Parallelizing pincer search using CUDA-A conceptual idea

Recent times have seen meteoric increase in the data that are available using which we can develop automated data-driven techniques of extracting useful knowledge. Data mining is the important step in this process of knowledge discovery. One of the key problem in most of the data mining applications is discovering the frequent item sets. Scanning of the huge data available to discover frequent item sets are computationally expensive. A conventional multi-core processor might not very effective multi-threading capabilities to be able to process large amounts of data leading to sequential implementation of a considerably large number of processes. Such sequential implementation leads to high computation times due to pipeline latency and other issues. Due to this limitation there is an increasing interest in the researchers to develop parallel data mining algorithms for faster implementation and efficient use of available GPU architectures. Pincer search is one the data mining algorithms which is used to discover the maximum frequent sets. Pincer search algorithm reduces both the number of times the database is scanned and also the number of candidate considered. In this study, we discuss a way to parallelize the pincer search algorithm to further speed up the process of discovering maximum frequent sets.