Fast Recursive Nonnegative Standard and Hierarchical Tucker Decomposition

This letter proposes a new approach to nonnegative Tucker decomposition, which assumes recursive updates of latent factors with any nonnegative matrix factorization algorithm. The proposed strategy is extended to the nonnegatively constrained hierarchical Tucker decomposition model. Numerical experiments confirmed that the proposed algorithms have lower computational complexities and demonstrate improved performance compared to their baseline counterparts.