Neural Network Verification with PyRAT

2024-10-31

arXiv.org

AI Summary

Neural networks lack formal safety guarantees in safety-critical domains such as healthcare and autonomous transportation. To address this, we propose PyRAT, a verification framework based on abstract interpretation. Its core innovation is a multi-granularity abstract domain coordination mechanism that jointly leverages interval, zonotope, and DeepPoly abstractions to enable both efficient and precise forward propagation and faithful modeling of nonlinear activations. This integration significantly improves verification speed and tightness of bounds. In the VNN-Comp 2024 international verification competition, PyRAT achieved second place in overall performance. The framework has been successfully deployed in multiple industrial collaboration projects, delivering formally verifiable robustness guarantees for real-world AI systems. PyRAT thus establishes a state-of-the-art solution for scalable, sound, and practical neural network verification.

Technology Category

Natural Language Processing: Safety and RobustnessComputer Vision: Adversarial Attacks & RobustnessMachine Learning: Deep Neural Architectures and Foundation Models

Application Category

Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsSystems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applicationsSearch and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

As AI systems are becoming more and more popular and used in various critical domains (health, transport, energy, ...), the need to provide guarantees and trust of their safety is undeniable. To this end, we present PyRAT, a tool based on abstract interpretation to verify the safety and the robustness of neural networks. In this paper, we describe the different abstractions used by PyRAT to find the reachable states of a neural network starting from its input as well as the main features of the tool to provide fast and accurate analysis of neural networks. PyRAT has already been used in several collaborations to ensure safety guarantees, with its second place at the VNN-Comp 2024 showcasing its performance.

Problem

• Verifying neural network safety and robustness
• Providing guarantees for AI systems in critical domains
• Analyzing reachable states using abstract interpretation

Innovation

• Abstract interpretation for neural network verification
• Reachable state analysis from input specifications
• Fast and accurate neural network analysis