Home

Welcome to the Adversarial Robustness Toolbox

Adversarial Robustness Toolbox (ART) is a Python library for Machine Learning Security. ART provides tools that enable developers and researchers to evaluate, defend, certify and verify Machine Learning models and applications against the adversarial threats of Evasion, Poisoning, Extraction, and Inference. ART supports all popular machine learning frameworks (TensorFlow, Keras, PyTorch, MXNet, scikit-learn, XGBoost, LightGBM, CatBoost, GPy, etc.), all data types (images, tables, audio, video, etc.) and machine learning tasks (classification, object detection, speech recognition, generation, certification, etc.).

Adversarial Threats

ART for Red and Blue Teams (selection)

Supported Machine Learning Libraries

• TensorFlow (v1 and v2) (www.tensorflow.org)
• Keras (www.keras.io)
• PyTorch (www.pytorch.org)
• MXNet (https://mxnet.apache.org)
• Scikit-learn (www.scikit-learn.org)
• XGBoost (www.xgboost.ai)
• LightGBM (https://lightgbm.readthedocs.io)
• CatBoost (www.catboost.ai)
• GPy (https://sheffieldml.github.io/GPy/)