NEURAL-NETWORK FOR PATTERN-RECOGNITION, BASED ON THE PRINCIPLES OF ADAPTIVE RESONANCE

This paper proposes a three-layer neural network for pattern recognition, with feedbacks and complex states of the neurons and the interconnections. The network uses adaptive resonance ideology and consists of layers for comparison, recognition, and selective attention. The patterns are compared in spectral space, and the recognition and selective attention is carried out in pattern space. Access to long-term memory is parallel-serial. Adaptation is accomplished by creating new categories and by varying the long-term memory, and the network is implemented with hybrid optoelectronics. The optical part is based on a combined transformation correlator with a dynamic holographic filter.