Analog Optical Neural Nets: A Noise Sensitivity Analysis
Abstract
The development of analog optical implementations of neural networks such as the multilayer perceptron with learning by backward error propagation (BEP) requires an understanding of the noise sensitivity of such architectures. In this program computer simulations were used to study the effects of component and system noise on the performance of such optical implementations. A hybrid optical/electronic parallel architecture capable of both the forward pass and backward error propagation steps of training data presentation was conceived and modeled. The simulations showed that the most significant effects were due to the nonlinear response of the spatial light modulators used to store and update the neural weights. Another conclusion of the simulation results was that increasing the hidden layer size increases noise immunity significantly.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 11, 1991
- Accession Number
- ADA242920
Entities
People
- James J. Levy
- Michael W. Haney
- Ravindra A. Athale
Organizations
- Braddock Dunn & McDonald