大数据机器学习在数据原料、算法设计及优化求解等不同层次存在多种不确定性，给传统方法带来严重挑战。本课题利用机器学习、随机优化、分布式计算等方面的最新研究成果，着重开展“针对不可靠数据的学习原料提炼”、“面向动态多变环境的学习算法设计”、“针对大规模复杂问题的优化求解”和“面向脑疾病大数据的应用”等研究工作。特别是研究多源异构数据融合、主动标记查询、样本分布变化跟踪、语义迁移建模、标记变化检测、大规模非凸优化等新方法，在大数据机器学习不确定性建模的技术和方法上取得创新。并基于以上研究成果，对阿尔茨海默病、颅内动脉瘤等脑疾病医疗大数据任务进行典型应用。

In big data machine learning, there are various uncertainties in data source, algorithm design and optimization, which pose serious challenges for traditional methods. This project employs recent advances in machine learning, stochastic optimization and distributed computing to study the following research topics: "refinement of unreliable data", "learning algorithm designing for dynamic environment", "optimization for large scale problem with complex structures" and "application on brain disease analysis". More specifically, the project will focus on multi-sourced heterogeneous data fusion, active label querying, tracking of data distribution change, modeling on concept shift, detection of class label evolving and large scale optimization for non-convex problems, and try to make technique breakthroughs on modeling uncertainty for big data machine learning. Also, the research achievements will be applied to medical data analysis, especially for Alzheimer's disease and intracranial aneurysm.