Part III - Practice

using state-of-the-art toolboxes for neural network verification.

tl;dr: We first give a very brief overview of the tools used in the tutorial, including the motivations, features and design philosophy of the tools. Our tutorial will first cover the versatile PyTorch based auto_LiRPA library for incomplete verification and certified defense. The library allows easy handling of general neural networks, including ResNet, LSTM, and Transformer. Then, we introduce the usage of one of the state-of-the-art complete verification tools, ɑ,β-CROWN (alpha-beta-CROWN), which outperformed 11 other tools and won the first prize in VNN-COMP 2021.