Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
The paper proposes a method to control the level set of the decision surface of neural networks. The reviewers found the approach to be novel and convincingly demonstrated to work on tasks such as CIFAR-10. Furthermore, the method is shown to increase robustness to adversarial perturbations. The work successfully tackles an important topic and as such would be of interesting to the NeurIPS community.