Adaptive integration of multiple experts
Abstract
A novel method of integrating multiple experts in an adaptive manner is proposed. Each expert specializes in a particular sub-domain but performs poorly on the entire domain. By combining several such experts, the overall performance can be boosted significantly. To that effect, a supervised learning method, known as the supervised clustering and matching (SCM) algorithm, is used to combine the decisions of these experts based on their performance profile. By the fast and incremental learning capability of SCM, expert integration can be performed both on-line and off-line. Experiments on a sample benchmark problem illustrate that expert integration improves significantly upon the performance of each expert.
This paper has been withdrawn.