AUTHOR RESPONSE TO THE REVIEWS OF "CORMORANT: COVARIANT MOLECULAR NEURAL NETWORKS" We thank all three reviewers for their insightful comments and positive evaluations of our manuscript. We will update the paper to reflect their suggestions by the camera ready deadline. We will shortly release a Python library that implements the Cormorant architecture. Currently we are just cleaning up and documenting the code. As for the other points brought up by the reviewers we have the following comments: 1. Structure and supplement: As suggested by Reviewer 2, we will move some details of the technical implementation in Section 4.4 to the supplement.

Neural Information Processing Systems http://nips.cc/

We investigate whether post-hoc model explanations are effective for diagnosing model errors-model debugging. In response to the challenge of explaining a model's prediction, a vast array of explanation methods have been proposed. Despite increasing use, it is unclear if they are effective.

Memory is an important aspect of intelligence and plays a role in many deep reinforcement learning models. However, little progress has been made in understanding when specific memory systems help more than others and how well they generalize.

Habibian et al., 2019, Y ang et al., 2020a], which can reduce a sizable amount of global internet traffic. State-of-the-art neural methods for lossy image compression [Ballé et al., 2018, Minnen et al., 2018, Lee et al., 2019] learn a mapping between images and latent variables with a variational

In contrast, human perception is much more robust to such perturbations.