Title :
The influence of ART1 parameters setting on the behavior of layer 2
Author :
Chen, Zhong ; Cai, Zi-Xing ; Li, Ze-Wen ; Wang, Xiao-Hua
Author_Institution :
Sch. of Inf. Sci. & Eng., Central South Univ. of Technol., Changsha, China
Abstract :
Due to the complexity of ART network, the setting of parameters is rather difficult. Based on analyzing the architecture and the membrane equation of layer 2 (L2) in ART1 network, this paper describes the oscillation possibility of the activities of neurons in L2 layer and studies the influence of parameters setting on the behavior of L2 layer, such as the number of neurons (S 2), the biases +b and -b, the initial value of L1-L2 connection matrix W 1:2, through a simulation case whose transfer function is faster-than-linear. It also suggests a few basic principles of the parameters setting.
Keywords :
ART neural nets; matrix algebra; transfer functions; ART network; L1-L2 connection matrix; membrane layer 2 equation; oscillation possibility; parameter setting; transfer function; Artificial neural networks; Biomembranes; Computer aided instruction; Computer architecture; Equations; Information science; Neural network hardware; Neurons; Resonance; Subspace constraints; ART Network; Adaptive Resonance Theory; Neural Network; Response;
Conference_Titel :
Machine Learning and Cybernetics, 2005. Proceedings of 2005 International Conference on
Conference_Location :
Guangzhou, China
Print_ISBN :
0-7803-9091-1
DOI :
10.1109/ICMLC.2005.1527788