenges in machine learning now lie at its intersection with game theory. These challenges arise either from explicitly strategic environments—such as multi-agent reinforcement learning systems, autobidding in auctions, and algorithmic pricing—or from problems that can be implicitly modeled as games, such as robustifying models against adversarial attacks, training generative models, and alignment.
These emerging problems present new challenges for both machine learning and game theory. On the one hand, it remains unclear what types of optimization algorithms are effective in these game-theoretic settings. On the other hand, the games arising from machine learning problems often involve non-concave utilities, whereas traditional game theory has largely focused on games with (quasi)-concave utilities. In this talk, we will discuss recent progress in tackling these challenges and explore the obstacles and opportunities that lie ahead for both machine learning and game theory.
Reply all
Reply to author
Forward
0 new messages