The Third International Verification of Neural Networks Competition (VNN-COMP 2022): Summary and Results

TL;DR

VNN-COMP 2022 advances fair comparison in neural network verification by standardizing inputs (ONNX), specifications (VNN-LIB), and hardware (AWS) with automated submission and evaluation. The competition reveals a convergence around GPU-accelerated bound-propagation with branch-and-bound (notably α,β-CROWN) as the most effective strategy on diverse benchmarks, while exposing gaps in tool reliability and safeties via counterexamples. It also broadens the benchmark suite to include large-scale, real-world architectures (Large ResNets, Carvana UNet) and domain-relevant tasks (RUL-CNN, RL benchmarks). Overall, the results demonstrate progress in verification capabilities and emphasize the value of automated workflows, witness sharing, and broader, more challenging benchmarks for driving future improvements.

Abstract

This report summarizes the 3rd International Verification of Neural Networks Competition (VNN-COMP 2022), held as a part of the 5th Workshop on Formal Methods for ML-Enabled Autonomous Systems (FoMLAS), which was collocated with the 34th International Conference on Computer-Aided Verification (CAV). VNN-COMP is held annually to facilitate the fair and objective comparison of state-of-the-art neural network verification tools, encourage the standardization of tool interfaces, and bring together the neural network verification community. To this end, standardized formats for networks (ONNX) and specification (VNN-LIB) were defined, tools were evaluated on equal-cost hardware (using an automatic evaluation pipeline based on AWS instances), and tool parameters were chosen by the participants before the final test sets were made public. In the 2022 iteration, 11 teams participated on a diverse set of 12 scored benchmarks. This report summarizes the rules, benchmarks, participating tools, results, and lessons learned from this iteration of this competition.

Figures (16)

• Figure 1: Cactus Plot for All Instances.
• Figure 2: Cactus Plot for Carvana Unet 2022.
• Figure 3: Cactus Plot for CIFAR100 Tiny ImageNet ResNet.
• Figure 4: Cactus Plot for CIFAR Biasfield.
• Figure 5: Cactus Plot for Collins Rul CNN.