A study on hill climbing algorithms for neural network training

Author

Chalup, Stephan ; Maire, Frederic

Author_Institution

Machine Learning Res. Centre, Queensland Univ. of Technol., Brisbane, Qld., Australia

Volume

3

fYear

1999

fDate

1999

Abstract

This study empirically investigates variations of hill climbing algorithms for training artificial neural networks on the 5-bit parity classification task. The experiments compare the algorithms when they use different combinations of random number distributions, variations in the step size and changes of the neural networks´ initial weight distribution. A hill climbing algorithm which uses inline search is proposed. In most experiments on the 5-bit parity task it performed better than simulated annealing and standard hill climbing

Keywords

evolutionary computation; learning (artificial intelligence); neural nets; pattern classification; random number generation; 5-bit parity classification task; 5-bit parity task; artificial neural networks; hill climbing algorithms; initial weight distribution; inline search; neural network training; random number distributions; simulated annealing; step size; Artificial neural networks; Backpropagation algorithms; Computer architecture; Computer networks; Evolutionary computation; Feedforward neural networks; Feedforward systems; Machine learning algorithms; Neural networks; Random variables;

fLanguage

English

Publisher

ieee

Conference_Titel

Evolutionary Computation, 1999. CEC 99. Proceedings of the 1999 Congress on

Conference_Location

Washington, DC

Print_ISBN

0-7803-5536-9

Type

conf

DOI

10.1109/CEC.1999.785522

Filename

785522