Reviews: Spectral Signatures in Backdoor Attacks

This paper provides a methodology called spectral signatures that analyzes the covariance of the feature representation learnt by the neural network to detect backdoor attacks. Detecting backdoor attacks in neural networks is an important and very recent problem that hasn't been solved properly. The paper is well-written and easy to follow. However, I found that the evaluation is limited and there are some inconsistencies in the paper. After reading the rebuttal, I am still concerned by the limitations in the experimental evaluation.