Presentation 2006-01-12
Design of Piecewise-Linear Neural Networks Generating Desired Periodic Sequences of Binary Vectors
Norikazu TAKAHASHI, Tsuyoshi YAMAKAWA, Tetsuo NISHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) How to design continuous-time neural networks generating prescribed sequences of binary vectors is studied in this report. It is assumed that state-output characteristic of each neuron is expressed by a piecewise linear function and that every two consecutive vectors in the sequence differ in exactly one component. We first analyze the state trajectories of such networks and derive a sufficient condition for the network to generate prescribed sequence of binary vectors. Since this sufficient condition is expressed in the form of linear inequalities, the network design can be formulated as an optimization problem having the sufficient condition as constraints. We will propose two design methods based on this approach : one is to minimize the sum of square of parameter values and the other is to maximize the robustness against parameter variations. Since the former and the latter are formulated as a quadratic programming problem and a linear programming problem, respectively, we can easily find parameter values by applying one of existing numerical solvers.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) neural networks / piecewise linear function / design / periodic sequence / optimization
Paper # CAS2005-82
Date of Issue

Conference Information
Committee CAS
Conference Date 2006/1/5(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 Circuits and Systems (CAS)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Design of Piecewise-Linear Neural Networks Generating Desired Periodic Sequences of Binary Vectors
Sub Title (in English)
Keyword(1) neural networks
Keyword(2) piecewise linear function
Keyword(3) design
Keyword(4) periodic sequence
Keyword(5) optimization
1st Author's Name Norikazu TAKAHASHI
1st Author's Affiliation Department of Computer Science and Communication Engineering, Kyushu University()
2nd Author's Name Tsuyoshi YAMAKAWA
2nd Author's Affiliation Department of Computer Science and Communication Engineering, Kyushu University:Nissan Motor Co., Ltd.
3rd Author's Name Tetsuo NISHI
3rd Author's Affiliation Faculty of Science and Engineering, Waseda University
Date 2006-01-12
Paper # CAS2005-82
Volume (vol) vol.105
Number (no) 503
Page pp.pp.-
#Pages 6
Date of Issue