摘要
条件随机场(Conditional Random Fields,CRF)是一种用于结构化数据标注的机器学习方法,可以应用于序列标注任务.样本训练中随着样本标签数量的增加,训练过程的计算时间呈非线性增长.利用GPU流处理器的多核计算单元和多级存储结构,在OPECNCL编程模型下采用并行计算方法提高样本训练的计算效率.实验结果表明,采用并行计算的性能相对于面向单核CPU环境下的单线程计算能获得16倍的计算加速比.
Conditional Random Field（CRF） is an approach of machine learning for structural data,such as computing sequence and so on.The cost of training computing will increase rapidly while the tagged samples are going up.Therefore,we proposed an effective approach to parallel computing conditional random field to enhance the efficiency of training samples based on OPENCL programming model.Thus,we can mostly use hundreds of computing cores and multiple levels of storing systems within GPU dataflow processors system.According to the results of experiments,we gain 16x speedup of computing performance compared to serial computing on single CPU.
出处
《小型微型计算机系统》
CSCD
北大核心
2011年第12期2392-2395,共4页
Journal of Chinese Computer Systems
基金
上海高校优秀青年教师科研基金项目(SLG10005)资助
上海理工大学光电学院教师创新基金项目(GDCX-Y-102)资助
AMD大学合作计划基金项目(SOW-02)资助
