ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS

ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS

In this paper we propose a new technic called etalons, which allows us to interpret the way how convolution network makes its predictions. This mechanism is very similar to voting among different experts. Thereby CNN could be interpreted as a variety of experts, but it acts not like a sum or product...

Full description

Bibliographic Details
Main Authors: V. V. Molchanov, B. V. Vishnyakov, V. S. Gorbatsevich, Y. V. Vizilter
Format: Article
Language:English
Published: Copernicus Publications 2018-05-01
Series:The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Online Access:https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLII-2/707/2018/isprs-archives-XLII-2-707-2018.pdf
id doaj-41eb0c074bdd4fc7a58bb3fa875c9fda
record_format Article
spelling doaj-41eb0c074bdd4fc7a58bb3fa875c9fda2020-11-25T01:01:32ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342018-05-01XLII-270771410.5194/isprs-archives-XLII-2-707-2018ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKSV. V. Molchanov0B. V. Vishnyakov1V. S. Gorbatsevich2Y. V. Vizilter3FGUP «State Research Institute of Aviation Systems», 125319, Moscow, Viktorenko street, 7, RussiaFGUP «State Research Institute of Aviation Systems», 125319, Moscow, Viktorenko street, 7, RussiaFGUP «State Research Institute of Aviation Systems», 125319, Moscow, Viktorenko street, 7, RussiaFGUP «State Research Institute of Aviation Systems», 125319, Moscow, Viktorenko street, 7, RussiaIn this paper we propose a new technic called etalons, which allows us to interpret the way how convolution network makes its predictions. This mechanism is very similar to voting among different experts. Thereby CNN could be interpreted as a variety of experts, but it acts not like a sum or product of them, but rather represent a complicated hierarchy. We implement algorithm for etalon acquisition based on well-known properties of affine maps. We show that neural net has two high-level mechanisms of voting: first, based on attention to input image regions, specific to current input, and second, based on ignoring specific input regions. We also make an assumption that there is a connection between complexity of the underlying data manifold and the number of etalon images and their quality.https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLII-2/707/2018/isprs-archives-XLII-2-707-2018.pdf
collection DOAJ
language English
format Article
sources DOAJ
author V. V. Molchanov
B. V. Vishnyakov
V. S. Gorbatsevich
Y. V. Vizilter
spellingShingle V. V. Molchanov
B. V. Vishnyakov
V. S. Gorbatsevich
Y. V. Vizilter
ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
author_facet V. V. Molchanov
B. V. Vishnyakov
V. S. Gorbatsevich
Y. V. Vizilter
author_sort V. V. Molchanov
title ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
title_short ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
title_full ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
title_fullStr ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
title_full_unstemmed ETALON IMAGES: UNDERSTANDING THE CONVOLUTION NEURAL NETWORKS
title_sort etalon images: understanding the convolution neural networks
publisher Copernicus Publications
series The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
issn 1682-1750
2194-9034
publishDate 2018-05-01
description In this paper we propose a new technic called etalons, which allows us to interpret the way how convolution network makes its predictions. This mechanism is very similar to voting among different experts. Thereby CNN could be interpreted as a variety of experts, but it acts not like a sum or product of them, but rather represent a complicated hierarchy. We implement algorithm for etalon acquisition based on well-known properties of affine maps. We show that neural net has two high-level mechanisms of voting: first, based on attention to input image regions, specific to current input, and second, based on ignoring specific input regions. We also make an assumption that there is a connection between complexity of the underlying data manifold and the number of etalon images and their quality.
url https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLII-2/707/2018/isprs-archives-XLII-2-707-2018.pdf
work_keys_str_mv AT vvmolchanov etalonimagesunderstandingtheconvolutionneuralnetworks
AT bvvishnyakov etalonimagesunderstandingtheconvolutionneuralnetworks
AT vsgorbatsevich etalonimagesunderstandingtheconvolutionneuralnetworks
AT yvvizilter etalonimagesunderstandingtheconvolutionneuralnetworks
_version_ 1725208856336793600

Similar Items