A formulation for 2-D array neural nets for electro-optical implementation

A description is given of a formulation for 2-D neural nets which is suitable for the processing of 2-D array images. The formulation essentially processes a 2-D input array to be premultiplied and postmultiplied by a matrix of weights. Then the resulting 2-D array of data is processed via the usual sigmoidal thresholding and feedback, which are characteristic of neural nets. The authors describe two electrooptical implementations of their formulation: the first includes adaptations of the available incoherent optical techniques for the real-time multiplication of three matrices, while the second includes adaptations of vector-matrix multiplication and the sum of the outer product of vectors. The second requires less optical hardware for its implementation