Research map / Learning theory and algorithms / Matrix and tensor methods
Scalable Robust Matrix Factorization with Nonconvex Loss
Quanming Yao, James Kwok
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 nonconvex LSP loss rather than the convex loss in RMF-MM | Matrix factorization · Synthetic data, m=250, m=500, and m=1000 | RMSE: 0.194$\pm$0.032 -> 0.110$\pm$0.004; 0.145$\pm$0.009 -> 0.073$\pm$0.001; 0.122$\pm$0.004 -> 0.047$\pm$0.002 | Improved |
| Use nonconvex LSP loss rather than the convex loss in RMF-MM | Matrix factorization · MovieLens-100K | RMSE: 0.901$\pm$0.003 -> 0.885$\pm$0.006 | Improved |
| Use nonconvex LSP loss rather than the convex loss in RMF-MM | Matrix factorization · MovieLens-1M and MovieLens-10M | RMSE | Improved |
| Use nonconvex LSP loss rather than the convex loss in RMF-MM(heuristic) | Matrix factorization · Dinosaur D1, D2, and D3 | MAE: 0.374$\pm$0.031 -> 0.323$\pm$0.012; 0.381$\pm$0.022 -> 0.332$\pm$0.005; 0.382$\pm$0.034 -> 0.316$\pm$0.006 | Improved |
| Exploit data sparsity in surrogate optimization (ablation) | RMFNL · Synthetic m=1000 | CPU time: 490.1$\pm$91.9 -> 24.6$\pm$3.2 | Improved |
| Replace APG with LADMPSAP for surrogate optimization (ablation) | RMFNL · Synthetic m=1000 | CPU time: 24.6$\pm$3.2 -> 950.8$\pm$138.8 | Hurt |
| Replace APG with APG(dense) for surrogate optimization (ablation) | RMFNL · Synthetic m=1000 | RMSE: 0.45$\pm$0.006 -> 0.45$\pm$0.007 | No clear effect |
| Replace LSP loss with Geman or Laplace loss (ablation) | RMFNL · Synthetic data and MovieLens | RMSE | No clear effect |
| Gradually increase the diagonal entries of the surrogate weighting matrices (ablation) | RMF-MM · Dinosaur D1, D2, and D3 | MAE: 0.442$\pm$0.096 -> 0.374$\pm$0.031; 0.458$\pm$0.043 -> 0.381$\pm$0.022; 0.466$\pm$0.072 -> 0.382$\pm$0.034 | Improved |
Closest papers by meaning
- Robust Matrix Factorization by Majorization Minimization (TPAMI 2017 · similarity 0.90)
- Non-Convex Matrix Completion and Related Problems via Strong Duality (JMLR 2019 · similarity 0.89)
- Factor Group-Sparse Regularization for Efficient Low-Rank Matrix Recovery (NeurIPS 2019 · similarity 0.89)
- Preconditioned Gradient Descent for Over-Parameterized Nonconvex Matrix Factorization (NeurIPS 2021 · similarity 0.88)
- RGNMR: A Gauss-Newton method for robust matrix completion with theoretical guarantees (NeurIPS 2025 · similarity 0.88)
- Guarantees of a Preconditioned Subgradient Algorithm for Overparameterized Asymmetric Low-rank Matrix Recovery (ICML 2025 · similarity 0.86)
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.
