Work

Pareto-Optimal Algorithms for Learning in Games

Year: 2024

Type: conference-paper

Abstract: We study the problem of characterizing optimal learning algorithms for playing repeated games against an adversary with unknown payoffs. In this problem, the first player (called the learner) commits to a learning algorithm against a second player (called the optimizer), and the optimizer best-responds by choosing the optimal dynamic strategy for their (unknown but well-defined) payoff. Classic learning algorithms (such as no-regret algorithms) provide some counterfactual guarantees for the learner, but might perform much more poorly than other learning algorithms against particular optimizer payoffs.

Source: ACM Conference on Economics and Computation (EC)

DOI: 10.1145/3670865.3673517

Authors: Eshwar Ram Arunachaleswaran, Natalie Collina, Jon Schneider

Institutions: University of Pennsylvania, Google (United States)

Language: English

Topic: Advanced Bandit Algorithms Research

Subfield: Management Science and Operations Research

Field: Decision Sciences

Domain: Social Sciences

Open Access status: green

Funders: National Science Foundation, Amazon Web Services

Award: CAREER: Learning and property testing -- a complexity theoretic perspective