Influence Decompositions For Neural Network Attribution
Kyle Reing, Greg Ver Steeg, Aram Galstyan
Abstract
Methods of neural network attribution have emerged out of a necessity for explanation and accountability in the predictions of black-box neural models. Most approaches use a variation of sensitivity analysis, where individual input variables are perturbed and the downstream effects on some output metric are measured. We demonstrate that a number of critical functional properties are not revealed when only considering lower-order perturbations. Motivated by these shortcomings, we propose a general framework for decomposing the orders of influence that a collection of input variables has on an output classification. These orders are based on the cardinality of input subsets which are perturbed to yield a change in classification. This decomposition can be naturally applied to attribute which input variables rely on higher-order coordination to impact the classification decision. We demonstrate that our approach correctly identifies higher-order attribution on a number of synthetic examples. Additionally, we showcase the differences between attribution in our approach and existing approaches on benchmark networks for MNIST and ImageNet.
BibTeX
@InProceedings{pmlr-v130-reing21a,
title = { Influence Decompositions For Neural Network Attribution },
author = {Reing, Kyle and Ver Steeg, Greg and Galstyan, Aram},
booktitle = {Proceedings of The 24th International Conference on Artificial Intelligence and Statistics},
pages = {2710--2718},
year = {2021},
editor = {Banerjee, Arindam and Fukumizu, Kenji},
volume = {130},
series = {Proceedings of Machine Learning Research},
month = {13--15 Apr},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v130/reing21a/reing21a.pdf},
url = {https://proceedings.mlr.press/v130/reing21a.html},
abstract = { Methods of neural network attribution have emerged out of a necessity for explanation and accountability in the predictions of black-box neural models. Most approaches use a variation of sensitivity analysis, where individual input variables are perturbed and the downstream effects on some output metric are measured. We demonstrate that a number of critical functional properties are not revealed when only considering lower-order perturbations. Motivated by these shortcomings, we propose a general framework for decomposing the orders of influence that a collection of input variables has on an output classification. These orders are based on the cardinality of input subsets which are perturbed to yield a change in classification. This decomposition can be naturally applied to attribute which input variables rely on higher-order coordination to impact the classification decision. We demonstrate that our approach correctly identifies higher-order attribution on a number of synthetic examples. Additionally, we showcase the differences between attribution in our approach and existing approaches on benchmark networks for MNIST and ImageNet. }
}