2024/01/18 by Fischer, Manfred M., Gopal, Sucharita
paper · doi:10.57938/c97073bd-c561-4504-a325-ca54c043ff9a
Leaming in neural networks has attracted considerable interest in recent years. Our focus is <br/>on learning in single hidden layer feedforward networks which is posed as a search in the <br/>network parameter space for a network that minimizes an additive error function of <br/>statistically independent examples. In this contribution, we review first the class of single <br/>hidden layer feedforward networks and characterize the learning process in such networks <br/>from a statistical point of view. Then we describe the backpropagation procedure, the leading <br/>case of gradient descent learning algorithms for the class of networks considered here, as <br/>well as an efficient heuristic modification. Finally, we analyse the applicability of these <br/>learning methods to the problem of predicting interregional telecommunication flows. <br/>Particular emphasis is laid on the engineering judgment, first, in choosing appropriate <br/>values for the tunable parameters, second, on the decision whether to train the network by <br/>epoch or by pattern (random approximation), and, third, on the overfitting problem. In <br/>addition, the analysis shows that the neural network model whether using either epoch-based <br/>or pattern-based stochastic approximation outperforms the classical regression approach to <br/>modelling telecommunication flows. (authors' abstract)