gpgpu下机器学习相关算法的高性能实现与分析-high performance implementation and analysis of machine learning related algorithms based on gp gpu.docx

目录第一章引言·······························11.1研究原因·····························21.1.1机器学习算法的实现与研究·················21.1.2大量训练数据的挑战与机遇·················31.2研究现状·····························41.2.1机器学习算法的应用场景··················41.2.2目前GPGPU机器学习算法实现5第二章CUDA与GPGPU运算62.1并行计算模型···························62.1.1并行硬件体系结构······················62.1.2并行模型和语言·······················8第三章哼唱搜索与动态时间规整算法实现与效果··········133.1哼唱搜索系统的架构·······················133.2哼唱搜索的音高提取算法····················143.3哼唱搜索的匹配算法·······················153.3.1动态时间规整·························153.3.2线性缩放···························173.3.3动态时间规整与线性缩放算法的效果对比·········183.4多显卡实现架构·························183.5算法实现细节与方法·······················193.5.1哼唱数据库的多显卡分发··················193.5.2哼唱任务的多显卡分发···················203.5.3如何做哼唱音高偏移·····················203.5.4通用任务求解流程······················213.5.5显卡选择方法·························223.5.6简单的多显卡动态时间规整算法实现············223.5.7多显卡动态时间规整的优化实现··············233.6实现效果对比···························253.6.1实验设计···························253.6.2动态时间规整计算时间加速·················253.6.3动态时间规整计算显存占用对比··············263.6.4实现结论···························27第四章单机多显卡神经网络GPGPU实现与效果············324.1感知器·······························324.1.1什么是感知器·························324.1.2感知器如何学习·······················344.1.3感知器的局限性·······················354.2多层神经网络···························354.2.1Logistics回归364.2.2多层神经网络的模型·····················384.2.3多层神经网络学习······················404.3深度学习·····························414.3.1为什么要深度？·······················414.3.2深度学习的学习算法·····················424.4使用Theano实现神经网络444.5多显卡并行····························454.6多显卡并行实现流程·······················454.7实现效果·····························474.8实验结论·····························48第五章结论与展望···························495.1结论································495.2展望································50参考文献·································51发表论文和参加科研情况说明·····················56致谢···································57第一章引言机器学习当今是目前计算机科学界的热门研究领域。ArthurSamuel定义