Experimental evaluation of a hybrid method for configuring ensemble encoding receptors

Abstract

Ensemble encoding is a biologically-motivated, distributed data representation scheme for MLP networks. Multiple overlapping receptive fields are used to enhance the locality of representation. The number, form, and placement of receptive fields has a great impact on performance. In this paper, we explore a technique in which clustering is used to determine receptive field centers, and a variance-based method is used to determine receptive field widths. The relative performance of this hybrid method to other published methods is evaluated experimentally on three benchmark data sets.