Implementing Contextual Neural Networks in Distributed Machine Learning Framework

Abstract

Contextual neural networks are generalization of multilayer neural networks. They possess interesting property of automatic selection of data attributes needed for correct processing of given input vectors. To achieve that they are using neurons with conditional, multistep aggregation functions and error generalized error backpropagation algorithm based on self-consistency paradigm. According to the literature of the subject, currently there are no implementations of those models in high-performance machine learning platforms like Mahout or MLlib. In this paper we present initial results of implementation of contextual neural networks in distributed machine learning framework called H2O. The motivation behind this work is the need to analyze properties of contextual neural networks and conditional multi-step aggregation functions while solving large classification problems.