Shellcode Dataset for Adversarial ML Robustness Evaluation

Sudip Saha's shellcode dataset was used to evaluate a hybrid adversarially-trained deep learning framework. The dataset is 5.5 KB in size, stored in XLS format, and was last updated on June 1, 2026. It was used in experiments demonstrating model performance against FGSM and PGD adversarial attacks.

Use Cases

• Benchmarking adversarial robustness of ML models based on reported clean and adversarial accuracy metrics.
• Training models for malware detection based on shellcode data mentioned in the description.
• Studying the trade-off between model accuracy and robustness based on the described adversarial loss components.
• Analyzing model convergence stability and calibration error based on the training dynamics mentioned.

Strengths

• Dataset is small (5.5 KB), facilitating quick download and initial experimentation.
• License is CC-BY-4.0, permitting sharing and adaptation with attribution.
• Specific performance benchmarks are provided (e.g., 97.88% clean accuracy, 84.9% under FGSM attack).

Limitations

• Row count and column definitions are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality and structure require manual inspection after download.
• The dataset's scope appears limited to shellcode, which may narrow its applicability.

Provenance

Source

Sudip Saha via figshare

Collection Method

Likely collected or generated for research on adversarial machine learning in cybersecurity.

Freshness

Last updated 2026-06-01 17:55:11; freshness should be verified.