The paper presents Wasserstein manifold nonnegative tensor factorization (WMNTF), a new approach to nonnegative tensor factorization (NTF), a tool for feature extraction from non-negative high-order data. The conventional NTF methods use Euclidean or Kullback-Leibler divergence, treating each feature equally, which often neglects the side-information of features. WMNTF overcomes this by minimizing the Wasserstein distance between the distribution of input tensorial data and the distribution of reconstruction, thereby accounting for correlation information of features and manifold information of samples. The method has been tested and has shown effectiveness compared to other NMF and NTF methods.
Publication date: 3 Jan 2024
Project Page: https://arxiv.org/abs/2401.01842v1
Paper: https://arxiv.org/pdf/2401.01842
