HEART-library Documentation

The Hardened Extension of the Adversarial Robustness Toolkit (HEART) is a modular open-source Python library that provides AI developers and researchers with Testing and Evaluation (T&E) tools to assess AI model performance under adversarial attacks and improve model resiliency.

Get Started with HEART

---

Quick Start

Use these resources to get an introduction to HEART and install the library

• Introduction to HEART

• Installation and Setup

Tutorials

If you are new to AI Security, the tutorials will introduce you to key concepts and workflows

• Drone Object Detection

How-to Guides

If you are familiar with AI Security and know what you want to do, the how-to guides will show you step-by-step how to do it with HEART tools

• Object Detection How-to Guides

• Image Classification How-to Guides

Explanations

The explanations provide in-depth descriptions of relevant technical concepts

• Attack Types

• Evaluation Metrics

• Overview - Creating Adversarial Patches

See all

Reference Materials

Reference Materials provide further resources for your understanding of both AI Security concepts and HEART tools

• Attack Cards

• Evaluation Pathways

• Modules

About HEART

Learn more about HEART, its background, mission, and contributors

• Contributors

• Legal

---

Acknowledgement

This material is based upon work supported by the Chief Digital and Artificial Intelligence Office under Contract No. W519TC-23-9-2037. The views and conclusions contained herein are those of the author(s) and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the U.S. Government.