一种基于MapReduce并行框架的大规模矩阵乘法运算的实现

IMPLEMENTATION FOR LARGE SCALE MATRIX MULTIPLICATION ON MAPREDUCE PARALLEL FRAMEWORK

在机器学习算法中,矩阵乘法运算是一种基本运算。而扩大矩阵乘法的运算规模并降低其运算时间,将有利于满足机器学习算法处理大规模数据的要求。将MapReduce并行框架用于分块矩阵乘法,实现一种用于大规模矩阵乘法运算的方法。理论分析和实验结果表明该方法在处理大规模矩阵乘法上具有极大的潜能,并且随着计算节点的增加从而获得较好的加速比。

Among machine learning algorithms,matrix multiplication is a fundamental operation.It will help satisfy machine learning algorithm’s requirement to handle a large amount of data to expand the matrix multiplication size and reduce its computation time.The article applies MapReduce parallel framework to block matrix multiplication to implement a large-scale matrix multiplication method.Both theoretical analysis and experimental results illustrate that the approach reveals great potentialities on large-scale matrix multiplication.As the number of nodes increases,it gains even better speedup ratio.