论文部分内容阅读
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 Fields (CRF) is a kind of machine learning method for structured data annotation, which can be applied to sequence annotation tasks. With the increase of the number of sample labels in training, the computational time of training process is nonlinear Growth.Using the multi-core computing unit and multi-level storage structure of GPU stream processor, parallel computing is used to improve the computational efficiency of the sample training under the OPECNCL programming model.The experimental results show that the performance of parallel computing is higher than that of single-core CPU Single-threaded calculation can get 16 times the calculated speedup.