Introduction to HEART

HEART is a Python extension library for Machine Learning Security that builds on the popular adversarial robustness algorithms in the Adversarial Robustness Toolbox (ART).

The extension library is operation-ready and tailored for real-world DoD use cases, offering essential adversarial robustness methods within the three evaluation tool dimensions: physical realizability, perturbation type, black/white box. HEART allows the user to leverage core ART algorithms, while providing additional benefits to the AI Test & Evaluation (T&E) engineer:

  • Support for T&E of models for DoD use cases (developers, researchers and evaluators focused on adversarial machine learning capabilities)

  • Alignment to MAITE protocols to access this subset of ART and other JATIC tools for seamless T&E workflows

  • Essential subset of adversarial robustness methods for targeted AI security coverage

  • Assessment quality assurance in the form of metadata

  • In-depth support for users in the form of guides and examples

  • Front-end application for low-code users: HEART Gradio Application

Additional Resources

HEART is a curated subset of tools from the Adversarial Robustness Toolbox, an open-source library maintained by Linux Foundation AI and Data Foundation.

HEART fits into a broader ecosystem of T&E tools as a part of the Joint AI Test Infrastructure Capability (JATIC). Learn about the program and other capabilities via the CDAO JATIC Public Documentation.