NeurIPS 2020

Pipeline PSRO: A Scalable Approach for Finding Approximate Nash Equilibria in Large Games

Meta Review

The paper presents an improvement to policy-space response oracles (PSRO) algorithm, in particular its parallel form DCH. The fix is straight-forward, but the effect it has in practice is important as demonstrated by the authors across known benchmarks and randomized games. Please take the reviews into account when preparing the final version. If possible, it might be useful to include a mention of Probe Stratego/Barrage agent, and a summary of the reasons that it could not be used as a benchmark at this time.