trusted-ai/adversarial-robustness-toolbox is a Python library for Adversarial Robustness Toolbox (ART), Machine Learning Security. The latest release is 1.20.1. It implements attacks and defenses for four threats — evasion, poisoning, extraction, and inference — letting you test the adversarial robustness of machine learning models and applications and evaluate the defenses themselves.
ART is hosted by the Linux Foundation AI & Data Foundation. It supports several machine learning frameworks such as TensorFlow, PyTorch, and scikit-learn, and covers image, tabular, audio, and video data along with classification, object detection, and speech recognition tasks. It is implemented in Python, licensed under MIT, and has collected 6,282 stars. Created on March 15, 2018, it covers a wide range of attacks and defenses from both red team and blue team perspectives inside a single library.
However, what changed in 1.20.1 is not included in the given materials. With no changelist, there is no way to tell whether this release is better than the previous version. The ratings are also thin: the momentum score is a low 1.6 out of 10, and no freshness, technical, or quality scores are given. The MIT license makes it easy to adopt and to build attack and defense scenarios. There are no benchmarks comparing it with other tools, so it is not known whether it leads on any particular task.