On research of optimization strategy for dynamic backtracking
Hongbo Li, Zhanshan Li, Yang Ai, Hui-ying Du
Abstract
Hongbo Li, Zhanshan Li, Yang Ai, Hui-ying Du
Abstract
Constraint Satisfaction Problem is an important branch of Artificial Intelligence, one typical algorithm to solve Constraint Satisfaction Problem is the searching algorithm based on backtracking. The Dynamic Backtracking algorithm proposed by Ginsberg in 1993 is an efficient algorithm which uses backtracking integrates with constraint propagation. Now, according to the basic idea of dynamic backtracking, we put forward four implementary strategies and demonstrate that the efficiency and backtracking times of these four strategies are different. The most efficient strategy of these four strategies is the Strategy2.1, it can significantly improve the efficiency and reduce the backtracking times. Anatomizing the results of experiments, we find the differences between these four strategies, then we propose an heuristic rules to improve dynamic backtracking algorithm on selecting a variable that has not been instantiated --- Successful Assignment Principle. According to the Failure First Principle, we propose an optimization strategy that combine the Successful Assignment Principle with the Failure First Principle --- Strategy 2.4.What is more, the final test results show that efficiency of Strategy 2.4 is 1.595 ~ 2.227 times more than the that of Strategy 2.1.
OpenAlex reports 5 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Constraint Satisfaction Problem is an important branch of Artificial Intelligence, one typical algorithm to solve Constraint Satisfaction Problem is the searching algorithm based on backtracking. The Dynamic Backtracking algorithm proposed by Ginsberg in 1993 is an efficient algorithm which uses backtracking integrates with constraint propagation. Now, according to the basic idea of dynamic backtracking, we put forward four implementary strategies and demonstrate that the efficiency and backtracking times of these four strategies are different. The most efficient strategy of these four strategies is the Strategy2.1, it can significantly improve the efficiency and reduce the backtracking times. Anatomizing the results of experiments, we find the differences between these four strategies, then we propose an heuristic rules to improve dynamic backtracking algorithm on selecting a variable that has not been instantiated --- Successful Assignment Principle. According to the Failure First Principle, we propose an optimization strategy that combine the Successful Assignment Principle with the Failure First Principle --- Strategy 2.4.What is more, the final test results show that efficiency of Strategy 2.4 is 1.595 ~ 2.227 times more than the that of Strategy 2.1.
Key concepts: Backtracking, Constraint learning, Beam stack search, Mathematical optimization, Computer science, Constraint (computer-aided design), Heuristic, Constraint satisfaction problem