Development and Analysis of the Technique for the Building of Artificial Neural Networks Based on Adaptive Elements
Technical cybernetics. Information technology. Computer facilities
Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia
AbstractThis paper relates to the feedforward artificial neural networks learned with error backpropagation supervised methods. Authors consider the generalized system approach to describe the main concept of structure diagram of interactive adaptive elements which are implementing the transfer functions of bidirectional – forward and backward – signal propagation. The prototype of the software, including the class library of the basic adaptive elements and their interconnections, was developed using Python (object oriented programming language) for SageMath (mathematical package with GPL-compatible license). As the demonstrative examples, two following problems were solved: the approximation of radar response and the classification of two random processes.
Keywords:artificial neural network; backpropogation; adaptive element; signals and systems; model simulation; gradient descent; SageMath; Python.