The authors demonstrate how to train a CNN to be more robust to random noise perturbations and adversarial attacks by regularizing the network to match the similarity matrix accrued from neural recordings from mice. Building robust CNNs is an extremely important application and leveraging ideas from neuroscience to make this happen is a clever and exciting intersection of the fields. Reviewers commented that the paper was clearly written; the experiments are properly designed and controlled; and the research questions is crisp. There was some minor issues that need to be addressed as the reviewers request including an improved related work section. Assuming all of the issues mentioned by the reviewers are addressed, this paper will be accepted into this conference.