Presentation 2007-06-15
Mixture of RNN experts for learning of temporal sequence given by rule dynamics
Jun NAMIKAWA, Jun TANI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper proposes a learning method of the "mixture of experts" type model, which can acquire the ability to generate desired sequences as switching functions governing change of states. Our method is similar to Tani and Nolfi (1991) in that both are maximum likelihood estimation based on gradient descent algorithm, though the likelihood function is different. We first show a numerical simulation in which the model can learn Markov chain switching nine Lissajous curves-using our method. Furthermore, we numerically examine generalization and training error to compare conventional method and proposed method. The simulation results shows that our method improves learning performance of the model.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) recurrent neural network / mixture of experts architecture / maximum likelihood estimation / rule dynamics
Paper # NC2007-16
Date of Issue

Conference Information
Committee NC
Conference Date 2007/6/7(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) Mixture of RNN experts for learning of temporal sequence given by rule dynamics
Sub Title (in English)
Keyword(1) recurrent neural network
Keyword(2) mixture of experts architecture
Keyword(3) maximum likelihood estimation
Keyword(4) rule dynamics
1st Author's Name Jun NAMIKAWA
1st Author's Affiliation RIKEN Brain Science Institute()
2nd Author's Name Jun TANI
2nd Author's Affiliation RIKEN Brain Science Institute
Date 2007-06-15
Paper # NC2007-16
Volume (vol) vol.107
Number (no) 92
Page pp.pp.-
#Pages 6
Date of Issue