SPADE: Sparsity-Guided Debugging for Deep Neural Networks

Arshia Soltani Moakhar, Eugenia Iofinova, Elias Frantar, Dan Alistarh
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:45955-45987, 2024.

Abstract

It is known that sparsity can improve interpretability for deep neural networks. However, existing methods in the area either require networks that are pre-trained with sparsity constraints, or impose sparsity after the fact, altering the network’s general behavior. In this paper, we demonstrate, for the first time, that sparsity can instead be incorporated into the interpretation process itself, as a sample-specific preprocessing step. Unlike previous work, this approach, which we call SPADE, does not place constraints on the trained model and does not affect its behavior during inference on the sample. Given a trained model and a target sample, SPADE uses sample-targeted pruning to provide a "trace" of the network’s execution on the sample, reducing the network to the most important connections prior to computing an interpretation. We demonstrate that preprocessing with SPADE significantly increases the accuracy of image saliency maps across several interpretability methods. Additionally, SPADE improves the usefulness of neuron visualizations, aiding humans in reasoning about network behavior. Our code is available at https://github.com/IST-DASLab/SPADE.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-soltani-moakhar24a, title = {{SPADE}: Sparsity-Guided Debugging for Deep Neural Networks}, author = {Soltani Moakhar, Arshia and Iofinova, Eugenia and Frantar, Elias and Alistarh, Dan}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {45955--45987}, 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/soltani-moakhar24a/soltani-moakhar24a.pdf}, url = {https://proceedings.mlr.press/v235/soltani-moakhar24a.html}, abstract = {It is known that sparsity can improve interpretability for deep neural networks. However, existing methods in the area either require networks that are pre-trained with sparsity constraints, or impose sparsity after the fact, altering the network’s general behavior. In this paper, we demonstrate, for the first time, that sparsity can instead be incorporated into the interpretation process itself, as a sample-specific preprocessing step. Unlike previous work, this approach, which we call SPADE, does not place constraints on the trained model and does not affect its behavior during inference on the sample. Given a trained model and a target sample, SPADE uses sample-targeted pruning to provide a "trace" of the network’s execution on the sample, reducing the network to the most important connections prior to computing an interpretation. We demonstrate that preprocessing with SPADE significantly increases the accuracy of image saliency maps across several interpretability methods. Additionally, SPADE improves the usefulness of neuron visualizations, aiding humans in reasoning about network behavior. Our code is available at https://github.com/IST-DASLab/SPADE.} }
Endnote
%0 Conference Paper %T SPADE: Sparsity-Guided Debugging for Deep Neural Networks %A Arshia Soltani Moakhar %A Eugenia Iofinova %A Elias Frantar %A Dan Alistarh %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-soltani-moakhar24a %I PMLR %P 45955--45987 %U https://proceedings.mlr.press/v235/soltani-moakhar24a.html %V 235 %X It is known that sparsity can improve interpretability for deep neural networks. However, existing methods in the area either require networks that are pre-trained with sparsity constraints, or impose sparsity after the fact, altering the network’s general behavior. In this paper, we demonstrate, for the first time, that sparsity can instead be incorporated into the interpretation process itself, as a sample-specific preprocessing step. Unlike previous work, this approach, which we call SPADE, does not place constraints on the trained model and does not affect its behavior during inference on the sample. Given a trained model and a target sample, SPADE uses sample-targeted pruning to provide a "trace" of the network’s execution on the sample, reducing the network to the most important connections prior to computing an interpretation. We demonstrate that preprocessing with SPADE significantly increases the accuracy of image saliency maps across several interpretability methods. Additionally, SPADE improves the usefulness of neuron visualizations, aiding humans in reasoning about network behavior. Our code is available at https://github.com/IST-DASLab/SPADE.
APA
Soltani Moakhar, A., Iofinova, E., Frantar, E. & Alistarh, D.. (2024). SPADE: Sparsity-Guided Debugging for Deep Neural Networks. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:45955-45987 Available from https://proceedings.mlr.press/v235/soltani-moakhar24a.html.

Related Material