大数据环境下基于马氏重抽样的分布式集成学习的理论与算法

分布式集成学习是目前大数据处理的一个基本方法。本项目拟通过对大数据分割后的小数据集进行马氏重抽样，使得多次马氏重抽样后的马氏样本序列较快达到数据的极限分布或平稳分布，从而应用数据分布的信息对小数据集进行价值提纯，优化分布式学习的负载均衡和计算效率。通过对基于马氏重抽样的基本学习机有较好泛化性能和较大差异性所满足的条件，以及基于马氏重抽样的基本学习机的集成方式和合并准则等问题的研究，建立大数据环境下基于马氏重抽样的分布式集成学习的理论框架，设计出适合大数据特性的基于马氏重抽样的分布式集成学习的新算法，并将理论和算法研究成果应用到大数据处理的实际问题中，为大数据挖掘提供新思路和新方法。

The distributed ensemble learning is a fundamental method of the big data processing at present. In this project we plan to deal with the small data sets that big data divided into using the method of Markov resampling such that the Markov chains reach quickly the limiting distribution or the stationary distribution after many times Markov resampling. Then we can use the information of the distribution to purify the small data sets, optimize the load balancing and the computational efficiency of the distributed ensemble learning. The aim of this project is to establish the theoretical framework and design novel algorithms for the distributed ensemble learning based on Markov resampling, and then use them to deal with the practical problem of the big data processing and present some new idea and novel method for mining big data by studying the sufficient conditions that the basic learning machines based on Markov resampling have good generalization ability and the basic learning machines based on Markov resampling have large otherness, and establilshing the basic principle for the ensemble fashion and the combined criterion of the basic learning machines based on Markov resampling.