Instantaneous Learning in Pattern Recognition

One of the main disadvantages of artificial neural networks is slow learning associated with the need to calculate a large number of coefficients. The article shows that learning can be substantially accelerated. Acceleration is achieved by a sharp reduction in the number of training patterns. In addition, the inverse pattern method was used both for the formation of features and for the subsequent recognition of objects; this has made it possible to dispense with coefficients, which significantly reduces the amount of calculations. In instantaneous learning, as in deep learning, features are generated automatically. Computational experiments have shown the invariance of the method with respect to not only scaling and rotations but also large deformations of objects to be recognized.