A linearization technique for linearly inseparable patterns
Sungkwon Park, J.H. Kim · 2002
This paper concerns a technique which transforms a set of linearly inseparable binary patterns to a set of linearly separable one. Using the technique, a framework to train multilayer perceptrons without iteration is introduced. The trained multilayer perceptrons using these ideas use only hard limiters as neutrons and integer weights and thresholds. Hence accurate hardware implementation of the networks can be realized using the readily available VLSI technology.>