Abstract | ||
---|---|---|
It is not uncommon to observe that the performance of constraint solving on a particular problem can be easily influenced by altering the search strategy, restart policy and their parameter settings, etc. In the multicore era, this lack of robustness can be exploited to speed up the constraint solving by designing a parallel portfolio search which simultaneously executes different incarnations of a sequential solver on the same problem. In this paper, we first investigate the techniques of existing single-solver-based portfolio approach in detail. On this basis, we gained insight into how to improve the portfolio approach. We then present the parallel stochastic portfolio search that benefits from the explicit early diversity resulted from randomization and parallelism. Performance evaluation on some classical constraint satisfaction problems benchmarks shows that our technique can solve harder instances compared to the existing portfolio; and not only that, it has advantages of simplicity for implementation and good scalability. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/ISPA-BDCloud-SustainCom-SocialCom48970.2019.00106 | ISPA/BDCloud/SocialCom/SustainCom |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Ke Liu | 1 | 20 | 16.97 |
Sven Löffler | 2 | 0 | 0.34 |
Petra Hofstedt | 3 | 50 | 18.83 |