Application of Fuzzy Answer set Programming in Hybrid Deep Neural-Symbolic Architecture

Abstract

Hybrid deep neural-symbolic architecture for event-detection employs a deep neural network at the back-end to perform low-level reasoning and a symbolic logical module to perform high-level cognitive reasoning. The currently known hybrid architectures use classical Answer Set Programming(ASP), which is unable to perform fuzzy reasoning with uncertainty. Moreover these systems don’t extract new rules from the available data. On the other hand, there are neuro-fuzzy systems that can extract fuzzy rules from data by means of Gaussian Restricted Boltzman Machines (GRBM). Both the aspects should be merged together to achieve human-like intelligent reasoning and learning from environment. But the success of such an integration depends upon the chosen logical system, that can support fuzzy reasoning with uncertainty, as well as, can support the extracted knowledge from GRBM. This work investigates the feasibility of using interval-valued fuzzy logic programming for this purpose. This work focuses on the theoretical aspects from logic programming perspective.