Automated and Formal Synthesis of Neural Barrier Certificates for Dynamical Models

TL;DR

This paper presents an automated, neural network-based method for synthesizing Barrier Certificates to verify the safety of complex dynamical systems, significantly improving speed and efficiency over existing techniques.

Contribution

It introduces a formal, counterexample-guided inductive synthesis approach for neural barrier certificates, outperforming state-of-the-art methods in speed and data efficiency.

Findings

Faster synthesis of sound Barrier Certificates (up to 100x faster)

Significant reduction in verification time (up to 100,000x faster)

Effective on high-dimensional and hybrid dynamical models

Abstract

We introduce an automated, formal, counterexample-based approach to synthesise Barrier Certificates (BC) for the safety verification of continuous and hybrid dynamical models. The approach is underpinned by an inductive framework: this is structured as a sequential loop between a learner, which manipulates a candidate BC structured as a neural network, and a sound verifier, which either certifies the candidate's validity or generates counter-examples to further guide the learner. We compare the approach against state-of-the-art techniques, over polynomial and non-polynomial dynamical models: the outcomes show that we can synthesise sound BCs up to two orders of magnitude faster, with in particular a stark speedup on the verification engine (up to five orders less), whilst needing a far smaller data set (up to three orders less) for the learning part. Beyond improvements over the state of the art, we further challenge the new approach on a hybrid dynamical model and on larger-dimensional models, and showcase the numerical robustness of our algorithms and codebase.