Rough restricted Boltzmann machine -New architecture for incomplete input data

Abstract

In the paper, a rough restricted Boltzmann machine (RRBM) is proposed. It is a hybrid architecture, which extends the restricted Boltzmann machine (RBM) using some elements of the Pawlak rough set theory. The main goal of such hybridization is to allow processing the imperfect input data and expressing the imperfection in the answer of the system. In the paper, one form of the imperfection is considered - missing values. However, the solutions similar to presented one can be designed also to handle e.g. imprecise data. The formal definition of RRBM is illustrated by experimental results on a handwritten digits reconstruction.