A simple genetic approach to adaptive processing of tree-structure patterns

Siu Yeung Cho, Zheru Chi

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

Abstract

This paper describes a learning scheme of a structural connectionist architecture based on a simple genetic approach to adaptive processing of tree-structures representation. Conventionally, one of the most popular supervised learning formulations of tree-structures processing is Backpropagation Through Structures (BPTS) [1]. The BPTS algorithm has been successfully applied to a number of learning tasks that involved complex symbolic structural patterns such as image semantic, internet behaviour, and chemical compound. However, this BPTS typed algorithm suffers from the long-term dependency problem in learning very deep tree structures. In this paper, we propose a simple genetic evolution approach for this processing. The idea of this algorithm is to tune the learning parameters by the genetic evolution with specified binary chromosome structures. Experimental results significantly support the capabilities of our proposed approach to classify and recognize structural patterns in terms of generalization capability.
Original languageEnglish
Title of host publicationProceedings of the International Conference on Artificial Intelligence, IC-AI'04 and Proceedings of the International Conference on Machine Learning; Models, Technologies and Applications, MLMTA'04)
Pages555-561
Number of pages7
Volume2
Publication statusPublished - 1 Dec 2004
EventProceedings of the International Conference on Artificial Intelligence, IC-AI'04 - Las Vegas, NV, United States
Duration: 21 Jun 200424 Jun 2004

Conference

ConferenceProceedings of the International Conference on Artificial Intelligence, IC-AI'04
CountryUnited States
CityLas Vegas, NV
Period21/06/0424/06/04

ASJC Scopus subject areas

  • Engineering(all)

Cite this