A training approach based on linear separability analysis for layered perceptrons
David D. Zhang, Mohamed S. Kamel, M.I. Elmasry · 1994
In this paper, we explore linear separability as a training approach for layered perceptrons. A training approach, called layer adaptation (LA), is presented. Its learning mechanism and implementation are described and examples are given to illustrate its effectiveness. Compared with the BP and the MRII algorithms, preliminary analysis shows that the LA is easily implemented using digital VLSI technology while the stability, the training time and the complexity in silicon are acceptable.>