Report available

[Kedar Babu Madineni]

The below mentioned technical report is now available. If you
would like to have postscript versions of this paper please
send your email request to me.
----------------------------------------------------------------
Technical Report ECE/LSU 94-41, April 25, 1994

TWO CORNER CLASSIFICATION ALGORITHMS FOR TRAINING 
THE KAK FEEDFORWARD NEURAL NETWORK


			Abstract
This paper presents two new algorithms ALG1 and ALG2 for training the Kak
feedforward neural network. The learning power and the
generalization capabilities for this class of algorithms are presented. 
Comparing with the results for backpropagation, we obtain the following
results: ALG2's generalization capability was sometimes better
to or comparable to that of backpropagation. ALG1's generalization
capability was also comparable to that of backpropagation.
 The main advantage of the proposed algorithms is in the 
training time. Both ALG1 and ALG2 outperformed backpropagation 
in learning time, as expected. From these
results, it can be said that ALG2 is superior to backpropagation in 
learning time and at least comparable in generalization.
The comparison experiments were performed on time-series prediction
of interest rates of corporate bonds over a time period of three years,
and creditworthiness of a customer.