robustness-verification

alpha-beta-CROWN: An Efficient, Scalable and GPU Accelerated Neural Network Verifier (winner of VNN-COMP 2021, 2022, 2023, 2024, 2025)

auto_LiRPA: An Automatic Linear Relaxation based Perturbation Analysis Library for Neural Networks and General Computational Graphs

Neural Network Verification Software Tool https://www.verivital.com Documentation:

Certified defense to adversarial examples using CROWN and IBP. Also includes GPU implementation of CROWN verification algorithm (in PyTorch).

Formal Verification of Neural Feedback Loops (NFLs)

A neural network verification tool based on the DPLL(T) SMT Solving algorithm.

β-CROWN: Efficient Bound Propagation with Per-neuron Split Constraints for Neural Network Verification

Reference implementations for RecurJac, CROWN, FastLin and FastLip (Neural Network verification and robustness certification algorithms) [Do not use this repo, use https://github.com/Verified-Intelligence/auto_LiRPA instead]

[NeurIPS 2019] H. Chen*, H. Zhang*, S. Si, Y. Li, D. Boning and C.-J. Hsieh, Robustness Verification of Tree-based Models (*equal contribution)

[ICLR 2020] Code for paper "Robustness Verification for Transformers"

[CCS 2021] TSS: Transformation-specific smoothing for robustness certification

This github repository contains the official code for the paper, "Evolving Robust Neural Architectures to Defend from Adversarial Attacks"

The official repo for GCP-CROWN paper

certifying robustness of neural network via convex optimization

WraLU is an artifact for the paper "ReLU Hull Approximation" (POPL'24), which provides a sound but incomplete neural network verifier by over-approximating ReLU function hull.

Sampling-based Scalable Quantitative Verification for DNNs

[NeurIPS 2021] Towards Better Understanding of Training Certifiably Robust Models against Adversarial Examples | ⛰️⚠️

Fast Adversarial Robustness Certification of Nearest Prototype Classifiers for Arbitrary Seminorms [NeurIPS 2020]

This github repository contains the official code for the papers, "Robustness Assessment for Adversarial Machine Learning: Problems, Solutions and a Survey of Current Neural Networks and Defenses" and "One Pixel Attack for Fooling Deep Neural Networks"

WraAct is an artifact for the paper "Convex Hull Approximation for Activation Functions" (OOPSLA'25), which provides a sound but incomplete neural network verifiers by over-approximating the function hulls of various activation functions (including leaky ReLU, ReLU, sigmoid, tanh, and maxpool).