Benign Executable Binaries for Adversarial Machine Learning Security Research

AsiaCCS 2021 research provides a collection of benign executable binaries used to study the robustness of machine learning-based static malware analysis. The dataset supports work on adversarial attacks where executable bytes are modified to evade detection. It is associated with the paper 'Malware Makeover: Breaking ML-based Static Analysis by Modifying Executable Bytes.'

Use Cases

• Training baseline classifiers for malware detection based on executable byte features
• Testing adversarial robustness of static analysis models by modifying executable bytes
• Benchmarking the performance of ML-based security tools against evasion techniques

Strengths

• Dataset size is explicitly stated as 5,091,270,656 bytes (approximately 4.7 GB)
• Directly linked to a peer-reviewed conference publication (AsiaCCS 2021)
• Cross-platform presence on figshare indicates established use in the research community

Limitations

• Column names and specific data structure are unknown, limiting detailed understanding of features
• Exact row count is not provided across any source
• License (CC-BY-NC-ND-4.0) is restrictive, prohibiting commercial use, modifications, and derivatives

Provenance

Source

Mahmood Sharif, associated with the AsiaCCS 2021 paper by Lucas, Sharif, Bauer, Reiter, and Shintre.

Collection Method

Collected as part of research for adversarial machine learning in security.

Time Range

Associated with research presented at AsiaCCS 2021.

Freshness

Last updated on 2026-03-19, indicating ongoing maintenance.

Geography

null