Designing and discovering new molecules is key for addressing some of world's most pressing

Generated samples on CelebA-HQ 256 256 (left) and unconditional CIFAR10 (right) 34th Conference on Neural Information Processing Systems (NeurIPS 2020), V ancouver, Canada.

On one hand, as other data-driven deep learning methods, our method, namely variational denoising network (VDN), can perform denoising efficiently due to its explicit form of posterior expression. On the other hand, VDN inherits the advantages of traditional model-driven approaches, especially the good generalization capability of generative models.

Recent work has shown that CNN-based depth and ego-motion estimators can be learned using unlabelled monocular videos. However, the performance is limited by unidentified moving objects that violate the underlying static scene assumption in geometric image reconstruction.

Despite its success, the influence of different view choices has been less studied.

The authors compare results of shallow networks versus deep networks on the task at hand and demonstrate better performance for deep networks. They also use the scores of the shallow versus deep networks to analyse the role of basic vs. handcrafted features. GENERAL OPINION: This is a very straightforward paper that applies an existing technique on a novel, relevant and interesting application. The technical quality and clarity are very sound. Even though there is no real novelty on the side of machine learning research itself, in my opinion the introduction of a new application domain for deep learning in an extremely relevant scientific field in itself warrants acceptance. One caveat is that this reviewer has a background in physics, and is hence quite familiar with (and enthusiastic about) the concepts explained in the introduction. Whereas I found this explanation very clear, I cannot really speak for the rest of the NIPS community. DETAILED COMMENTS: - If possible, it might be informative to provide a depiction that illustrates the raw data one gets from a typical collision event (an illustration of the detected particles emerging from a single collision). This might give a better feel of the kind of data that is dealt with.

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

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. Summary: This paper proposes a deep learning based spatial attention mechanism based on a probabilistic generative model. The approach enables identifying novel objects in large images, and hence might allow for better exploitation of unlabeled, uncropped data. Main Comments: This paper takes a solid step towards being able to apply deep learning methods to large uncropped, unlabeled images. The experimental section contains a number of interesting experiments verifying that the approximate inference method is working, and that HMC is still worthwhile beyond this.