Frequent itemset mining on graphics processors

摘要: We present two efficient Apriori implementations of Frequent Itemset Mining (FIM) that utilize new-generation graphics processing units (GPUs). Our take advantage the GPU's massively multi-threaded SIMD (Single Instruction, Multiple Data) architecture. Both employ a bitmap data structure to exploit parallelism and accelerate frequency counting operation. One implementation runs entirely on GPU eliminates intermediate transfer between memory CPU memory. The other employs both for processing. It represents itemsets in trie, uses trie traversing incremental maintenance. preliminary results show achieve speedup up orders magnitude over optimized PC with an NVIDIA GTX 280 quad-core CPU.