Title :
Technique of learning rate initialization for efficient training of MLP: Using for computing of Lyapunov exponents
Author :
Savitsky, Yury ; Savitsky, Anton
Author_Institution :
Dept. of Electron. Inf. Syst., Brest State Tech. Univ., Brest, Belarus
Abstract :
A new approach for the training of multilayer feedforward neural networks is proposed. Simulation results are presented in the problem of the calculation of the largest Lyapunov exponent to demonstrate an efficiency of the method, which solves the problem of the training step choice in multilayer perceptrons.
Keywords :
Lyapunov methods; learning (artificial intelligence); multilayer perceptrons; Lyapunov exponent; MLP; learning rate initialization; multilayer feedforward neural network; multilayer perceptron; training step choice; Conferences; Decision support systems; Tin; Training; Lyapunov exponent; backward propagation of errors; chaotic processes; learning rate; multilayer neural networks;
Conference_Titel :
Information and Digital Technologies (IDT), 2015 International Conference on
Conference_Location :
Zilina
DOI :
10.1109/DT.2015.7222987
