The Workshop on Youth Statistician Forum: Prospect and Perspective in Statistical Inference and its Application

Quantile Correlation Screening for Ultrahigh Dimensional Heterogeneous Data Xiaolin Chen, China University of Petroleum In this talk, we systematically investigate the quantile correlation sure independence screening (QC-SIS) for ultrahigh dimensional heterogeneous data. We unveil the question under which assumptions and correlation structures QC-SIS enjoys the ranking consistency and sure screening properties when the number of features grows at an exponential rate of the sample size. Due to the nature of marginal independence screening, QC-SIS may miss truly active features which are marginally independent of the response, but contribute to the response jointly with other predictors. In addition, when there are many irrelevant predictors which are highly correlated with some strongly active predictors, QC-SIS may fails to identify other active predictors with relatively weak marginal signals. To enhance the finite sample performance, an iterative quantile correlation sure independence screening (QC-ISIS) is suggested based on the partial quantile correlation. Extensive simulation studies are carried out to examine the performances of QC-SIS and QC-ISIS with its main competitors. We also illustrate the proposed QC-SIS and QC-ISIS by real data examples. Joint work with Dr. Chunling Liu and Prof. Kam Chuen Yuen. Some Thoughts on Probabilistic Data Integration Xiaodan Fan, The Chinese University of Hong Kong Many scientific truths can be probed from multiple available datasets produced by different research groups. How to weight their relative trustworthiness is a common challenge. This problem becomes more and more common in the ear of big data. Probabilistic models for integrating these datasets can be a principled way for weighting, but there are many unsolved aspects. Based on our experience on integrative clustering for large heterogeneous data sets, we will discuss on the prospects and challenges of probabilistic data integration.