Max-min convolutional neural networks for image classification

Michael Blot 1 Matthieu Cord 1 Nicolas Thome 1
1 MLIA - Machine Learning and Information Access
LIP6 - Laboratoire d'Informatique de Paris 6
Abstract : Convolutional neural networks (CNN) are widely used in computer vision, especially in image classification. However, the way in which information and invariance properties are encoded through in deep CNN architectures is still an open question. In this paper, we propose to modify the standard convo-lutional block of CNN in order to transfer more information layer after layer while keeping some invariance within the network. Our main idea is to exploit both positive and negative high scores obtained in the convolution maps. This behavior is obtained by modifying the traditional activation function step before pooling. We are doubling the maps with specific activations functions, called MaxMin strategy, in order to achieve our pipeline. Extensive experiments on two classical datasets, MNIST and CIFAR-10, show that our deep MaxMin convolutional net outperforms standard CNN.
Liste complète des métadonnées

Cited literature [24 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01372216
Contributor : Michael Blot <>
Submitted on : Tuesday, October 4, 2016 - 1:43:25 PM
Last modification on : Thursday, March 21, 2019 - 2:33:07 PM
Document(s) archivé(s) le : Thursday, January 5, 2017 - 12:57:37 PM

File

article.pdf
Files produced by the author(s)

Licence


Distributed under a Creative Commons Attribution 4.0 International License

Identifiers

Citation

Michael Blot, Matthieu Cord, Nicolas Thome. Max-min convolutional neural networks for image classification. ICIP 2016 - IEEE International Conference on Image Processing, Sep 2016, Phoenix, United States. pp.3678-3682, ⟨10.1109/ICIP.2016.7533046⟩. ⟨hal-01372216⟩

Share

Metrics

Record views

384

Files downloads

459