Nonlinear mapping with minimal supervised learning

Abstract
The problem of interpolating unknown mappings from known mappings is addressed. This problem arises when a large number of mappings must be learned and it is impractical to train the network on all possible mappings. Described is a network model that can learn nonlinear mappings with a minimal amount of supervised training. A combination of supervised and supervised learning is used to train the network. It is shown that the network is able to interpolate mappings on which it has not been previously trained.

This publication has 10 references indexed in Scilit: