Satisfiability Logic Programming Incorporating Metaheuristics In Hopfield Neural Networks
Loading...
Date
2017-08
Authors
Mohd Kasihmuddin, Mohd Shareduwan
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
The representation of 2 Satisfiability problem or 2SAT is increasingly viewed as
a significant logical rule in order to synthesize many real life applications. Although
there were many researchers proposed the solution of 2SAT, little attention has been
paid to the significance of the 2SAT logical rule itself. It can be hypothesized that,
2SAT property can be used as a logical rule in the intelligent system. To verify this
claim, different 2 Satisfiability logic programming namely 2SAT, maximum 2 Satisfiability
(MAX2SAT) and Pattern Satisfiability (Pattern-SAT) were embedded to Hopfield
neural network (HNN) as a single unit. Maximum 2 Satisfiability is a type of 2SAT
logical rule that is unsatisfiable in nature. On the other hand, Pattern-SAT is a new
perspective of doing logic programming in HNN-2SAT by helping the user to represent
the pattern in terms of 2SAT programming. Learning in HNN will be inspired
by Wan Abdullah method since the conventional Hebbian learning is inefficient when
dealing with large number of constraints. As the number of 2SAT clauses increased,
the efficiency and effectiveness of the learning phase in HNN deteriorates. Inspired
by evolutionary algorithm and swarm intelligence metaheuristic algorithm has been
introduced to reduce the learning complexity of the network. There are two newly improved
metaheuristic algorithms were proposed namely, enhanced genetic algorithm
(GA) and enhanced artificial bee colony (ABC) algorithm. These two algorithm will
combine with HNN-2SAT model as single unit namely HNN-2SATGA and HNN-
2SATABC. The performance of the all models will be tested by using simulated data
set. In terms of real life application, the newly improved 2 satisfiability based Reverse
Analysis Method (2SATRA) will be implemented to do logic mining in 16 different
real life data sets ranging from finance to medical field. 2SATRA will implement all
the proposed hybrid HNN-2SAT models during learning phase. The performance comparison
indicates that most of the time, HNN-2SATABC with its variant outperformed
other hybrid HNN-2SAT models in doing simulated and real data set.
Description
Keywords
Satisfiability logic programming Incorporating metaheuristics , In Hopfield Neural Networks.