Adversarial Robustness Toolbox
The Adversarial Robustness Toolbox (ART) is a Python library for Machine Learning Security. It provides tools that enable developers and researchers to defend and evaluate 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, 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.). The library is under continuous development and is hosted by the Linux Foundation AI & Data Foundation (LF AI & Data). ART is designed to be used by both Red and Blue Teams, providing a comprehensive set of tools for evaluating and improving the security of Machine Learning models. The library includes a range of features, including attacks, defences, estimators, and metrics, as well as a range of examples and notebooks to help users get started. ART is widely used in the field of Machine Learning Security and has been cited in numerous research papers. The library is open-source and is available on GitHub, where it has received over 6,000 stars and has been forked over 1,400 times. ART is licensed under the MIT License and is available for use in commercial and non-commercial applications. The library has a strong community of developers and users, with a Slack channel and a range of documentation and resources available to help users get started and contribute to the project. Overall, ART is a powerful and widely-used library for Machine Learning Security, providing a comprehensive set of tools for evaluating and improving the security of Machine Learning models.
- Category
- Model Serving
- Pricing
- Open Source
Problem solved
The Adversarial Robustness Toolbox (ART) solves the problem of evaluating and improving the security of Machine Learning models against adversarial threats. Machine Learning models are vulnerable to a range of attacks, including Evasion, Poisoning, Extraction, and Inference, which can compromise their security and reliability. ART provides a comprehensive set of tools for evaluating and improving the security of Machine Learning models, enabling developers and researchers to defend against these threats and improve the overall security of their models. By using ART, developers and researchers can identify vulnerabilities in their models, evaluate the effectiveness of different defences, and improve the overall security of their models.
Use cases
- Evaluating the security of Machine Learning models against adversarial threats
- Improving the security of Machine Learning models using defences and estimators
- Developing and testing new attacks and defences for Machine Learning models
- Using Machine Learning models in security-critical applications, such as autonomous vehicles and medical diagnosis
Audience
- Developers
- ML Engineers
- Data Scientists
- Researchers
Pros
- Comprehensive set of tools for evaluating and improving the security of Machine Learning models
- Supports all popular machine learning frameworks and data types
- Wide range of examples and notebooks to help users get started
- Strong community of developers and users
Cons
- Steep learning curve due to the complexity of the library and the field of Machine Learning Security
- Requires a strong background in Machine Learning and programming
- Limited support for certain machine learning frameworks and data types
Official website: https://github.com/Trusted-AI/adversarial-robustness-toolbox
Canonical LambdaBase page: https://www.lambdabase.com/tools/adversarial-robustness-toolbox