基于CUDA的概念生成算法研究-计算机软件与理论专业论文.docxVIP

摘要 摘要 概念格理论，是在形式背景中进行数据分析和规则提取的强有力工具。在应 用概念格理论进行相关领域研究时首先要从所给定领域的形式背景中生成概念并 建立格结构。然而在大数据集中进行这些操作非常复杂，因此生成概念并建立概 念格算法的效率问题就成为影响该理论应用的关键因素。 随着计算机技术的发 展，在高性能计算领 域出现了众多实现技 术，其中 NVIDIA 公司推出的高性能计算技术 CUDA，充分利用了 GPU 的众核、并行特性 来挖掘其潜在的通用计算效能，能够带来较高的性价比，因此得到广泛应用。 本文首先介绍了高性能计算技术 CUDA 的技术原理，概念格的基本理论以及 在各个领域的应用情况以及目前所采用的多种概念生成算法，文中重点阐述了概 念格生成算法 CloseByOne 的数学原理，并结合 CUDA 的技术特点详细分析了采 用 CUDA 对该算法进行并行化的理论可行性。在此基础上，设计并实现了五种基 于 CUDA 的概念生成算法，给出了这五种算法的基本思想和算法描述，最后结合 实验数据详细分析了造成算法实现效能未能得到有效提高的因素。 尽管本文中提出的五种算法都没有能够实现对概念生成算法加速的效果，然 而文中结合 CloseByOne 算法的结构特点重点分析了影响加速性能的技术因素，对 实现基于 CUDA 的概念格生成算法具有指导意义。 关键字：形式概念分析 CloseByOne CUDA 并行计算 基于 基于 CUDA 的概念生成算法研究 Abst Abstract Abstract The theory of concept lattice is becoming a powerful tool for data analysis and rules extraction. When the theory is applied on the relative rearch, concept generation and the establishment of concept lattice must first be done. However, these operations will become much complicated in big data. So the efficiency of algorithms for generating concepts and establishing concept lattice becomes the key factor which affects the application of the theory. Along with the development of computer technology, many technologies appeare in the high performance computing area.CUDA, as one of them, is brought out by NVIDIA and based on graghics processor produced by the company. CUDA fully makes use of the the characteristics of many cores and parallelism in the processor to dig out its potential general calculation efficiency and can bring high performance-to-price ratio, so it can be widely used. In this paper the principle of high performance computing technology CUDA, the basic theory of concept lattice and the application in different fields and a variety of concept generation algorithms are firstly introduced. based on these, the mathematical theory 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 of CUDA is analyzed