INSTITUTE OF INFORMATION TECHNOLOGIES - BAS

Cybernetics and Information Technologies
Volume 2, No 2. Sofia, 2002, Bulgarian Academy of Sciences


Entropy Based Delta Rule for Supervised Training of Temporal Sequence Sensitive Neuron

Stefan Sarnev
Institute of Information Technologies, 1113 Sofia


Abstract:
A robust method for recognition of specific temporal correlations in the input pattern is presented. The method is based on a modified model of spiking neuron. A supervised single neuron training algorithm is proposed. The training rule could be used with both types of input patterns - rate-coded spatial pattern and temporal coded pattern. A Cascade-Correlation architecture enables complex temporal sequence recognition.

Keywords: spiking neuron model, entropy based Delta rule, supervised neuron training algorithm.