Optical Conceptual Computing and Associative Memory (OCCAM)

摘要

The Optical Conceptual Computing and Associative Memory (OCCAM) Program applied the techniques of neural-network dynamical systems analysis and fuzzy theory to problems in conceptual computing. Key contributions are detailed in the ten technical papers included in the Appendix. Three fundamental results were: the development of bidirectional associative memories (BAMs), a fuzzy knowledge combination scheme that allows arbitrarily many neural (casual) network 'expert systems' from experts with arbitrary credibility to be naturally synthesized into a single, representative associative knowledge network, and the development of pure fuzzy associative memories (FAMs). UCSD focused on the theory and application of optical neurocomputing. Four fundamental research thrusts were: the design and implementation of optical neural networks, especially optical BAMs, the investigation of physical properties important to neural network implementations, the application of optics to fuzzy knowledge processing and to fuzzy computing (approximate reasoning) in general, and the application of neural network principles to optical pattern recognition. Several holographic and nonholographic BAM systems were devised, with emphasis on volume holography. A backpropagation (hierarchial supervised learning) network was taught to perform rotation-invariant pattern recognition.