Extraordinary amounts of data are generated across disciplines owing to advanced technology. Such data richness, however, may yield difficulties in statistical model fitting and prediction either in terms of time cost or numerical stability. Subsampling has been an effective remedy to this problem over the past decade. In this talk, I will introduce two subsampling methods, model-dependent and model-free, for big data prediction. Theoretical and numerical results are provided to show the benefits of the proposed methods.
 
 
Keywords: Gaussian process regression, Partitioning estimate, Bagging