Estimation of correctness region using clustering in mixture of experts

This paper investigates a mixture of experts system with a gating network that estimates the knowledge space of each classifier in the feature space. A clustering algorithm is used to estimate the correctness region of each classifier in a feature space. Unlike other mixture of experts systems that use a priori information on the classification of each expert classifier, this system uses the information about the feature space region where the expert is estimated to classify correctly. The clustering nature of the gating function of this system allows fast partitioning and soft splitting of the knowledge space that cannot be linearly partitioned.