Bimodal stochastic optical learning machine

The authors present the results of numerical and experimental study aimed at optoelectronic realization of networks that can be switched between two distinct operating modes, depending on noise level in the network. The results include fast simulated annealing by noisy thresholding, which demonstrates that the global energy minimum of a small analog test network can be reached in a matter of a few tens of neuron time constants, and stochastic learning with binary weights, which paves the way for the use of fast binary and nonvolatile spatial light modulators to realize synaptic modifications.<>