Using a neural network apparatus for study an adaptive learning trajectory

Abstract

A theoretical analysis of the current state of neural network technology was performed, which allowed us to identify the known algorithms for working with neural networks and figure out the advantages and disadvantages of their application to solve various scientific and technological problems. The relevance of the use of neural networks for the study of adaptive learning trajectories based on the analysis of current characteristics of the assimilation of information quanta by students, which ensures the successful conduct of distance learning using open communication platforms. The adaptive process of mastering new knowledge by the student of multilayer perceptron architecture is presented, in which each neuron is considered as a certain lesson, and the educational process is presented as the movement of the student in separate lessons. The application of the inverse error propagation algorithm and the least squares method has been implemented for network training. This made it possible to iteratively adjust the synaptic weights of each neuron by minimizing the network parameters based on minimizing the standard deviation between the correct and actual network responses. The general algorithm of adaptive training with use of a neural network is constructed and stages of its functioning are described. Prospects for further research aimed at applying modifications of the backpropagation algorithm and improving the selection of rules for correcting weights are identified.