Data Fusion for Heterogeneous Data and Its Applications

With the rapid development of data storage and cloud computing facilities, volume and velocity are no longer the bottlenecks of big data applications. Variety poses more challenges, as the data we obtain may come from extremely heterogeneous sources. Clearly, simple integration of different databases by collating data is not enough. Innovative data fusion approaches open up a wide range of research opportunities in big data research. This thesis will cover data fusion for large-scale data analysis in the following three levels. Feature level

fusion through semi-parametric model for heterogeneous data, data level fusion through optimal transport map for medical image data, and decision level fusion through ensemble learning for medical studies.