What question did this study set out to answer?

To establish a model-agnostic, black-box feature attribution framework capable of generating human-interpretable explanations without requiring fully invertible input transformation functions.

August 19, 2025Machine LearningOpen Access

Toward practical human-interpretable explanations

View Full Paper

Ask AI

Bookmark

Authors

AMAlon Malach AMAmiel MeiselesBen-Gurion University of the Negev RBRon BittonBen-Gurion University of the Negev

Discussion

Member takes

Overview

Algorithm evaluation demonstrates faithful human-interpretable explanations in complex machine learning models, highlighting practical black-box explainability without invertible mappings.

Key Points

To establish a model-agnostic, black-box feature attribution framework capable of generating human-interpretable explanations without requiring fully invertible input transformation functions.
Introduced Latent SHAP, an explainability framework bridging low-level model inputs to high-level conceptual features without invertible mappings.
Conducted quantitative faithfulness assessments on a self-generated artificial dataset and the dSprites benchmark dataset.
Evaluated practical utility across ensemble models, deep neural networks, and large language models within computer vision, natural language processing, and cybersecurity tasks.
Faithfulness evaluations on controlled synthetic and benchmark datasets verified the fidelity of the generated feature attribution explanations.
Latent SHAP produced actionable, human-interpretable explanations across real-world vision, language, and cybersecurity domains lacking invertible transformation functions.

Cite This Study

Malach et al. (2025) studied this question.

synapsesocial.com/papers/6a0ebeeea14f152feaf9c8ae https://doi.org/10.1007/s10994-025-06852-8

View Full Paper

Ask AI

Bookmark

August 19, 2025Machine LearningOpen Access

Toward practical human-interpretable explanations

View Full Paper

Ask AI

Bookmark

Authors

AMAlon Malach AMAmiel MeiselesBen-Gurion University of the Negev RBRon BittonBen-Gurion University of the Negev

Discussion

Member takes

Overview

Algorithm evaluation demonstrates faithful human-interpretable explanations in complex machine learning models, highlighting practical black-box explainability without invertible mappings.

Key Points

To establish a model-agnostic, black-box feature attribution framework capable of generating human-interpretable explanations without requiring fully invertible input transformation functions.
Introduced Latent SHAP, an explainability framework bridging low-level model inputs to high-level conceptual features without invertible mappings.
Conducted quantitative faithfulness assessments on a self-generated artificial dataset and the dSprites benchmark dataset.
Evaluated practical utility across ensemble models, deep neural networks, and large language models within computer vision, natural language processing, and cybersecurity tasks.
Faithfulness evaluations on controlled synthetic and benchmark datasets verified the fidelity of the generated feature attribution explanations.
Latent SHAP produced actionable, human-interpretable explanations across real-world vision, language, and cybersecurity domains lacking invertible transformation functions.

Cite This Study

Malach et al. (2025) studied this question.

synapsesocial.com/papers/6a0ebeeea14f152feaf9c8ae https://doi.org/10.1007/s10994-025-06852-8

Also Consider

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

View Full Paper

Ask AI

Bookmark