Quantum noise in optical neural networks

Summary form only given. Optical implementation of neural networks is accompanied by an intrinsic quantum noise which causes the optical network to deviate from its desired characteristics. This noise is modeled by a shot-noise process and is exhibited in the processes of recording the network weights (recording noise) and detecting the optical beams that are transmitted through the network (computing noise), Shot noise is governed by the impulse response of the detection process (which governs the processing time), the system´s spatial resolution, and the optical power (photon flux) used for recording and computing. We have established theoretical results that characterize the dependence of the average probability of incorrect mapping P, of the network input to the desired output in terms of these noise parameters for binary-input/binary-output networks with hard-limiting nonlinearities