| Summary: | 碩士 === 國立交通大學 === 多媒體工程研究所 === 102 === NCTU6 is a Connect6 AI program developed in our lab. NCTU6 and its variations won gold medals in ICGA tournaments three times and also defeated many expert players in Man-Machine Connect6 championships. The objective of this thesis is to improve the program’s strength by parallelization. First, this thesis designs the architecture of parallel program for NCTU6. Based on this architecture, we modified NCTU6 into a baseline parallel program by parallelizing some threat-space searches (TSSs). Second, this thesis applies the technique of dovetailing to parallelize NCTU6 by dovetailing two search algorithms, TSSs and alpha-beta search. The technique of dovetailing is useful and relatively easy especially for parallelizing programs which are already very complicated.
The result of the experiment shows that the baseline parallel program with 12 cores reaches about 79.6% win rate against NCTU6, and, the time taken by the baseline parallel program is about 110% of the time taken by the original program. Appling the technique of dovetailing, the dovetailing parallel program with 12 cores reaches 77.5% win rate against NCTU6, and 61.9% win rate against the baseline parallel program, and, the time taken by the dovetailing parallel program is about 80% of the time taken by the original program. Besides, the version with dovetailing finds more winning solutions. This thesis is also the first one that applies the technique of dovetailing to the non-proof search algorithms such as alpha-beta search.
|
|---|
