Research map / Learning theory and algorithms / Matrix and tensor methods
Generalized Separable Nonnegative Matrix Factorization
Junjun Pan, Nicolas Gillis
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use GS-FGM instead of GSPA | generalized separable NMF · Documents: tr23 (r=6) | Relative approximation quality (%): 63.66 -> 70.70 | Improved |
| use GS-FGM instead of GSPA | generalized separable NMF · Faces: Yale (r=38) | Relative approximation quality (%): 57.52 -> 68.26 | Improved |
| use GS-FGM instead of GSPA | generalized separable NMF · Faces: CBCL (r=49) | Relative approximation quality (%): 80.73 -> 83.10 | Improved |
| use generalized separable model instead of separable SPA-R | GS-FGM · Faces: CBCL (r=49) | Relative approximation quality (%): 83.10 -> 84.57 | Hurt |
| use generalized separable model instead of separable SPA-R | GS-FGM · Faces: Yale (r=38) | Relative approximation quality (%): 62.94 -> 68.26 | Improved |
| use GS prior instead of unconstrained NMF at low noise | GS-FGM · Synthetic middle-point with adversarial noise | Relative approximation error | Improved |
| use GS prior instead of unconstrained NMF at high noise | GS-FGM · Synthetic middle-point with adversarial noise | Relative approximation error | Hurt |
| use GS-FGM instead of NMF/MV-NMF for ground-truth recovery | GS-FGM · Synthetic random GS matrices | Distance to ground truth | Improved |
| use GS-FGM instead of GSPA for index recovery | GS-FGM · Synthetic random GS matrices | Accuracy: 100% for all epsilon <= 0.483 | Improved |
| use generalized separable factorization instead of NMF on real data | GS-FGM · Documents: NG10 (r=10) | Relative approximation quality (%): 91.64 -> 92.41 | Hurt |
| use GS-FGM instead of GSPA in run time | GS-FGM · Documents (average time) | Computation time (s): 0.091 -> 1.249 | Hurt |
| use two-sided SPA* heuristic instead of GSPA | SPA* · Documents: tr45 (r=10) | Relative approximation quality (%): 69.08 -> 68.20 | Hurt |
Closest papers by meaning
- Distributionally Robust and Multi-Objective Nonnegative Matrix Factorization (TPAMI 2021 · similarity 0.93)
- Dropping Symmetry for Fast Symmetric Nonnegative Matrix Factorization (NeurIPS 2018 · similarity 0.91)
- An Exterior Method for Nonnegative Matrix Factorization (ICML 2026 · similarity 0.91)
- Non-negative Matrix Factorization under Heavy Noise (ICML 2016 · similarity 0.91)
- Characterizing the Loss Landscape in Non-Negative Matrix Factorization (AAAI 2021 · similarity 0.90)
- Statistically Optimal $K$-means Clustering via Nonnegative Low-rank Semidefinite Programming (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
