Practical optoelectronic neural network realizations based on the fault tolerance of the backpropagation algorithm

Abstract

This paper describes how the fault tolerance of the backpropagation algorithm can be used to accommodate the realistic (nonideal) transfer characteristics of the optical communication links used, between neural layers, in optoelectronic neural networks. In particular the authors demonstrate that networks, utilizing MSM (metal-semiconductor-metal) photodiodes (PDs) and either LED (light emitting diode) or MQW (multiple quantum well) laser transmitters within these intraneural links, are capable of performing satisfactorily even in the presence of such nonideal device phenomena as: 60% optical crosstalk, 50% optoelectronic device variation, or a thresholded (I(th)=0.5*I(max)) laser output characteristic. Subsequent to this, the authors then show how it is possible to use this fault tolerance to simplify the neuron architecture, to the extent that it consists only of MSM PDs a current amplifier, and an MQW laser. The overall neuron transfer function is then a first-order approximation to the original sigmoidal function.

Stabilizing and robustifying the learning mechanisms of artificial neural networks in control engineering applications

“…and a scaling factor denoted by η φ , for the selection of which, a detailed analysis is presented in the subsequent discussion. Using Euler's first order approximation for the derivative term, one obtains the following relation, which obviously validates the constructed model in (5) and which leads to the representation in (7).…”

“…In most applications of artificial neural networks, EBP method constitutes the central part of the learning [3][4][5][6][7][8][9][10][11][12][13][14][15][16][18][19][20]. In this section, the technique is briefly reviewed for systems in which the outputs are differentiable with respect to the parameter of interest.…”

“…Feedforward and recurrent neural networks, Gaussian radial basis function neural networks, dynamical neural networks [3], and Runge-Kutta neural networks [4][5] constitute typical structurally different models. The approaches mentioned have widely been used in applications extending from speech recognition and optoelectronics to identification and control of nonlinear systems [3][4][5][6][7][8][9][10][11][12]. Most of the studies reported in the literature adopt the Error Backpropagation (EBP) method for tuning the parameters of the network structure [3][4][5][6][7][8][9][10][11][12][13][14][15][16].…”

“…The approaches mentioned have widely been used in applications extending from speech recognition and optoelectronics to identification and control of nonlinear systems [3][4][5][6][7][8][9][10][11][12]. Most of the studies reported in the literature adopt the Error Backpropagation (EBP) method for tuning the parameters of the network structure [3][4][5][6][7][8][9][10][11][12][13][14][15][16]. After the discovery of the algorithm in 1986 by Rumelhart et al [17], EBP method has become the standard method for training of artificial neural networks.…”

Computational intelligence in photonics technology and optical networks: A survey and future perspectives

“…During the same period Giles [44] introduced the concept of routing WDM optical networks by the employment of optical neural networks. A variety of implemented optical neural networks [10,43] formed the basis for the evolution of this work by other researchers. So, Kurokawa in [67] introduced the concept of network routing based on a neural network implementation of a parallel decentralized network.…”