Towards Relevance Dendritic Computing

Dendritic Computing (DC) is lattice computing approach to classifier building that uses only additive and lattice operators. Constructive algorithms that provide perfect fitting for arbitrary training data have been provided, however they do not avoid overfitting. In this paper we propose to embed the DC in the sparse bayesian learning framework in order to improve the generalization of DC classifiers. The proposed Relevance DC searches for relevant dendritic structures in a Bayesian framework. This paper provides results of some computational experiments comparing Relevance DC with Relevance Vector Machines (RVM) where RDC provides comparable results with much more parsimonious models.