NeurIPS 2020

Few-shot Visual Reasoning with Meta-Analogical Contrastive Learning


Meta Review

Reviewers were split on this paper, with two recommending accept and two recommending reject. Reviewers and the AC appreciated the strong performance and found the method to be intriguing if a bit complicated. A main concern was that the methods were hard to understand and the paper was not clearly written. The rebuttal addressed several confusions and partially satisfied the reviewers. The AC concurs that clarity is still an issue and urges the authors to revise the writing for the final version, especially in Section 3, making clear how each aspect of the method relates to the idea of analogies. See also the excellent suggestions on simplifying the diagrams (R4) and clarifying the technical details and notation (R1, R2), among other comments from the reviewers.