Feature Attribution for Deep Learning Models through Total Variance Decomposition

Yinzhu Jin, Shen Zhu, Tom Fletcher

Keywords: feature attribution, explainable AI, counterfactual explanation, generative diffusion model

Abstract:

This paper introduces a new approach to feature attribution for deep learning models, quantifying the importance of specific features in model decisions. By decomposing the total variance of model decisions into explained and unexplained fractions, conditioned on the target feature, we define the feature attribution score as the proportion of explained variance. This method offers a solid statistical foundation and normalized quantitative results. When ample data is available, we compute the score directly from test data. For scarce data, we use constrained sampling with generative diffusion models to represent the conditional distribution at a given feature value. We demonstrate the method’s effectiveness on both a synthetic image dataset with known ground truth and OASIS-3 brain MRIs.

Primary Subject Area: Interpretability and Explainable AI

Secondary Subject Area: Interpretability and Explainable AI

Paper Type: Methodological Development

Registration Requirement: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 239