LGGTMay 27, 2021

Conic Blackwell Algorithm: Parameter-Free Convex-Concave Saddle-Point Solving

arXiv:2105.13203v35 citations
Originality Incremental advance
AI Analysis

This provides a scalable, parameter-free solution for optimization problems in game theory and robust decision-making, though it is incremental as it builds on ideas from existing regret minimization methods.

The paper tackles convex-concave saddle-point problems by developing a parameter-free algorithm, the Conic Blackwell Algorithm⁺ (CBA⁺), which achieves O(1/√T) average regret and outperforms state-of-the-art methods in experiments on matrix games and distributionally robust optimization.

We develop new parameter-free and scale-free algorithms for solving convex-concave saddle-point problems. Our results are based on a new simple regret minimizer, the Conic Blackwell Algorithm$^+$ (CBA$^+$), which attains $O(1/\sqrt{T})$ average regret. Intuitively, our approach generalizes to other decision sets of interest ideas from the Counterfactual Regret minimization (CFR$^+$) algorithm, which has very strong practical performance for solving sequential games on simplexes. We show how to implement CBA$^+$ for the simplex, $\ell_{p}$ norm balls, and ellipsoidal confidence regions in the simplex, and we present numerical experiments for solving matrix games and distributionally robust optimization problems. Our empirical results show that CBA$^+$ is a simple algorithm that outperforms state-of-the-art methods on synthetic data and real data instances, without the need for any choice of step sizes or other algorithmic parameters.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes