It is common to say that optics offer advantages in realizing the parallelism, massive interconnectivity, and plasticity required in the design and construction of large scale neurocomputers. This suitability is discussed in the case of stochastic neural nets such as those implementing the simulated annealing algorithm and the Boltzmann Machine. An optoelectronic random number generator is proposed. It is based on speckle properties and allows the design of a small stochastic unit on silicon. Experimental results of probabilistic updating are given.
Philippe Lalanne
and Pierre H. Chavel
"Optoelectronic hardware issues for implementation of simulated annealing or Boltzmann machines", Proc. SPIE 1621, Optical Memory and Neural Networks, (1 November 1991); doi: 10.1117/12.50446; https://doi.org/10.1117/12.50446
The alert did not successfully save. Please try again later.
Philippe Lalanne, Pierre H. Chavel, "Optoelectronic hardware issues for implementation of simulated annealing or Boltzmann machines," Proc. SPIE 1621, Optical Memory and Neural Networks, (1 November 1991);