Use of optical matrix inversion algorithm implementation for pattern classification and recognition

Neural networks do not need to be programmed for solving a task but can learn via examples. The training consists in updating weights of interconnections between neurons. During the learning process, the weight matrix is sequentially modified, finally taking a certain form which is generally unknown at the start. It is shown that it is possible to predict in advance what form the trained matrix will take and thereby either speed up the learning process or obtain a faster algorithm. A single-layer optical neural network for solving a pattern classification task is considered