Time-series attribution maps with regularized contrastive learning

Steffen Schneider, Rodrigo González Laiz, Anastasiia Filippova, Markus Frey, Mackenzie W Mathis
Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, PMLR 258:2521-2529, 2025.

Abstract

Gradient-based attribution methods aim to explain decisions of deep learning models but so far lack identifiability guarantees. Here, we propose a method to generate attribution maps with identifiability guarantees by developing a regularized contrastive learning algorithm trained on time-series data plus a new attribution method called Inverted Neuron Gradient (collectively named xCEBRA). We show theoretically that xCEBRA has favorable properties for identifying the Jacobian matrix of the data generating process. Empirically, we demonstrate robust approximation of zero vs. non-zero entries in the ground-truth attribution map on synthetic datasets, and significant improvements across previous attribution methods based on feature ablation, Shapley values, and other gradient-based methods. Our work constitutes a first example of identifiable inference of time-series attribution maps and opens avenues to a better understanding of time-series data, such as for neural dynamics and decision-processes within neural networks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v258-schneider25a, title = {Time-series attribution maps with regularized contrastive learning}, author = {Schneider, Steffen and Laiz, Rodrigo Gonz{\'a}lez and Filippova, Anastasiia and Frey, Markus and Mathis, Mackenzie W}, booktitle = {Proceedings of The 28th International Conference on Artificial Intelligence and Statistics}, pages = {2521--2529}, year = {2025}, editor = {Li, Yingzhen and Mandt, Stephan and Agrawal, Shipra and Khan, Emtiyaz}, volume = {258}, series = {Proceedings of Machine Learning Research}, month = {03--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v258/main/assets/schneider25a/schneider25a.pdf}, url = {https://proceedings.mlr.press/v258/schneider25a.html}, abstract = {Gradient-based attribution methods aim to explain decisions of deep learning models but so far lack identifiability guarantees. Here, we propose a method to generate attribution maps with identifiability guarantees by developing a regularized contrastive learning algorithm trained on time-series data plus a new attribution method called Inverted Neuron Gradient (collectively named xCEBRA). We show theoretically that xCEBRA has favorable properties for identifying the Jacobian matrix of the data generating process. Empirically, we demonstrate robust approximation of zero vs. non-zero entries in the ground-truth attribution map on synthetic datasets, and significant improvements across previous attribution methods based on feature ablation, Shapley values, and other gradient-based methods. Our work constitutes a first example of identifiable inference of time-series attribution maps and opens avenues to a better understanding of time-series data, such as for neural dynamics and decision-processes within neural networks.} }
Endnote
%0 Conference Paper %T Time-series attribution maps with regularized contrastive learning %A Steffen Schneider %A Rodrigo González Laiz %A Anastasiia Filippova %A Markus Frey %A Mackenzie W Mathis %B Proceedings of The 28th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2025 %E Yingzhen Li %E Stephan Mandt %E Shipra Agrawal %E Emtiyaz Khan %F pmlr-v258-schneider25a %I PMLR %P 2521--2529 %U https://proceedings.mlr.press/v258/schneider25a.html %V 258 %X Gradient-based attribution methods aim to explain decisions of deep learning models but so far lack identifiability guarantees. Here, we propose a method to generate attribution maps with identifiability guarantees by developing a regularized contrastive learning algorithm trained on time-series data plus a new attribution method called Inverted Neuron Gradient (collectively named xCEBRA). We show theoretically that xCEBRA has favorable properties for identifying the Jacobian matrix of the data generating process. Empirically, we demonstrate robust approximation of zero vs. non-zero entries in the ground-truth attribution map on synthetic datasets, and significant improvements across previous attribution methods based on feature ablation, Shapley values, and other gradient-based methods. Our work constitutes a first example of identifiable inference of time-series attribution maps and opens avenues to a better understanding of time-series data, such as for neural dynamics and decision-processes within neural networks.
APA
Schneider, S., Laiz, R.G., Filippova, A., Frey, M. & Mathis, M.W.. (2025). Time-series attribution maps with regularized contrastive learning. Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 258:2521-2529 Available from https://proceedings.mlr.press/v258/schneider25a.html.

Related Material