Work

Experimental outputs for "Spike Density Divergence Attribution: A Baseline-Relative Explainability Framework for Spiking Neural Network Autoencoders in IoT Anomaly Detection"

Year: 2026
Type: dataset
Abstract: Large per-sample outputs that accompany the code repository of the article. The code, configurations, aggregate tables, and revision analyses are in the GitHub repository; this record holds the files that exceed the GitHub size limit. pipeline_raw_per_sample_outputs.tar.gz: per-sample outputs of the main evaluation (2 datasets x 3 seeds x 10 folds = 60 SNN-AE models): faithfulness, deletion/insertion, stability, runtime, local explanation, SARED, feature behavior, suppression, and the split audit. pipeline_trained_models.tar.gz: the 60 trained SNN-AE models. revision_eq9_ablation_replay.tar.gz: replay of all 60 seed-fold units for the signal-projection ablation of Eq. (9), with per-sample attributions and metrics and parity checks against the original evaluation. FILE_MANIFEST.sha256: SHA-256 of every file inside the archives; SHA256SUMS: SHA-256 of the archives. Each archive keeps its relative paths, so it extracts into the results/ folder of the repository. The Edge-IIoTset and CICIoT2023 datasets are not redistributed; they are available from their original sources.
Language: English
Open Access status: green
OpenAlex ID: W7221453817
Cites:
Cited by:
Related to: 10