Structural Laplacian Eigenmaps for Modeling Sets of Multivariate Sequences

摘要: A novel embedding-based dimensionality reduction approach, called structural Laplacian Eigenmaps, is proposed to learn models representing any concept that can be defined by a set of multivariate sequences. This approach relies on the expression intrinsic structure sequences in form constraints, which are imposed process generate compact and data-driven manifold low dimensional space. mathematical representation nature interest regardless stylistic variability found its instances. In addition, this extended model jointly several related concepts within unified creating continuous space between manifolds. Since generated encodes unique characteristic interest, it employed for classification unknown instances concepts. Exhaustive experimental evaluation different datasets confirms superiority methodology other state-of-the-art methods. Finally, practical value method demonstrated three challenging computer vision applications, i.e., view-dependent view-independent action recognition as well human–human interaction classification.