This repository is a curated list of academic papers and resources focused on data poisoning and backdoor attacks in machine learning, along with their corresponding defenses. It serves as a comprehensive reference for researchers and practitioners in AI security, providing a structured overview of the evolving threat landscape and mitigation strategies.

How It Works

The repository organizes research papers by conference or publication year, categorizing them into attacks, defenses, surveys, and benchmarks. It links directly to papers and, where available, associated code repositories, enabling users to quickly access and evaluate specific research contributions. The collection highlights key findings and methodologies in the field.

Quick Start & Requirements

This repository is a curated list and does not require installation or execution. It serves as a knowledge base.

Highlighted Details

• Comprehensive coverage of data poisoning and backdoor attacks across major ML conferences (NeurIPS, ICML, ICLR, CVPR, etc.) from 2015 to 2024.
• Includes links to over 300 research papers and numerous code repositories for practical implementation and verification.
• Features dedicated sections for surveys, benchmarks, and specific attack/defense methodologies.
• Covers a wide range of ML domains, including computer vision, natural language processing, federated learning, and reinforcement learning.

Maintenance & Community

This repository is noted as "no longer maintained" by the original author, with the last update pertaining to ACL 2024. However, contributions via pull requests are welcomed.

Licensing & Compatibility

The repository itself is not software and thus does not have a software license. The linked papers are subject to their respective publication licenses and copyright.

Limitations & Caveats

The repository explicitly states it may not include all relevant papers and encourages community contributions to fill gaps. Its "no longer maintained" status suggests potential for outdated information or unaddressed new research.