What question did this study set out to answer?

To benchmark post hoc explainable artificial intelligence methods against an intrinsically interpretable compositional pipeline for regulatory clove quality grading under domain shift.

September 17, 2026ETRI JournalOpen Access

Explainable AI for clove quality grading: Benchmarking post hoc XAI and compositional interpretability under domain shift

View Full Paper

Ask AI

Bookmark

Authors

INInnocent Nyalala PNPatrick Vincent Ndowo

Discussion

Member takes

Overview

Benchmarking study reveals superior auditing fidelity in a compositional pipeline over post hoc XAI for clove grading, highlighting its potential for regulatory compliance.

Key Points

To benchmark post hoc explainable artificial intelligence methods against an intrinsically interpretable compositional pipeline for regulatory clove quality grading under domain shift.
Evaluated 4,603 expert-graded clove images across four Zanzibar State Trading Corporation quality grades using eight convolutional neural network and vision transformer architectures.
Benchmarked seven post hoc explainers (Grad-CAM, Grad-CAM++, ScoreCAM, LIME, GradientSHAP, CLS-Attention, and Chefer LRP) against a compositional segmentation–classification pipeline using a novel Explanation Energy Ratio metric and synthetic background-replacement domain shifts.
Post hoc explanation quality varied substantially across model architectures, with LIME showing high visual alignment on transformers despite high computational latency, while GradientSHAP proved too computationally intensive for edge deployment.
The compositional pipeline achieved an F1 score of 99.45% alongside a 100% Explanation Energy Ratio by design, making it the preferred approach for regulatory grading audits.

Cite This Study

Nyalala et al. (2026) studied this question.

synapsesocial.com/papers/6aabb7155f706d05830e5f42 https://doi.org/10.4218/etrij.2026-0225

View Full Paper

Ask AI

Bookmark

September 17, 2026ETRI JournalOpen Access

Explainable AI for clove quality grading: Benchmarking post hoc XAI and compositional interpretability under domain shift

View Full Paper

Ask AI

Bookmark

Authors

INInnocent Nyalala PNPatrick Vincent Ndowo

Discussion

Member takes

Overview

Benchmarking study reveals superior auditing fidelity in a compositional pipeline over post hoc XAI for clove grading, highlighting its potential for regulatory compliance.

Key Points

To benchmark post hoc explainable artificial intelligence methods against an intrinsically interpretable compositional pipeline for regulatory clove quality grading under domain shift.
Evaluated 4,603 expert-graded clove images across four Zanzibar State Trading Corporation quality grades using eight convolutional neural network and vision transformer architectures.
Benchmarked seven post hoc explainers (Grad-CAM, Grad-CAM++, ScoreCAM, LIME, GradientSHAP, CLS-Attention, and Chefer LRP) against a compositional segmentation–classification pipeline using a novel Explanation Energy Ratio metric and synthetic background-replacement domain shifts.
Post hoc explanation quality varied substantially across model architectures, with LIME showing high visual alignment on transformers despite high computational latency, while GradientSHAP proved too computationally intensive for edge deployment.
The compositional pipeline achieved an F1 score of 99.45% alongside a 100% Explanation Energy Ratio by design, making it the preferred approach for regulatory grading audits.

Cite This Study

Nyalala et al. (2026) studied this question.

synapsesocial.com/papers/6aabb7155f706d05830e5f42 https://doi.org/10.4218/etrij.2026-0225

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark