Advanced neural-network training algorithm with optimized error based on modified gram schmidt with reorthogonalization

Abstract EN

In this paper we purpose an advanced supervised training method for neural networks.

It is based on Modified Gram-Schmidt with reorthogonalization (MGSR) process.

And it is formulated, in some sense, in the spirit of the Feed-Forward Neural Network (FFNN) algorithm.

It outperforms (in terms of training accuracy, convergence properties, overall training time, etc.) the basic back propagation and its variations with variable learning rate significantly.

The new method developed in this paper is aiming at improving convergence properties, in supervised training of neural networks.

Simulation results are provided to demonstrate the superior performance of the purposed algorithm over the FFNN algorithm.