Exploring Neural Turing Machines Applicability in Neural-Symbolic Decision Support Systems

Abstract

The task of building hybrid decision support systems that combine symbolic and connectionist approaches is actual and challenging. In particular, decision support systems operate with symbolic structures that describe the problem situation, stakeholders, assessment criteria, etc. Integrating connectionist approaches into certain parts of the decision-making process bring robustness, fixed response speed and ability to generalize. This paper examines Neural Turing Machines - a special case of Memory-Augmented Neural Networks - and demonstrates that such an architecture can be integrated into the Decision Support Systems. It was also shown that Neural Turing Machine can solve arithmetic sum task for numbers represented as binary vectors of length 10.