Date and time: Oct 24, 2025, 12-1:45pm
Location: Rice 109
Title: Swap Regret and Strategic Learning
Abstract:
In the seminar, I will present Swap Regret and Strategic Learning. The presentation is based on the tutorial at this workshop at EC 2025, which in turn builds primarily on the following two papers:
Yuan Deng, Jon Schneider, Balusubramanian Sivan. Strategizing against No-regret Learners. NeurIPS 2019.
Eshwar Ram Arunachaleswaran, Natalie Collina, Jon Schneider. Pareto-Optimal Algorithms for Learning in Games. EC 2024.
The talk will discuss how learners running standard no-regret algorithms (e.g., gradient descent, multiplicative weights update) can be manipulated by strategic optimizers --- opponents who know the learner’s algorithm and optimize based on it --- and how no-swap-regret algorithms mitigate this vulnerability.
We’ll also see how no-swap-regret algorithms are, in some sense, Pareto-optimal among all learning algorithms.