Double-Matched Matrix Decomposition for Multi-View Data

We consider the problem of extracting joint and individual signals from multi-view data, that is, data collected different sources on matched samples. While existing methods for decomposition explore single matching by samples, we focus double-matched (matched both samples source features). Our motivating example is miRNA primary tumor normal tissues same subjects; measurements two are thus subjects miRNAs. proposed matrix allows us to simultaneously extract across subjects, as well estimation approach takes advantage double-matching formulating a new type optimization with explicit row space column constraints, which develop an efficient iterative algorithm. Numerical studies indicate taking leads superior signal performance compared based single-matching. apply our method English Premier League soccer matches find align domain-specific knowledge. Supplementary materials this article available online.