Abstract
The Boltzmann machine, a hardware implementation of the simulated annealing algorithm, easily and directly takes advantage of highly interconnected systems for parallel processing. Optical systems for matrix operations and random number generation greatly relax the computational burden of the digital electronics and increase the speed by many orders of magnitude over the software implementation. The algorithm is inherently fault-tolerant, but simple use of an analog optical system for matrix operations might critically limit available dynamic range and space-bandwidth product. We therefore present a method for significantly increasing the available dynamic range with little sacrifice in the overall grain-processing speed. We also show how under some circumstances the usable space-bandwidth product of the reconstruction can be increased.
© 1986 Optical Society of America
PDF ArticleMore Like This
Anthony J. Ticknor and Harrison H. Barrett
TUY3 OSA Annual Meeting (FIO) 1985
Anthony J. Ticknor, Harrison H. Barrett, and Roger L. Easton
PD3 Optical Computing (IP) 1985
A. J. Ticknor and Harrison H. Barrett
MW6 OSA Annual Meeting (FIO) 1987