著者

伊東 英彦 増田 一 大堀 隆文 渡辺 一央

出版者

一般社団法人 映像情報メディア学会

雑誌

テレビジョン学会技術報告 (ISSN:03864227)

巻号頁・発行日

vol.16, no.11, pp.37-42, 1992

---

This paper clarifies that a Multi-Tandem Perceptron (MTP) outputs the dominant K-L transform components of each sample of arbitrary input samples set. A R-stages MTP consists of R Unit-Perceptrons (UPs). Each UP is a 3-layerred linear perceptron with a hidden unit for identity mapping and transmits the mapping error vector to the following one. The hidden unit of the r-th UP outputs the r-th dominant component. This paper also discribes a rapidly learning method. The simulation results indicate that the method performs more than 100 times as rapidly as the conventional back propagation.