Reconnaissance Dataset for Adversarial Machine Learning Model Evaluation

Sudip Saha's dataset, published on figshare in 2026, is used for evaluating adversarial robustness of machine learning models. The description reports model performance metrics, including 97.88% accuracy on clean data and 84.9% accuracy under FGSM attacks. The dataset is 5.5 KB in size and is available in XLS format under a CC-BY-4.0 license.

Use Cases

• Benchmarking adversarial robustness of models based on reported accuracy metrics under FGSM and PGD attacks.
• Training hybrid adversarially-trained models based on the described framework integrating reinforcement learning and knowledge-driven regularization.
• Analyzing the trade-off between clean accuracy and adversarial robustness mentioned in the paper's experimental results.

Strengths

• Specific performance metrics are provided, such as 97.88% clean accuracy and 81.75% accuracy under PGD attacks.
• The dataset is associated with a published research methodology describing a hybrid adversarial training framework.
• It is openly available under a permissive CC-BY-4.0 license.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's small size (5.5 KB) suggests a very limited scope or a sample rather than a comprehensive evaluation corpus.

Provenance

Source

Sudip Saha via figshare

Collection Method

Likely generated or collected as part of academic research on adversarial machine learning.

Freshness

Last updated 2026-06-01 17:55:10; freshness should be verified.