A possible genetic-algorithm based method for optimizing a class of ANN transfer functions

M.P. Beddoess1, R.K. Ward2
1Department of ECE, UBC-TRIUMF, Vancouver, Canada
2Elect. Engineering and CICR, UBC-TRIUMF, Vancouver, Canada

Tóm tắt

This paper proposes a hybrid of two methods to determine the weight-constants in a class of artificial neuron networks, ANNs. The class of ANNs we are interested in are characterized by feed-forward processing elements. One of the methods is the genetic algorithm, GA; the other is "training through" back-propagation of the error, BPE. We expect our hybrid scheme to be faster than using BPE alone.

Từ khóa

#Optimization methods #Transfer functions #Artificial neural networks #Circuits #Neurons #Genetics #Backpropagation #Resource management #Weight measurement #Error correction

Tài liệu tham khảo

fausett, 1994, Fundamentals of Neural Networks Architectures Algorithms and Applications dorji, 2001, Diagnostic Aid for the Vocal Tract, Research Report EE 466 Electrical and Computer Engineering haataja, 1999, Using Genetic Algorithms for Optimization: Technology Transfer in Action, Evolutionary Algorithms in Engineering and Computer Science go1dberg, 1989, Genetic Algorithm in Search Optimization and Machine Learning vose, 1999, The Simple Geenertic AI gorithm, A Bradford Book, 402 5, 68v 10.1007/978-3-662-04199-4 beddoes, 2001, ANNmodels for jaw-research, EMBC Conference 2001 10.1109/NAFIPS.2001.944283