Optical Computing Based on the Hopfield Model for Neural Nets.
Abstract
Neural net models and their analogs present a new approach to signal processing that is collective, robust, and fault tolerant. The collective nature of processing means also high throughput where bandwidth or speed of the processing elements i.e., temporal degrees of freedom are traded by spatial degrees of freedom via massive interconnectivity of the elements (neurons) in the network. As a result neural nets can solve computationally extensive problems like those encountered for example in optimization, nearest neighbor search, inverse scattering in a matter of a few time-constants of the processing elements. In biological systems (the brain in particular) this is of the order of a tens of milliseconds since neurons operate with electronic conduction and hence can be made considerably faster with time constants approaching nano-seconds. Collective processing such artificial nets can therefore be extremely fast. The remarkable ability of neural nets in handling sketchy (erroneous or incomplete) information and their fault tolerance (graceful degradation in performance with element failure) make them particularly attractive in pattern recognition, robotics, and autonomous system intended to operate virtually unattended for long time periods.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1986
- Accession Number
- ADA168767
Entities
People
- N. H. Hart
Organizations
- Moore School of Electrical Engineering