In recent decades, research on the structure and function of brain has led to the development of a powerful new paradigm for solving problems of pattern and speech recognition, classification, optimization, and so on. Neural network in brain is known to store informations in the manner of associative memory and to be massively interconnected and parallel. Optical holography has inherently those properties.
In this thesis, linear and quadratic associative memories are holographically implemented. The holographic associative memories are prepared by using coherent scattered light and analyzed quantitatively in optical experiment. In the linear associative memory, picture elements(pixels) of stored patterns are encoded by two processes: transparent/opaque coding process and polarization coding process. The experimental and computer simulation results of both coding processes, are presented and analytically compared with Hopfield model.
As a higher order correlation model for associative memory, the experimental setup of quadratic associative memory is proposed and implemented holographically by using the scattered light and the polarization coding process. The polarization coding process is made by liquid crystal light modulator. Experimental results of retrieving the stored patterns from an incomplete input are also presented. The holographic implementation of quadratic associative memory makes the capacities of storage and error correction greater than those of the linear associative memory.
