Generalization Comparison

[Praveen Raina]

The following report is available that examines the generalization
performance of two feedforward learning techniques.

----
Technical Report ECE 93-03, LSU, March 31, 1993
-----

COMPARISON OF LEARNING AND GENERALIZATION CAPABILITIES OF
THE KAK AND THE BACKPROPAGATION ALGORITHMS

	Praveen Raina
	Dept. of Electrical & Computer Engineering
	LSU, Baton Rouge, LA 70803-5901
	E-mail: raina at max.ee.lsu.edu

Abstract: This report compares the learning and generalization 
capabilities of the Backpropagation and the Kak algorithm. It 
is observed that the Backpropagation algorithm is much more 
computation intensive than the Kak algorithm. The generalization 
performance with respect to an error criterion is better for the 
Backpropagation algorithm for intermediate values of error. But 
when considered together with the number of extra iterations that
the Backpropagation algorithm entails one concludes that the Kak 
algorithm has overall superior performance.