A comment on Guo et al. [arXiv:2206.11228]

TL;DR

This paper critically examines Guo et al.'s claim that adversarially trained neural networks match primate neural robustness, questioning the interpretation and presentation of their experimental results.

Contribution

It provides a critical analysis of Guo et al.'s findings, highlighting potential issues in their interpretation of neural robustness comparisons.

Findings

Raises doubts about the robustness equivalence claim

Highlights possible misinterpretation of experimental results

Calls for careful reconsideration of the conclusions

Abstract

In a recent article, Guo et al. [arXiv:2206.11228] report that adversarially trained neural representations in deep networks may already be as robust as corresponding primate IT neural representations. While we find the paper's primary experiment illuminating, we have doubts about the interpretation and phrasing of the results presented in the paper.