NeurIPS 2020

Quantifying Learnability and Describability of Visual Concepts Emerging in Representation Learning

Meta Review

The reviewers generally agreed that this paper provides a fresh perspective on a timely problem, i.e., evaluating the quality of learned representations via learnability and describability. After the rebuttal, all the reviewers (except R3) recommended Accept. R3 initially raised some concerns and the authors properly addressed them in the rebuttal. I concur with the reviewers' recommendation to accept the paper.