Determining the dimension and structure of the subspace correlated across multiple data sets

Tanuj Hasija, Timothy Marrinan, Christian Lameiro, Peter J. Schreier

Abstract: Highlights•The paper proposes a novel method for determining the complete correlation structure among the components of multiple data sets.•It solves the model-selection problem of determining which components are common or correlated across all sets and which are correlated across a subset of data sets.•It derives theoretical identifiability conditions using tools from the graph theory.•It can be applied for jointly analyzing multimodal data in brain imaging, remote sensing, oceanography, genomics, where common, partially-common and distinct information is of interest.