Research map / Learning theory and algorithms / Combinatorial and search optimization
DC3: A learning method for optimization with hard constraints
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 equality completion so the network predicts only partial variables and the remaining variables are solved from equality constraints (ablation) | DC3 · Convex QP | max equality violation: 0.35 (0.00) -> 0.00 (0.00) | Helped |
| Add correction during training (ablation) | DC3 · Convex QP | max inequality violation: 0.02 (0.02) -> 0.00 (0.00) | Helped |
| Add correction during both training and testing (ablation) | DC3 · Convex QP | max inequality violation: 0.09 (0.13) -> 0.00 (0.00) | Helped |
| Retain equality and inequality violation terms in the training loss (ablation) | DC3 · Convex QP | max inequality violation: 23.83 (0.11) -> 0.00 (0.00) | Helped |
| Add equality completion (ablation) | DC3 · Simple non-convex optimization | max equality violation: 0.35 (0.00) -> 0.00 (0.00) | Helped |
| Add correction during training (ablation) | DC3 · Simple non-convex optimization | max inequality violation: 0.01 (0.01) -> 0.00 (0.00) | Helped |
| Retain equality and inequality violation terms in the training loss (ablation) | DC3 · Simple non-convex optimization | max inequality violation: 15.21 (0.04) -> 0.00 (0.00) | Helped |
| Add equality completion (ablation) | DC3 · ACOPF | max equality violation: 0.14 (0.01) -> 0.00 (0.00) | Helped |
| Add correction during training (ablation) | DC3 · ACOPF | max inequality violation: 0.00 (0.00) -> 0.00 (0.00) | No clear effect |
| Add correction during both training and testing (ablation) | DC3 · ACOPF | max inequality violation: 0.01 (0.00) -> 0.00 (0.00) | Helped |
| Retain equality and inequality violation terms in the training loss (ablation) | DC3 · ACOPF | max equality violation: 2.60 (0.35) -> 0.00 (0.00) | Helped |
| Use DC3 instead of the neural network trained with a soft constraint loss | Deep-learning optimization solver · Convex QP | max equality violation: 0.35 (0.00) -> 0.00 (0.00) | Helped |
Closest papers by meaning
- Inverse Optimization via Learning Feasible Regions (ICML 2025 · similarity 0.84)
- The Multi-Block DC Function Class: Theory, Algorithms, and Applications (NeurIPS 2026 · similarity 0.83)
- FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees (NeurIPS 2025 · similarity 0.83)
- AT-SKM-Net: An Accelerated Trainable Sampling Kaczmarz-Motzkin Framework for Linear Hard-Constraint Feasibility on Dynamic Graphs (NeurIPS 2026 · similarity 0.83)
- Compact Optimality Verification for Optimization Proxies (ICML 2024 · similarity 0.82)
- Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules (NeurIPS 2025 · 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.