International Journal of applied mathematics and computer science

online read us now

Paper details

Number 3 - September 1994
Volume 4 - 1994

Estimation of the hidden layer size based on analysis of neural networks for handwritten digits recognition

Zbigniew Mikrut

Abstract
The paper contains quantitative information about parameters generated in hidden layers of back-propagation type neural networks for handwritten digits recognition. On the basis of this information, a method of determination of the lower limit for the number of hidden layer elements for such networks is given. The method applies a suitable formula of combinatorics. The above-mentioned information is used to create an algorithm for the training set analysis aimed for more effective neural network initialization.

Keywords
-