HyperNTF: A hypergraph regularized nonnegative tensor factorization for dimensionality reduction

Tensor decomposition is an effective tool for learning multi-way structures and heterogeneous features from high-dimensional data, such as the multi-view images multichannel electroencephalography (EEG) signals, are often represented by tensors. However, most of tensor methods linear feature extraction techniques, which unable to reveal nonlinear structure within data. To address problem, a lot algorithms have been proposed simultaneously performs non-linear extraction. A representative algorithm Graph Regularized Nonnegative Matrix Factorization (GNMF) image clustering. normal 2-order graph can only model pairwise similarity objects, cannot sufficiently exploit complex samples. Thus, we propose novel method, named Hypergraph (HyperNTF), utilizes hypergraph connections among samples employs factor matrix corresponding with last mode Canonical Polyadic (CP) low-dimensional representation original Extensive experiments on synthetic manifolds, real-world datasets, EEG demonstrating that HyperNTF outperforms state-of-the-art in terms dimensionality reduction, clustering, classification.