Proof Minimization in Neural Network Verification

TL;DR

This paper introduces algorithms to minimize proofs of unsatisfiability in neural network verification, significantly reducing proof size and checking time while maintaining verification reliability.

Contribution

The authors develop novel algorithms for proof minimization in DNN verification, improving proof efficiency and reliability.

Findings

Proof size reduced by up to 82%

Proof checking time decreased by up to 88%

Runtime overhead of 7%-20% during verification

Abstract

The widespread adoption of deep neural networks (DNNs) requires efficient techniques for verifying their safety. DNN verifiers are complex tools, which might contain bugs that could compromise their soundness and undermine the reliability of the verification process. This concern can be mitigated using proofs: artifacts that are checkable by an external and reliable proof checker, and which attest to the correctness of the verification process. However, such proofs tend to be extremely large, limiting their use in many scenarios. In this work, we address this problem by minimizing proofs of unsatisfiability produced by DNN verifiers. We present algorithms that remove facts which were learned during the verification process, but which are unnecessary for the proof itself. Conceptually, our method analyzes the dependencies among facts used to deduce UNSAT, and removes facts that did not contribute. We then further minimize the proof by eliminating remaining unnecessary dependencies, using two alternative procedures. We implemented our algorithms on top of a proof producing DNN verifier, and evaluated them across several benchmarks. Our results show that our best-performing algorithm reduces proof size by 37%-82% and proof checking time by 30%-88%, while introducing a runtime overhead of 7%-20% to the verification process itself.