Research map / Optimization / Nonconvex and smooth optimization
Efficient Low-Rank Semidefinite Programming With Robust Loss Functions
Quanming Yao, Hansi Yang, Enliang Hu, James Tin-Yau 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 l1 loss with SDP-RL instead of square loss | low-rank PSD completion · Synthetic (m=2000) | testing RMSE: 0.615 -> 0.164 | Improved |
| use leaky-MCP nonconvex loss instead of l1 loss | SDP-RL · Synthetic (m=2000) | testing RMSE: 0.164 -> 0.113 | Improved |
| use SDP-RL(l1) instead of ADMM(l1) for the same l1 loss | PSD completion · Synthetic (m=2000) | CPU time (s): 2387 -> 37 | Improved |
| use SDP-RL(l1) instead of ADMM(l1) | PSD completion · Synthetic (m=2000) | testing RMSE: 0.332 -> 0.164 | Improved |
| use SDP-RL(MCP) instead of SDP-RL(l1) with increasing outlier fraction | SDP-RL · Synthetic (o=0.20) | testing RMSE: 0.161 -> 0.134 | Improved |
| use nonconvex MCP loss when no outliers | SDP-RL · Synthetic (o=0) | testing RMSE: 0.005 -> 0.007 | Hurt |
| increase maximum outlier amplitude sigma from 2.5 to 20 (ablation) | SDP-RL(MCP) · Synthetic | testing RMSE: 0.113 -> 0.112 | No clear effect |
| use SDP-RL(MCP) instead of SimpleNPKL | robust NPKL · 10% flipped labels | testing RMSE: 0.60 -> 0.26 | Improved |
| use SDP-RL(MCP) instead of SimpleCMVU | robust CMVU · 10% large outliers | testing RMSE: 0.97 -> 0.30 | Improved |
| use SDP-RL(MCP) instead of SDP-RL(l1) | sparse PCA · colon cancer r=200 | explained variance: 66.44 -> 67.92 | Improved |
| use SDP-RL(l1) instead of ADMM(l1) | sparse PCA · colon cancer r=200 | CPU time (s): 118.28 -> 7.42 | Improved |
| initialize X by PCA instead of zero (ablation) | SDP-RL(l1) · sparse PCA r=50 | CPU time (s): 0.21 -> 0.11 | Improved |
Closest papers by meaning
- Robust Optimization for Non-Convex Objectives (NeurIPS 2017 · similarity 0.87)
- Robust high dimensional learning for Lipschitz and convex losses (JMLR 2020 · similarity 0.87)
- Better generalization with less data using robust gradient descent (ICML 2019 · similarity 0.87)
- Descending, lifting or smoothing: Secrets of robust cost optimization (ECCV 2018 · similarity 0.87)
- Semi-infinite Nonconvex Constrained Min-Max Optimization (NeurIPS 2025 · similarity 0.85)
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.85)
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.
