NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:5207
Title:A Convex Relaxation Barrier to Tight Robustness Verification of Neural Networks

An interesting paper on robustness verification. The paper proposes a general framework for layer-wise LP relaxations and shows which relaxation is tighter. Further it shows that there is a theoretical barrier to layer-wise LP relaxations.