Deep 2D-Neural Network and its Fast Learning
Yevgeniy V. Bodyanskiy, Iryna Pliss, Daria S. Kopaliani, Олена Бойко · 2018
A deep 2D-neural network and its learning algorithm are proposed. This system is based on the 2D analogue of the elementary Rosenblatt's perceptron, the error backpropagation procedure, and the matrix analogue of the Kaczmarz-Widrow-Hoff algorithm. The distinguishing feature of the proposed system is that it saves information between the rows of the processed 2D observations.