Neural Network Attributions: A Causal Perspective

Aditya Chattopadhyay, Piyushi Manupriya, Anirban Sarkar, Vineeth N Balasubramanian
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:981-990, 2019.

Abstract

We propose a new attribution method for neural networks developed using first principles of causality (to the best of our knowledge, the first such). The neural network architecture is viewed as a Structural Causal Model, and a methodology to compute the causal effect of each feature on the output is presented. With reasonable assumptions on the causal structure of the input data, we propose algorithms to efficiently compute the causal effects, as well as scale the approach to data with large dimensionality. We also show how this method can be used for recurrent neural networks. We report experimental results on both simulated and real datasets showcasing the promise and usefulness of the proposed algorithm.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-chattopadhyay19a, title = {Neural Network Attributions: A Causal Perspective}, author = {Chattopadhyay, Aditya and Manupriya, Piyushi and Sarkar, Anirban and Balasubramanian, Vineeth N}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {981--990}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/chattopadhyay19a/chattopadhyay19a.pdf}, url = {https://proceedings.mlr.press/v97/chattopadhyay19a.html}, abstract = {We propose a new attribution method for neural networks developed using first principles of causality (to the best of our knowledge, the first such). The neural network architecture is viewed as a Structural Causal Model, and a methodology to compute the causal effect of each feature on the output is presented. With reasonable assumptions on the causal structure of the input data, we propose algorithms to efficiently compute the causal effects, as well as scale the approach to data with large dimensionality. We also show how this method can be used for recurrent neural networks. We report experimental results on both simulated and real datasets showcasing the promise and usefulness of the proposed algorithm.} }
Endnote
%0 Conference Paper %T Neural Network Attributions: A Causal Perspective %A Aditya Chattopadhyay %A Piyushi Manupriya %A Anirban Sarkar %A Vineeth N Balasubramanian %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-chattopadhyay19a %I PMLR %P 981--990 %U https://proceedings.mlr.press/v97/chattopadhyay19a.html %V 97 %X We propose a new attribution method for neural networks developed using first principles of causality (to the best of our knowledge, the first such). The neural network architecture is viewed as a Structural Causal Model, and a methodology to compute the causal effect of each feature on the output is presented. With reasonable assumptions on the causal structure of the input data, we propose algorithms to efficiently compute the causal effects, as well as scale the approach to data with large dimensionality. We also show how this method can be used for recurrent neural networks. We report experimental results on both simulated and real datasets showcasing the promise and usefulness of the proposed algorithm.
APA
Chattopadhyay, A., Manupriya, P., Sarkar, A. & Balasubramanian, V.N.. (2019). Neural Network Attributions: A Causal Perspective. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:981-990 Available from https://proceedings.mlr.press/v97/chattopadhyay19a.html.

Related Material