Presentation 2014-12-13
Consideration of Dynamic Binary Neural Networks based on the Feature Quantity Plane
Ryuji SATO, Jungo MORIYASU, Toshimichi SAITO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper studies learning the 2-layer dynamic binary neural network that can generate various binary periodic orbit. The learning aims at storage of a desired periodic orbit (or fixed point) and stabilization of the stored periodic orbits. For the storage, we apply a simple correlation-based learning algorithm. For the stabilization, we sparsify the network connection. In order to evaluate the network behavior, we introduce feature quantities of steady states (periodic orbits) and transient states (eventually periodic points). On the plane of the two feature quantities, the network dynamics can be visualized. Performing basic numerical experiments, we investigate the effectiveness of learning algorithm and feature quantities plane.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Binary Neural Networks / Correlation Learning / Sparsification
Paper # NC2014-51
Date of Issue

Conference Information
Committee NC
Conference Date 2014/12/6(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Consideration of Dynamic Binary Neural Networks based on the Feature Quantity Plane
Sub Title (in English)
Keyword(1) Binary Neural Networks
Keyword(2) Correlation Learning
Keyword(3) Sparsification
1st Author's Name Ryuji SATO
1st Author's Affiliation EE Dept., Hosei University()
2nd Author's Name Jungo MORIYASU
2nd Author's Affiliation EE Dept., Hosei University
3rd Author's Name Toshimichi SAITO
3rd Author's Affiliation EE Dept., Hosei University
Date 2014-12-13
Paper # NC2014-51
Volume (vol) vol.114
Number (no) 362
Page pp.pp.-
#Pages 5
Date of Issue