Pruning of recurrent neural models: an optimal brain damage approach

Patryk Chaber , Maciej Ławryńczuk

Abstract

This paper considers the problem of prun- ing recurrent neural models of perceptron type with one hidden layer which may be used for modelling of dynamic system. In order to reduce the number of model parameters (i.e. the number of weights), the Optimal Brain Damage (OBD) pruning algorithm is adopted for the recurrent neural models. Efficiency of the OBD algorithm is demonstrated for pruning neural models of a neutralisation reactor benchmark process. For the considered neutralisation system, the OBD algorithm makes it possible to reduce as many as 60% of model parameters and reduce the validation error by some 30% when compared to the full (unpruned) models
Author Patryk Chaber (FEIT / AK)
Patryk Chaber,,
- The Institute of Control and Computation Engineering
, Maciej Ławryńczuk (FEIT / AK)
Maciej Ławryńczuk,,
- The Institute of Control and Computation Engineering
Journal seriesNonlinear Dynamics, ISSN 0924-090X, e-ISSN 1573-269X, (A 40 pkt)
Issue year2018
Vol92
No2
Pages763-780
Publication size in sheets0.85
Keywords in EnglishNeural networks · Dynamic systemsModel pruning · Model structure optimisation
ASJC Classification2208 Electrical and Electronic Engineering; 2604 Applied Mathematics; 2210 Mechanical Engineering; 2212 Ocean Engineering; 2202 Aerospace Engineering; 2207 Control and Systems Engineering
DOIDOI:10.1007/s11071-018-4089-1
URL https://link.springer.com/content/pdf/10.1007/s11071-018-4089-1.pdf
projectDevelopment of methodology of control, decision support and production management. Project leader: Ogryczak Włodzimierz, , Phone: 6190, start date 12-06-2017, end date 31-12-2018, 504/statut2017/1031, Completed
WEiTI Działalność statutowa
Languageen angielski
File
Chaber Lawr Nonlinear Dyn 18.pdf 1.42 MB
Score (nominal)40
ScoreMinisterial score = 40.0, 10-07-2019, ArticleFromJournal
Publication indicators Scopus Citations = 0; WoS Citations = 0; Scopus SNIP (Source Normalised Impact per Paper): 2016 = 1.523; WoS Impact Factor: 2017 = 4.339 (2) - 2017=3.906 (5)
Citation count*1 (2019-07-29)
Cite
Share Share

Get link to the record


* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.
Back