Classification systems, which have long been realized through rule-based systems, are now being implemented with neural-expert system hybrids because of the well known difficulties associated with the formation of rules and the ease with which neural networks can capture the relationships between attributes and classifications. Backpropagation, which is a commonly used neural network training algorithm, suffers from slow training and the possibility of local minima trapping. In this paper we propose a new neural network training algorithm, ternary synaptic weights (TSW) algorithm, which offers fast, guaranteed learning and automatic topology determination. Also, it provides an insight into the interrelationships between the attributes and the classifications, which is very suitable for hybridization of neural networks and expert systems. Furthermore, it takes don't care attributes, which are an essential part of rule-based systems, into consideration during training.<>
Ternary synaptic weights algorithm: neural network training with don't care attributes
01.01.1994
256400 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Ternary Synaptic Weights Algorithm: Neural Network Training With Don't Care Attributes
British Library Conference Proceedings | 1994
|A Learning Algorithm of Fuzzy Neural Networks with Fuzzy Weights
British Library Online Contents | 1993
|Memristive synaptic crosstalk effects on Hopfield neural network
British Library Conference Proceedings | 2022
|Fuzzy Neural Networks with Fuzzy Weights
British Library Online Contents | 1993
|