Reviews: Optimistic Regret Minimization for Extensive-Form Games via Dilated Distance-Generating Functions

Through that lens, I liked this paper. This paper considers the application of optimistic regret-minimization algorithms to extensive form games. It follows a similar setup as CFR, the current SOTA algorithm for solving large games, which decomposes the overall regret-minimization problem into independent subproblems at each information set. Theoretically, the new algorithms should have faster convergence than the CFR variant of CFR. Empirically, the authors show that these methods converge dramatically faster than CFR in small games, but slower in the slightly larger game of Leduc hold'em.