论文部分内容阅读
在多对称处理器集群体系结构下进行独立成分分析并行算法研究,在共享内存模型一级并行算法基础上,通过同步、异步迭代两种方式并行计算固定点函数,分别提出具有两级并行特性的二级同步、二级异步并行端元提取算法,并结合两者的优势,进一步提出二级分组并行算法。实验评价表明,二级同步、分组并行算法在保持原算法精度的同时,大大提高了原算法的效率,体现出良好的并行计算性能,而二级异步并行算法可在节点数较少的情况下适用。
Under the architecture of symmetric processor cluster, the independent component analysis parallel algorithm is studied. Based on the first level parallel algorithm of shared memory model, fixed point functions are calculated in parallel by both synchronous and asynchronous iteration, Two-stage synchronous and two-stage asynchronous parallel end-point extraction algorithm, combined with the advantages of both, further proposed a two-level packet parallel algorithm. The experimental evaluation shows that the second-order synchronization and packet parallel algorithm improve the efficiency of the original algorithm while maintaining the accuracy of the original algorithm, and show good parallel computing performance. However, the second-order asynchronous parallel algorithm can reduce the number of nodes Be applicable.