Explaining nonlinear classification decisions with deep Taylor decomposition

Citation data:

Pattern Recognition, ISSN: 0031-3203, Vol: 65, Page: 211-222

Publication Year:
2017
Usage 210
Abstract Views 133
Clicks 72
Link-outs 5
Captures 388
Readers 387
Exports-Saves 1
Mentions 3
Comments 1
Blog Mentions 1
Q&A Site Mentions 1
Social Media 88
Shares, Likes & Comments 62
Tweets 26
Citations 44
Citation Indexes 44
Ratings
Reddit 1
arXiv Id:
1512.02479
DOI:
10.1016/j.patcog.2016.11.008
Repository URL:
http://arxiv.org/abs/1512.02479
Author(s):
Montavon, Grégoire; Bach, Sebastian; Binder, Alexander; Samek, Wojciech; Müller, Klaus-Robert
Publisher(s):
Elsevier BV
Tags:
Computer Science; Computer Science - Machine Learning; Statistics - Machine Learning
Most Recent Tweet View All Tweets
Most Recent Blog Mention
article description
Nonlinear methods such as Deep Neural Networks (DNNs) are the gold standard for various challenging machine learning problems such as image recognition. Although these methods perform impressively well, they have a significant disadvantage, the lack of transparency, limiting the interpretability of the solution and thus the scope of application in practice. Especially DNNs act as black boxes due to their multilayer nonlinear structure. In this paper we introduce a novel methodology for interpreting generic multilayer neural networks by decomposing the network classification decision into contributions of its input elements. Although our focus is on image classification, the method is applicable to a broad set of input data, learning tasks and network architectures. Our method called deep Taylor decomposition efficiently utilizes the structure of the network by backpropagating the explanations from the output to the input layer. We evaluate the proposed method empirically on the MNIST and ILSVRC data sets.