Abstraction-Based Proof Production in Formal Verification of Neural Networks

TL;DR

This paper introduces a new framework for proof-producing abstraction-based verification of neural networks, enabling scalable and trustworthy proofs by combining existing verifiers with novel formal proof methods for abstraction soundness.

Contribution

It presents the first method for generating formal proofs of abstraction soundness in neural network verification, bridging the gap between scalability and provable guarantees.

Findings

Framework successfully separates correctness and soundness proofs.

First formal proof method for abstraction soundness in DNN verification.

Supports scalable, trustworthy verification with abstraction techniques.

Abstract

Modern verification tools for deep neural networks (DNNs) increasingly rely on abstraction to scale to realistic architectures. In parallel, proof production is becoming a critical requirement for increasing the reliability of DNN verification results. However, current proofproducing verifiers do not support abstraction-based reasoning, creating a gap between scalability and provable guarantees. We address this gap by introducing a novel framework for proof-producing abstraction-based DNN verification. Our approach modularly separates the verification task into two components: (i) proving the correctness of an abstract network, and (ii) proving the soundness of the abstraction with respect to the original DNN. The former can be handled by existing proof-producing verifiers, whereas we propose the first method for generating formal proofs for the latter. This preliminary work aims to enable scalable and trustworthy verification by supporting common abstraction techniques within a formal proof framework.