Holographic implementation of a fully connected neural network
Article Abstract:
A fully connected neural network can be optically implemented as a hologram. The example network is single-layered with feedback and resembles a Hopfield network. Two-dimensional image interconnection is described by a four-dimensional kernel that uses a 2-D array of spatial frequency multiplexed holograms. Images in the optical system are separated and stored in correlators as reference images. When an input pattern is presented to the system, a correlator produces the auto-correlation and the cross-correlations, which are sampled by a pinhole array at the pattern center. The signals passing through the pinholes reconstruct the images stored in another correlator. The reconstructed images are superimposed on the input plane, where the strongest correlation will have the brightest image.
Publication Name: Proceedings of the IEEE
Subject: Electronics
ISSN: 0018-9219
Year: 1990
User Contributions:
Comment about this article or add new information about this topic:
Ground states of partially connected binary neural networks
Article Abstract:
The outer products of vectors over the set (-1, 0, 1) define partially-connected neural networks that are internally strongly connected and externally weakly connected. The subnetworks are defined by patterns over the set (-1, 1), and the vectors composed of nonzero bits from the patterns are subwords. The subwords associated with the same bits make up each subnetwork. When one subword from each subnetwork is combined, the combinations that agree in common bits are called permissible words. The permissible words are locally-stable network states when there are two or fewer subwords or each subnetwork's subwords are mutually orthogonal. When each subnetwork holds two orthogonal subwords, the permissible words are the network energy function's ground states.
Publication Name: Proceedings of the IEEE
Subject: Electronics
ISSN: 0018-9219
Year: 1990
User Contributions:
Comment about this article or add new information about this topic:
Imaging with reduced holographic data - an application of the AR model to a multifrequency hologram
Article Abstract:
The AR model, a method of extrapolation of a complete spectrum of data from a band-limited spectrum of data, is successfully applied to the generation of quality, high-resolution multifrequency holographic images from a reduced holographic data set. The coefficients of the model are determined by the maximum entropy method. The extrapolated hologram is shown to be stable and effective. The methodology was validated by experimental examples.
Publication Name: Proceedings of the IEEE
Subject: Electronics
ISSN: 0018-9219
Year: 1988
User Contributions:
Comment about this article or add new information about this topic:
- Abstracts: Localized wave representations of acoustic and electromagnetic radiation
- Abstracts: CMAC: an associative neural network alternative to backpropagation. Backpropagation through time: what it does and how to do it
- Abstracts: A performance comparison of trained multilayer perceptrons and trained classification trees. On the decision regions of multilayer Perceptrons