3SAT and Fuzzy-HornSAT in Hopfield Neural Network

Abstract

In artificial intelligence, logic programming and neural networks are two significant theories. Logic programming has appeared as a platform for learning abstraction and data mining. The initial propositional logic programming studied uses the HornSAT problem, which is the most direct case of logic programming. The advent of the 3SAT problem has resulted in more comprehensive works in logic and data mining. The 3SAT problem is considered a non-horn clause. Despite their utility, both networks have downsides; some of the greatest troublings is that the results are occasionally local minimum solutions rather than global minimum solutions. This paper describes an improved method for HornSAT to achieve greater energy relaxation and avoid locally minimum solutions by combining Hopfield networks and fuzzy logic techniques, namely fuzzy-HornSAT. The hybrid approaches’ performance was evaluated using a simulated data set. Matlab 2020b was used to train, simulate, and validate the suggested network’s performance. The network for fuzzy-HornSAT and 3SAT was measured using indicators of RMSE, SSE and MAE in the training phase. The energy analysis is also used to compare their robustness by using the ratio of global minima, hamming distance, and processing period.