A Semidefinite Relaxation Based Branch-and-Bound Method for Tight Neural Network Verification

We introduce a novel method based on semidefinite program (SDP) for the tight and efficient verification of neural networks. The proposed SDP relaxation advances present state art in SDP-based network by adding set linear constraints eigenvectors. extend this combining it with branch-and-bound that can provably close gap up to zero. show formally approach leads tighter solution than art. report experimental results showing outperforms baselines terms verified accuracy while retaining an acceptable computational overhead.