Fast Deep Learning Neural Networks
Alexander Yu. Dorogov · 2019
Algorithm of network topology construction and training of two-dimensional fast neural networks with additional switched planes is considered. The constructed topology is ideologically close to the topology of convolutional neural networks of deep learning, but it has regular topology with the number of layers established by the factor representation of the dimensions of the image and the output plane of the classes. The learning algorithm has an analytical representation; it is absolutely stable and converges in a finite number of steps. Additional planes extend the information capacity of the tunable transformation to the maximum possible. Control of the planes in the training and processing mode is realized by numerical coordinate codes of the output plane.