Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks

The present paper aims to extend the potential learning method to overcome the problem of collective interpretation, which aims to interpret multi-layered neural networks by compressing them into the simplest ones. In the process of compression, positive, negative, and complicated weights have had u...

Full description

Saved in:
Bibliographic Details
Main Authors: Ryotaro Kamimura, Haruhiko Takeuchi
Format: Article
Language:English
Published: Taylor & Francis Group 2020-04-01
Series:Connection Science
Subjects:
Online Access:http://dx.doi.org/10.1080/09540091.2019.1674245
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850225613560872960
author Ryotaro Kamimura
Haruhiko Takeuchi
author_facet Ryotaro Kamimura
Haruhiko Takeuchi
author_sort Ryotaro Kamimura
collection DOAJ
description The present paper aims to extend the potential learning method to overcome the problem of collective interpretation, which aims to interpret multi-layered neural networks by compressing them into the simplest ones. In the process of compression, positive, negative, and complicated weights have had unfavourable effects for interpretation. To deal with the problems of collective interpretation, the potential learning is extended only to use positive weights. In addition, to obtain more appropriate weights for interpretation, the number of candidate weights for higher potentialities is first increased as much as possible. Then, from among many candidates, more appropriate weights are selected as more important ones. This extended potentiality learning is expected to produce more stable and more simple representations for easy interpretation. The extended method was applied to three datasets, namely, an artificial dataset, a real eye-tracking dataset, and a student evaluation dataset. In all cases, it was observed that the selectivity of connection weights could be increased. Correspondingly, the majority of connection weights became positive, and the collective weights were quite similar to the regression coefficients of the logistic regression analysis. Finally, for the third dataset (student evaluations), the extended method could extract more explicit input-output relations, compared with the logistic regression analysis, while improving generalisation performance.
format Article
id doaj-art-518f9ea32f7d45a0a3a22a90df13cd8c
institution OA Journals
issn 0954-0091
1360-0494
language English
publishDate 2020-04-01
publisher Taylor & Francis Group
record_format Article
series Connection Science
spelling doaj-art-518f9ea32f7d45a0a3a22a90df13cd8c2025-08-20T02:05:18ZengTaylor & Francis GroupConnection Science0954-00911360-04942020-04-0132217420310.1080/09540091.2019.16742451674245Improving collective interpretation by extended potentiality assimilation for multi-layered neural networksRyotaro Kamimura0Haruhiko Takeuchi1Kumamoto Drone Technology and Development Foundation, Techno Research ParkNational Institute of Advanced Industrial Science and Technology (AIST)The present paper aims to extend the potential learning method to overcome the problem of collective interpretation, which aims to interpret multi-layered neural networks by compressing them into the simplest ones. In the process of compression, positive, negative, and complicated weights have had unfavourable effects for interpretation. To deal with the problems of collective interpretation, the potential learning is extended only to use positive weights. In addition, to obtain more appropriate weights for interpretation, the number of candidate weights for higher potentialities is first increased as much as possible. Then, from among many candidates, more appropriate weights are selected as more important ones. This extended potentiality learning is expected to produce more stable and more simple representations for easy interpretation. The extended method was applied to three datasets, namely, an artificial dataset, a real eye-tracking dataset, and a student evaluation dataset. In all cases, it was observed that the selectivity of connection weights could be increased. Correspondingly, the majority of connection weights became positive, and the collective weights were quite similar to the regression coefficients of the logistic regression analysis. Finally, for the third dataset (student evaluations), the extended method could extract more explicit input-output relations, compared with the logistic regression analysis, while improving generalisation performance.http://dx.doi.org/10.1080/09540091.2019.1674245neural networksgeneralizationautoencoderpotentialitycollective interpretation
spellingShingle Ryotaro Kamimura
Haruhiko Takeuchi
Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
Connection Science
neural networks
generalization
autoencoder
potentiality
collective interpretation
title Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
title_full Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
title_fullStr Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
title_full_unstemmed Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
title_short Improving collective interpretation by extended potentiality assimilation for multi-layered neural networks
title_sort improving collective interpretation by extended potentiality assimilation for multi layered neural networks
topic neural networks
generalization
autoencoder
potentiality
collective interpretation
url http://dx.doi.org/10.1080/09540091.2019.1674245
work_keys_str_mv AT ryotarokamimura improvingcollectiveinterpretationbyextendedpotentialityassimilationformultilayeredneuralnetworks
AT haruhikotakeuchi improvingcollectiveinterpretationbyextendedpotentialityassimilationformultilayeredneuralnetworks