Online Rank-Revealing Block-Term Tensor Decomposition

The so-called block-term decomposition (BTD) tensor model, especially in its rank-(Lr,Lr,1) version, has been recently receiving increasing attention due to enhanced ability represent systems and signals that are composed of block components rank higher than one, a scenario encountered numerous diverse applications. Its uniqueness approximation have thus thoroughly studied. challenging problem estimating the BTD model structure, namely number terms (rank) their individual (block) ranks, is crucial importance practice only started attract significant attention. In data-streaming scenarios and/or big data applications, where size grows time or processing can be done incrementally, it essential able perform selection computation recursive (online/incremental) manner. this paper, novel approach tracking proposed, based on idea imposing column sparsity jointly factors ranks as numbers factor columns non-negligible magnitude. vein, using new rank-revealing batch algorithm starting point, an online method alternating reweighted least squares (RLS) type developed shown computationally efficient fast converging, also allowing change time. memory efficiency evaluated favorably compared with those reported scheme a-priori knowledge ranks. Simulation results both synthetic real (video) reported, demonstrate effectiveness proposed selecting correct model.