Research map / Optimization / Nonconvex and smooth optimization
Homeomorphic Projection to Ensure Neural-Network Solution Feasibility for Constrained Optimization
Enming Liang, Minghua Chen, Steven H. Low
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 |
|---|---|---|---|
| add homeomorphic projection (H-Proj) as post-processing | Eq-NN-S and Eq-NN-U predictors · Convex QP/QCQP/SOCP/SDP | feasibility rate: 90.72 -> 100 (QP, Eq-NN-S); 49.32 -> 100 (QP, Eq-NN-U) | Improved |
| add homeomorphic projection (H-Proj) as post-processing | Eq-NN-S predictor · AC-OPF 30/57/118-node | feasibility rate: 77.73 -> 100 (30-node) | Improved |
| replace gradient-based D-Proj with H-Proj | Eq-NN-S · QP | feasibility rate: 90.82 -> 100 | Improved |
| use H-Proj instead of standard orthogonal projection (Proj) | Eq-NN-S · Convex QCQP | average speedup: 21.3 -> 8353 | Improved |
| use H-Proj instead of standard orthogonal projection (Proj) | Eq-NN-S · SOCP | average speedup: 13.6 -> 6724 | Improved |
| use H-Proj instead of Proj | Eq-NN-S · SDP | average speedup: 27.4 -> 369 | Improved |
| add H-Proj post-processing | Eq-NN-S · QP | objective error (all predictions): 0.45 -> 0.47 | Mixed |
| add H-Proj post-processing | Eq-NN-U · Convex QCQP | objective error (corrected infeasible only): 0.54 -> 0.78 | Hurt |
| add H-Proj post-processing | Eq-NN-S · 118-node AC-OPF | objective error (corrected infeasible only): 0.59 -> 2.44 | Hurt |
| use warm start (WS) instead of H-Proj | Eq-NN-S · QP | speedup and solution error: speedup 21.1 vs 184; solution error 1.46 vs 1.63 | Mixed |
| use 3-block INN rather than deeper (ablation) | INN approximating MDH mapping · 2-D parametric constraint set | constraint approximation | Improved |
| use infinity-norm ball instead of 2-norm ball as initial set (ablation) | INN MDH mapping · 2-D parametric constraint set | constraint approximation quality | No clear effect |
Closest papers by meaning
- Efficient Bisection Projection to Ensure Neural-Network Solution Feasibility for Optimization over General Set (ICML 2025 · similarity 0.88)
- FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees (NeurIPS 2025 · similarity 0.83)
- Inverse Optimization via Learning Feasible Regions (ICML 2025 · similarity 0.83)
- On the Power of Small-size Graph Neural Networks for Linear Programming (NeurIPS 2024 · similarity 0.82)
- A Study of First-Order Methods with a Deterministic Relative-Error Gradient Oracle (ICML 2024 · similarity 0.82)
- Projection-Free Online Convex Optimization with Time-Varying Constraints (ICML 2024 · similarity 0.82)
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.
