Prospector Heads: Generalized Feature Attribution for Large Models & Data

Gautam Machiraju, Alexander Derry, Arjun D Desai, Neel Guha, Amir-Hossein Karimi, James Zou, Russ B Altman, Christopher Re, Parag Mallick
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:34115-34144, 2024.

Abstract

Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for ML models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based attribution methods that can be applied to any encoder and any data modality. Prospector heads generalize across modalities through experiments on sequences (text), images (pathology), and graphs (protein structures), outperforming baseline attribution methods by up to 26.3 points in mean localization AUPRC. We also demonstrate how prospector heads enable improved interpretation and discovery of class-specific patterns in input data. Through their high performance, flexibility, and generalizability, prospectors provide a framework for improving trust and transparency for ML models in complex domains.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-machiraju24a, title = {Prospector Heads: Generalized Feature Attribution for Large Models & Data}, author = {Machiraju, Gautam and Derry, Alexander and Desai, Arjun D and Guha, Neel and Karimi, Amir-Hossein and Zou, James and Altman, Russ B and Re, Christopher and Mallick, Parag}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {34115--34144}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/machiraju24a/machiraju24a.pdf}, url = {https://proceedings.mlr.press/v235/machiraju24a.html}, abstract = {Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for ML models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based attribution methods that can be applied to any encoder and any data modality. Prospector heads generalize across modalities through experiments on sequences (text), images (pathology), and graphs (protein structures), outperforming baseline attribution methods by up to 26.3 points in mean localization AUPRC. We also demonstrate how prospector heads enable improved interpretation and discovery of class-specific patterns in input data. Through their high performance, flexibility, and generalizability, prospectors provide a framework for improving trust and transparency for ML models in complex domains.} }
Endnote
%0 Conference Paper %T Prospector Heads: Generalized Feature Attribution for Large Models & Data %A Gautam Machiraju %A Alexander Derry %A Arjun D Desai %A Neel Guha %A Amir-Hossein Karimi %A James Zou %A Russ B Altman %A Christopher Re %A Parag Mallick %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-machiraju24a %I PMLR %P 34115--34144 %U https://proceedings.mlr.press/v235/machiraju24a.html %V 235 %X Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for ML models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based attribution methods that can be applied to any encoder and any data modality. Prospector heads generalize across modalities through experiments on sequences (text), images (pathology), and graphs (protein structures), outperforming baseline attribution methods by up to 26.3 points in mean localization AUPRC. We also demonstrate how prospector heads enable improved interpretation and discovery of class-specific patterns in input data. Through their high performance, flexibility, and generalizability, prospectors provide a framework for improving trust and transparency for ML models in complex domains.
APA
Machiraju, G., Derry, A., Desai, A.D., Guha, N., Karimi, A., Zou, J., Altman, R.B., Re, C. & Mallick, P.. (2024). Prospector Heads: Generalized Feature Attribution for Large Models & Data. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:34115-34144 Available from https://proceedings.mlr.press/v235/machiraju24a.html.

Related Material