Skip to Main content Skip to Navigation
Conference papers

Delving Deep into Interpreting Neural Nets with Piece-Wise Affine Representation

Abstract : Deep convolutional neural networks (CNNs) are now ubiquitous in computer vision problems. However, these models usually describe very complicated functions of the input images. For a number of application, it is of utmost importance to be able to explain the decisions of a network, e.g. by highlighting the most relevant pixels in an image or a feature map w.r.t. a particular class. In this paper, we show that CNNs locally describe piece-wise affine functions of each pixel, whose coefficient and bias can be retrieved analytically. We apply our methodology on several popular CNNs and draw interesting conclusions on the relative contributions of pixels and biases for these networks.
Complete list of metadatas

Cited literature [11 references]  Display  Hide  Download

https://hal.sorbonne-universite.fr/hal-02507512
Contributor : Yifu Chen <>
Submitted on : Friday, March 13, 2020 - 11:39:11 AM
Last modification on : Friday, January 8, 2021 - 5:36:04 PM
Long-term archiving on: : Sunday, June 14, 2020 - 1:33:18 PM

File

ICIP2019.pdf
Files produced by the author(s)

Identifiers

Citation

Yifu Chen, Antoine Saporta, Arnaud Dapogny, Matthieu Cord. Delving Deep into Interpreting Neural Nets with Piece-Wise Affine Representation. 2019 IEEE International Conference on Image Processing (ICIP), Sep 2019, Taipei, Taiwan. pp.609-613, ⟨10.1109/ICIP.2019.8803817⟩. ⟨hal-02507512⟩

Share

Metrics

Record views

344

Files downloads

917