policy-representations

This paper investigates learning useful policy representations (embeddings) in two-player zero-sum imperfect-information games, introducing methods for creating policy datasets, learning embeddings, and evaluating them on downstream tasks using Kuhn and Leduc Poker.