No abstract

No abstract

No abstract

No abstract

A three-layerd neural network for pattern recognition with feedback and complex states of neurons and interconnections is suggested.It consists of comparisons recognition and selective attention layers. Comparison is realized in spectral space, recognition and selective attention are realized in image space. Recognition layer works as "winner takes all". Parallel-sequential accessing to long term memory is used. Adaptation is realized by creation of a new recognition categories and by long term memory change when input pattern are similar enough. Joint transform correlator with dynamic holographic filter is used in optical real ization of adaptive resonance neural network.