Search
A neural net for prediction prodlems
![Thumbnail](/bitstream/handle/data/37083/49-52.pdf.jpg?sequence=2&isAllowed=y)
View/ Open document files
Date
1996Publisher
Politehnica university of TimişoaraCitation
Golovko, V. A. A neural net for prediction prodlems / V. A. Golovko, J. V. Savitsky, V. B. Gladishchuk // Computer science and engineering : proceeding the international conference of technical informatics, Timişoara, November 14–16, 1996 / Politehnica university of Timişoara, Faculty of automation and computer science and engineering. – Timişoara, 1996. – P. 49–52. – Vol. 2. – Bibliogr.: p. 52 ( 3 titles).Abstract
In this paper are presents and investigates the various modifications of gradient descent method for Back-Propagation algorithm. Opportunity on reduction of neural network training time and increasing of training algorithm stability are considered The particular expressions for choosing of optimum steps of training neurons with sigmoid and linear functions of output activity are received. Results of experiments on predicting of various mathematical series and functions by using neural networks are presented The analysis of speed training neural network for offered training algorithm demonstrates its advantage before classical Back-Propagation algorithm.
Collection
![Лицензия Creative Commons](https://i.creativecommons.org/l/by-nc/4.0/88x31.png)
Это произведение доступно по лицензии Creative Commons «Attribution-NonCommercial» («Атрибуция-Некоммерчески») 4.0 Всемирная.