Parallel Execution of Neural Networks for Solving Satisfiability Problem
スポンサーリンク
概要
- 論文の詳細を見る
We have proposed a neural network named Langrage programming neural network with polarized high-order connections (LPPH) for solving the SAT, together with parallel execution of LPPHs to increase efficiency. Experimental results demonstrate a high speedup ratio using this parallel execution of LPPHs. Furthermore, it is easy to realize by hardware. LPPH dynamics has an important parameter named attenuation coefficient which strongly affects LPPH execution speed. For parallel execution of LPPHs, it is important to increase diversity of the set of LPPHs. We have proposed a method in which LPPHs have mutually different attenuation coefficients generated by a probabilistic generating function. Experimental results show the efficiency of this method. We also have proposed a LPPH dynamics with a bias. In this paper, to increase the diversity we propose a parallel execution in which LPPHs have mutually different kinds of biases, e.g., a bias toward 1 (positive bias), a bias toward 0 (negative bias), and a bias toward 0.5 (centripetal bias). For some problems, a positive bias has advantage if percentage of 1s is high in a solution, and negative bias if percentage of 0s is high. However the speed of the dynamics of LPPH does not completely depend on the percentage of 1s or 0s. So it is difficult to decide which bias is better before solving a problem. In this paper, we introduce mixed biases to parallel execution of LPPHs. Experimental results show the efficiency of the method.
- 社団法人電子情報通信学会の論文
- 2005-03-22
著者
-
Nagamatu Masahiro
Graduate School Of Life Science And Systems Engineering Kyushu Institute Of Technology
-
Zhang Kairong
Graduate School of Life Science and Systems Engineering, Kyushu Institute of Technology
-
Zhang Kairong
Graduate School Of Life Science And Systems Engineering Kyushu Institute Of Technology
関連論文
- 9P-E-5 Using the Internal Rewards for a Mobile Robot to Survive and Perform a Task Effectively.(Room E International session)
- Controlling the Diversive and Specific Exploration in Solving Disjunctive Linear Constraint Satisfaction Problem(Contribution to 21 Century Intelligent Technologies and Bioinformatics)
- A Continuous Valued Neural Network with a New Evaluation Function of Degree of Unsatisfaction for Solving CSP(Biocybernetics, Neurocomputing)
- Lagrange Neural Network for Solving Constraint Satisfaction Problem
- Parallel Execution of Neural Networks for Solving Satisfiability Problem
- Simple but Efficient Method for Parallel Execution of State Space Search
- Lagrangian Method for Solving Software-MCU Assignment Problem in Car Electronic System Design