The Application of Genetic Algorithms to Artificial Neural Networks

• Main Authors: Jyh-Iuan Sheu, 許智淵
• Other Authors: Dartzi Pan
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/39295925844029378815

碩士 === 國立成功大學 === 航空太空工程學系 === 88 === The objective of this research is to apply GA(Genetic Algorithms) to the learning of ANN(Artificial Neural Network). When the network structure has been determined, the capability of a particular ANN depends on the learning rule it adapts. One of the most commonly used learning rule is the Back Propagation(BP) method. It is well known that BP is a steepest descent method that relies on the system gradient. It does not guarautee to reach the system global optimal. On the other hand, Genetic Algorithms(GA) are based on the simulation of natural evolution process. It follows the Darwin''s principal of "Survival of the fittest", which always searchs for the global optimal under the system constraint. In this steady, a hybird method of BP and GA will be developed to imrove the robustness of learning rule for ANN.