Statistical Mechanical Analysis of Simultaneous Perturbation Learning

Seiji MIYOSHI  Hiroomi HIKAWA  Yutaka MAEDA  

IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E92-A   No.7   pp.1743-1746
Publication Date: 2009/07/01
Online ISSN: 1745-1337
DOI: 10.1587/transfun.E92.A.1743
Print ISSN: 0916-8508
Type of Manuscript: LETTER
Category: Neural Networks and Bioengineering
simultaneous perturbation,  on-line learning,  statistical mechanical method,  generalization error,  

Full Text: PDF>>
Buy this Article

We show that simultaneous perturbation can be used as an algorithm for on-line learning, and we report our theoretical investigation on generalization performance obtained with a statistical mechanical method. Asymptotic behavior of generalization error using this algorithm is on the order of t to the minus one-third power, where t is the learning time or the number of learning examples. This order is the same as that using well-known perceptron learning.