NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7160
Title:Learning to Correlate in Multi-Player General-Sum Sequential Games

This paper proves that coarse correlated equilibria can be learned by a variant of CFR, called CFR-Jr. The main concerns of the reviewers are actually the CFR-S algorithm that seems quite naive and whose purpose is not clear. Is it just to make CFR-Jr looks better ? This could be a good "oral presentation" technique, but we believe that this paper would actually greatly benefit from being an important reorganization where CFR-S is more or less deleted (and replaced by interesting, short remarks, comments and discussion). We strongly suggest to the authors to perform this reorganization and incorporate the reviewers suggestions to the updated and revised version of the papers; it will definitely have a better impact that way.