Publications [#280391] of David J. Brady
- David, B, Dynamic holographic interconnections for optical neural computers,
Neural Networks : the Official Journal of the International Neural Network Society, vol. 1 no. 1 SUPPL
pp. 374, Elsevier BV, ISSN 0893-6080 [doi]
(last updated on 2019/11/18)
A neural system comprised of p neurons with c interconnections per neuron requires cp degrees of freedom in its interconnections. If these interconnections are formed by interactions between narrow-band signals emitted by the p active elements, only p degrees of freedom are available at any given instant for modifying the interconnections. Thus a full and independent update of the interconnections requires that they be addressed c times. Constraints of this sort are natural to volume holography. While the number of interconnections which may be stored in a volume hologram scales as the volume of the system, N3, the number of optical beams which may be used to record holograms scales as the area enclosing the system, N2. This means that in holographic systems cp ≤ N3 and p ≤ N2. High density storage of holographic interconnections is therefore by necessity a dynamic process. The authors have developed dynamic recording processes which use photorefractive holography to record cp independent interconnections using p inputs. In implementing an optical neural architecture they use their control over the signals emitted by the p active elements to force the system to an interconnection pattern consistent with the needs of the neural system.