Improved backpropagation learning rates in neural nets

Abstract

The paper introduces a method to improve the convergence rate of the Backpropation (BP) Algorithm. In this method the calculation of the learning rate is based on the Inverse Gradient (IG). This method is compared with the usual BP algorithm and the Gradient Range Based Heuristic (GRBH) method [1][4]. The results show that,the IG method has considerably improved the convergence rates of the BP and GRBH methods for the chosen examples.

Keywords:

• Backpropagation
• neural nets
• gradient range based heuristic (GRBH)
• inverse gradient (IG)

C.R.Categories:

• F 1.1
• I 2.8