Optical Neural Networks Using Volume Holograms

Author: Gu, Claire Xiang-Guang

Year: 1990

Degree: Dissertation (Ph.D.)

Advisor: Psaltis, Demetri

Committee Member: Unknown, Unknown

Option: Physics

DOI: 10.7907/HW3F-PK12

Abstract

The optical implementation of neural networks utilizing volume holograms is investigated. The intrinsic degeneracy effect that limits the number of independent interconnections are identified and analyzed by applying the K-space analysis. Basic relationships between the number of neurons, the number of interconnections and the size of the optical system that is used to implement the neural network are derived. Systematic methods for selecting the positions of the neurons to achieve the maximum number of independent interconnections are described. Experiments of global and local connectivities accomplished by using fractal sampling grids for eliminating the degenerate interconnections are presented. The degrees of freedom of the volume hologram and of the planar hologram are compared.

Files