Fast 1-itemset frequency count using CUDA

• Main Authors: Uy, Roger Luis, Marcos, Nelson
• Format: text
• Published: Animo Repository 2017
• Subjects: Data mining; Big data; CUDA (Computer architecture); Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/2187; https://animorepository.dlsu.edu.ph/context/faculty_research/article/3186/type/native/viewcontent/TENCON.2016.7847991
• Institution: De La Salle University

Frequent itemset mining is one of the main and compute-intensive operations in the field of data mining. The said algorithm is use in finding frequent patterns in transactional databases. The 1-itemset frequent count is used as basis for finding succeeding k-itemset mining. Thus there is a need to speed-up this process. One of the techniques to speed-up the process is using the Single Instruction Multiple Thread (SIMT) architecture. This architecture allows a single instruction to be applied to multiple threads at the same time. Current graphics processing unit (GPU), which contains multiple streaming processing units, uses SIMT architecture. In order to abstract the GPU hardware from the programming model, NVIDIA introduces the compute unified device architecture (CUDA) as an extension to existing programming languages in order to support SIMT. This paper discusses how 1-itemset frequent count is implemented in SIMT using CUDA.