Fast Neural Network Verification via Shadow Prices

TL;DR

This paper introduces a shadow price-based technique to improve the efficiency of neural network verification, significantly reducing the number of input splits and computation time in safety-critical applications.

Contribution

It presents a novel shadow price approach that enhances input-splitting verification methods, leading to faster and more efficient neural network safety verification.

Findings

Reduces the number of input partitions needed for verification

Significantly decreases computation times on ACAS benchmark tasks

Potentially applicable to various machine learning safety applications

Abstract

To use neural networks in safety-critical settings it is paramount to provide assurances on their runtime operation. Recent work on ReLU networks has sought to verify whether inputs belonging to a bounded box can ever yield some undesirable output. Input-splitting procedures, a particular type of verification mechanism, do so by recursively partitioning the input set into smaller sets. The efficiency of these methods is largely determined by the number of splits the box must undergo before the property can be verified. In this work, we propose a new technique based on shadow prices that fully exploits the information of the problem yielding a more efficient generation of splits than the state-of-the-art. Results on the Airborne Collision Avoidance System (ACAS) benchmark verification tasks show a considerable reduction in the partitions generated which substantially reduces computation times. These results open the door to improved verification methods for a wide variety of machine learning applications including vision and control.