Learning to Explain: An Information-Theoretic Perspective on Model Interpretation

Jianbo Chen, Le Song, Martin Wainwright, Michael Jordan
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:883-892, 2018.

Abstract

We introduce instancewise feature selection as a methodology for model interpretation. Our method is based on learning a function to extract a subset of features that are most informative for each given example. This feature selector is trained to maximize the mutual information between selected features and the response variable, where the conditional distribution of the response variable given the input is the model to be explained. We develop an efficient variational approximation to the mutual information, and show the effectiveness of our method on a variety of synthetic and real data sets using both quantitative metrics and human evaluation.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-chen18j, title = {Learning to Explain: An Information-Theoretic Perspective on Model Interpretation}, author = {Chen, Jianbo and Song, Le and Wainwright, Martin and Jordan, Michael}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {883--892}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/chen18j/chen18j.pdf}, url = {https://proceedings.mlr.press/v80/chen18j.html}, abstract = {We introduce instancewise feature selection as a methodology for model interpretation. Our method is based on learning a function to extract a subset of features that are most informative for each given example. This feature selector is trained to maximize the mutual information between selected features and the response variable, where the conditional distribution of the response variable given the input is the model to be explained. We develop an efficient variational approximation to the mutual information, and show the effectiveness of our method on a variety of synthetic and real data sets using both quantitative metrics and human evaluation.} }
Endnote
%0 Conference Paper %T Learning to Explain: An Information-Theoretic Perspective on Model Interpretation %A Jianbo Chen %A Le Song %A Martin Wainwright %A Michael Jordan %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-chen18j %I PMLR %P 883--892 %U https://proceedings.mlr.press/v80/chen18j.html %V 80 %X We introduce instancewise feature selection as a methodology for model interpretation. Our method is based on learning a function to extract a subset of features that are most informative for each given example. This feature selector is trained to maximize the mutual information between selected features and the response variable, where the conditional distribution of the response variable given the input is the model to be explained. We develop an efficient variational approximation to the mutual information, and show the effectiveness of our method on a variety of synthetic and real data sets using both quantitative metrics and human evaluation.
APA
Chen, J., Song, L., Wainwright, M. & Jordan, M.. (2018). Learning to Explain: An Information-Theoretic Perspective on Model Interpretation. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:883-892 Available from https://proceedings.mlr.press/v80/chen18j.html.

Related Material