Skip to content
Keras implementation for DASP: Deep Approximate Shapley Propagation (ICML 2019)
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
dasp Fix wrong computation on mean and variance in Dense input. Closes #1 Jun 21, 2019
examples Delete cifar10.py Mar 25, 2019
supplementary included supplementary material as PDF Apr 12, 2019
tests Working on Feb 4, 2019
utils Parkinson example Feb 19, 2019
.gitignore included supplementary material as PDF Apr 12, 2019
README.md Update README.md Jun 7, 2019

README.md

Deep Approximate Shapley Propagation

This repository contains the Keras implementation for Deep Approximate Shapley Propagation (DASP), "Explaining Deep Neural Networks with a Polynomial Time Algorithm for Shapley Values Approximation", ICML 2019.

The problem of explaining the behavior of deep neural networks has gained a lot of attention over the last years. Shapley values as a unique way of assigning relevance scores such that certain desirable properties are satisfied. Unfortunately, the exact evaluation of Shapley values is prohibitively expensive, exponential in the number of input features. DASP is a polynomial-time approximation of Shapley values designed for deep neural networks. It relies on uncertainty propagation using Lightweight Probabilistic Deep Networks (LPDN) to approximate Shapley values. This libraries relies on a Keras implementation of such probabilistic framework.

How to use

Please see the example folder for practical usage example and results.

Citation

@InProceedings{ancona19a,
  title = 	 {Explaining Deep Neural Networks with a Polynomial Time Algorithm for Shapley Value Approximation},
  author = 	 {Ancona, Marco and Oztireli, Cengiz and Gross, Markus},
  booktitle = 	 {Proceedings of the 36th International Conference on Machine Learning},
  pages = 	 {272--281},
  year = 	 {2019},
  editor = 	 {Chaudhuri, Kamalika and Salakhutdinov, Ruslan},
  volume = 	 {97},
  series = 	 {Proceedings of Machine Learning Research},
  address = 	 {Long Beach, California, USA},
  month = 	 {09--15 Jun},
  publisher = 	 {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v97/ancona19a/ancona19a.pdf},
  url = 	 {http://proceedings.mlr.press/v97/ancona19a.html},
  abstract = 	 {The problem of explaining the behavior of deep neural networks has recently gained a lot of attention. While several attribution methods have been proposed, most come without strong theoretical foundations, which raises questions about their reliability. On the other hand, the literature on cooperative game theory suggests Shapley values as a unique way of assigning relevance scores such that certain desirable properties are satisfied. Unfortunately, the exact evaluation of Shapley values is prohibitively expensive, exponential in the number of input features. In this work, by leveraging recent results on uncertainty propagation, we propose a novel, polynomial-time approximation of Shapley values in deep neural networks. We show that our method produces significantly better approximations of Shapley values than existing state-of-the-art attribution methods.}
}

License

MIT

You can’t perform that action at this time.