Presentation 1996/6/21
Annealed RNN Learning of Finite State Automata
Ken-ichi Arai, Ryohei Nakano,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In recurrent neural network (RNN) learning of finite state automata (FSA), we discuss how a neuro gain (J) influences the stability of the state representation and the performance of the learning. We formally show that the existence of the critical neuro gain (β_0) : any β larger than β_0 makes an RNN maintain the stable representation of states of an acquired FSA. Considering the existence of β_0 and avoidance of local minima, we propose a new RNN learning method with the scheduling of β, called an annealed RNN learning. Our experiments show that the annealed RNN learning went beyond a constant β learning.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Recurrent Neural Network / Finite State Automata / Stability / annealing
Paper # NC96-12
Date of Issue

Conference Information
Committee NC
Conference Date 1996/6/21(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) Annealed RNN Learning of Finite State Automata
Sub Title (in English)
Keyword(1) Recurrent Neural Network
Keyword(2) Finite State Automata
Keyword(3) Stability
Keyword(4) annealing
1st Author's Name Ken-ichi Arai
1st Author's Affiliation NTT Communication Science Laboratories()
2nd Author's Name Ryohei Nakano
2nd Author's Affiliation NTT Communication Science Laboratories
Date 1996/6/21
Paper # NC96-12
Volume (vol) vol.96
Number (no) 117
Page pp.pp.-
#Pages 8
Date of Issue