Presentation 1996/7/24
A Study of Back Propagation Learning with Periodic Activation Function
Takashi Okada, Masahiro Nakagawa,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this report, we shall propose a back propagation learning model with a periodic activation function, and investigate the learning ability. The present neural network model is found to avoid an unfavorable trapping at local minima and remarkably assure the convergence to the global minimum as a periodic activation function, which promotes considerably the learning ability. In practice we shall investigate a 3-layered neural network model applied to the XOR problem. From the present simulation results, it is found that the learning speed can be relatively improved in comparison with the conventional back propagation learning model. In addition, it is also found that a chaotic noise in the learning scheme results in an advantage to the learning speed.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Periodic Activation Function / Chaos / Back Propagation / Learning Speed
Paper # NC96-23
Date of Issue

Conference Information
Committee NC
Conference Date 1996/7/24(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) A Study of Back Propagation Learning with Periodic Activation Function
Sub Title (in English)
Keyword(1) Periodic Activation Function
Keyword(2) Chaos
Keyword(3) Back Propagation
Keyword(4) Learning Speed
1st Author's Name Takashi Okada
1st Author's Affiliation Department of Electrical Engineering, Faculty of Engineering,Nagaoka University ofTechnohogy()
2nd Author's Name Masahiro Nakagawa
2nd Author's Affiliation Department of Electrical Engineering, Faculty of Engineering,Nagaoka University ofTechnohogy
Date 1996/7/24
Paper # NC96-23
Volume (vol) vol.96
Number (no) 178
Page pp.pp.-
#Pages 8
Date of Issue