NeurIPS 2020

Theoretical Insights Into Multiclass Classification: A High-dimensional Asymptotic View


Meta Review

The paper studies the statistical behaviour of certain multiclass classification algorithms in the doubly-asymptotic limit of n, d -> ∞. The results elucidate certain differences compared to the analysis of binary classifiers, such as dependence on class-correlation matrices. One reviewer raised concerns about the results not providing insight into generalisation performance. The response indicates this is not the case, and this was corroborated by other reviews and my own reading. One critique raised by a couple of reviewers was regarding the specialised nature of the results, which are for linear classifiers and specific data models. These critiques are valid, but given that the paper is the first extension of existing work to the multiclass setting, it seems permissible to make tractable simplifications. Beyond these, three reviewers found the results to be interesting and well-presented. This was corroborated by an additional review sought after the author feedback. From my own reading of the paper, I concur that it is well written and the theoretical results seem of interest to the community. We thus recommend acceptance.