YRAN2SAT: A novel flexible random satisfiability logical rule in discrete hopfield neural network. (September 2022)

• Record Type: Journal Article
• Authors: Guo, Yueling; Kasihmuddin, Mohd Shareduwan Mohd; Gao, Yuan; Mansor, Mohd. Asyraf; Wahab, Habibah A.; Zamri, Nur Ezlin; Chen, Ju
• Is Part Of: Advances in engineering software. Volume 171(2022)
• Publication Date: 2022-09
• Subjects: Discrete hopfield neural network -- Non-systematic logic -- Flexible logical rule -- Y-Type random 2 satisfiability -- Random dynamics; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2022.103169
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: A novel Satisfiability namely YRAN2SAT is introduced by generating first and second order logic with randomized clauses combinations. By introducing both features, the proposed Satisfiability is able to explore the search space. YRAN2SAT will be embedded into Discrete Hopfield Neural Network by minimizing the inconsistency of the logical rule that leads to zero cost function. The obtained cost function will be mapped to satisfied assignments which contributes to optimal synaptic weight management for the network. Suitable metrics is applied to demonstrate the performance of both learning and retrieval phase of YRAN2SAT in the network. Several factors such as synaptic weight management, learning and retrieval errors, energy profile, and quality of final neuron states will be analysed. Abstract: The current development of the satisfiability logical representation in Discrete Hopfield Neural Network has two prominent perspectives which are systematic and non-systematic logic. In general, systematic logic was reported to provide us with more consistent and very predictable behaviour whereas non-systematic logic provides broader solution space with a non-predictable pattern. However, the main weaknesses of the existing satisfiability logical rule are the non-flexibility of the logical composition and a low capability to represent the feature's intersection between systematic and non-systematic logic. In this paper, a flexible logical rule called Y-Type Random 2 Satisfiability was proposed by combining the structure of the systematic and non-systematic logic, as a hybrid satisfiability logical rule. The proposed logical rule enumerates randomly based on the structure of either first-order, second-order, or both orders of the clause. The proposed logical rule will be implemented into Discrete Hopfield Neural Network with the aim of minimizing the cost function. The behaviour of the proposed logical rule is compared with the existing satisfiability logic in terms of various performance evaluation metrics. The results show the improvement of the solution capacity for our proposed logic in terms of synaptic weight assessments, various error analyses, energy evaluation, the effects of different states, similarity index, and variability of states. The findings confirm that the proposed logical rule serves as a flexible symbolic instruction for the Discrete Hopfield Neural Network by obtaining the optimal synaptic weights and more variation of neuron states in solution spaces.