Concept Generation Algorithms Based On CUDA

The theory of concept lattice is becoming a powerful tool for data analysis andrules extraction. When the theory is applied on the relative rearch, concept generationand the establishment of concept lattice must first be done. However, these operationswill become much complicated in big data. So the efficiency of algorithms forgenerating concepts and establishing concept lattice becomes the key factor whichaffects the application of the theory.Along with the development of computer technology, many technologies appearein the high performance computing area.CUDA, as one of them, is brought out byNVIDIA and based on graghics processor produced by the company. CUDA fullymakes use of the the characteristics of many cores and parallelism in the processor todig out its potential general calculation efficiency and can bring highperformance-to-price ratio, so it can be widely used.In this paper the principle of high performance computing technology CUDA, thebasic theory of concept lattice and the application in different fields and a variety ofconcept generation algorithms are firstly introduced. based on these, the mathematicaltheory of concept algorithm CloseByOne is described in detail. According to all these,the theoretical feasibility of speeding up the algorithm in parallel by making use ofCUDA is analyzed detailedly. Next, the basic idea of five improved algorithms and theirvariants for concept generation which are combined with CUDA are described. At last,combined with experimental data, detailed analysis of the factors that causes failure ofeffective improvement on algorithm performance is given.Although the five algorithms put forward in this paper fail to speed up the processof concept generation, the paper combines the structure characteristics of theCloseByOne algorithm to analyse the technologic factors that impact performanceacceleration. It is of guiding significance to the implementation of the concept latticegeneration algorithms based on CUDA.