Initializing RBF-Networks with Small Subsets of Training Examples

An important research issue in RBFnetworks is howto determine the ganssian centers of the radial-basis functions. Weinvestigate a technique that identifies these centers withcarefully selected training examples, with the objective to minimize the network’ssize. Theessence is to select three very small subsets rather than one larger subset whose size wouldexceedthe size of the three small subsets unified. The subsets complementeach other in the sense that whenused by a nearestneighborclassifier, each of themincurs errors in a different part of the instance space. Thepaper describes the example-selectionalgorithm and shows, experimentally, its merits in the design of RBFnetworks.

Authors

• Miroslav Kubat
• Martin Cooperson
• Jr. Battle Command Battle Lab (BCBL); William Turmel and Kenneth D. Forbus
• University of Southwestern Louisiana

Keywords

No keywords are indexed for this paper.