Geometrical Formulation of the Nonnegative Matrix Factorization

Shotaro Akaho, Hideitsu Hino, Neneka Nara, Noboru Murata

Abstract

Nonnegative matrix factorization (NMF) has many applications as a tool for dimension reduction. In this paper, we reformulate the NMF from an information geometrical viewpoint. We show that a conventional optimization criterion is not geometrically natural, thus we propose to use more natural criterion. By this formulation, we can apply a geometrical algorithm based on the Pythagorean theorem. We also show the algorithm can improve the existing algorithm through numerical experiments.
BibTeX
@inproceedings{akaho2018geometrical,
  author = {Shotaro Akaho and Hideitsu Hino and Neneka Nara and Noboru Murata},
  title = {{Geometrical Formulation of the Nonnegative Matrix Factorization}},
  booktitle = {ICONIP},
  year = {2018},
  doi = {10.1007/978-3-030-04182-3\_46},
}

Loading related content...

Figure