Reachability Analysis of Deep Neural Networks with Provable Guarantees

TL;DR

This paper introduces a novel reachability analysis method for deep neural networks that provides provable guarantees on output ranges, enhancing safety verification and robustness assessment.

Contribution

It presents a new adaptive nested optimization algorithm for reachability analysis, improving efficiency, scalability, and applicability over existing verification methods.

Findings

Demonstrates effectiveness on various DNNs

Outperforms state-of-the-art verification approaches

Handles a broader class of networks

Abstract

Verifying correctness of deep neural networks (DNNs) is challenging. We study a generic reachability problem for feed-forward DNNs which, for a given set of inputs to the network and a Lipschitz-continuous function over its outputs, computes the lower and upper bound on the function values. Because the network and the function are Lipschitz continuous, all values in the interval between the lower and upper bound are reachable. We show how to obtain the safety verification problem, the output range analysis problem and a robustness measure by instantiating the reachability problem. We present a novel algorithm based on adaptive nested optimisation to solve the reachability problem. The technique has been implemented and evaluated on a range of DNNs, demonstrating its efficiency, scalability and ability to handle a broader class of networks than state-of-the-art verification approaches.