基于互信息的并行约简

Parallel reduction based on mutual information

将并行约简理论和信息论相结合,提出了信息论意义下的并行约简概念(原并行约简理论是代数意义下的并行约简),并利用互信息的概念提出了基于互信息的并行约简算法.信息论意义下的并行约简的优点在于:处理不一致数据时可以保留比代数意义下的并行约简更多的分类信息.最后从理论上探讨了信息论并行约简和代数并行约简的一致性和差异性,将两者进行了有机统一.

It was proposed parallel reduction in information view （ the original parallel reduction was in algebraic view） by combining the information theory with the parallel reduction theory, and the corresponding algorithm with the concept of mutual information was also presented. The parallel reduction in information view could preserve more classification information than parallel reducts in algebraic view when used to deal with inconsistent data. Moreover, the consistency and differences of these two parallel reduction were discussed and proposed a unified dealing.