Research map / Learning theory and algorithms / Combinatorial and search optimization
Two-Stage Predict+Optimize for MILPs with Unknown Parameters in Constraints
Xinyi HU, Jasper C.H. Lee, Jimmy H.M. Lee
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 |
|---|---|---|---|
| train with 2S instead of IntOpt-C | MILP predict+optimize · Alloy production, Brass | mean post-hoc regret: 45.27 -> 43.87 | Improved |
| train with 2S instead of IntOpt-C | MILP predict+optimize · Alloy production, Titanium-alloy | mean post-hoc regret: 4.72 -> 4.52 | Improved |
| evaluate with Two-Stage framework instead of Hu et al. framework (ablation) | IntOpt-C/2S · Alloy production, Brass | mean post-hoc regret: 68.16 -> 43.87 | Improved |
| evaluate with Two-Stage framework instead of Hu et al. framework (ablation) | 2S · Alloy production, Titanium-alloy | mean post-hoc regret: 6.45 -> 4.52 | Improved |
| train with 2S instead of best classical regression (NN) | predict+optimize · Alloy production, Brass, penalty 8 | mean post-hoc regret: 355.96 -> 194.06 | Improved |
| train with 2S instead of best classical regression (RF) | predict+optimize · Alloy production, Titanium-alloy, penalty 8 | mean post-hoc regret: 32.16 -> 19.69 | Improved |
| train with 2S instead of CombOptNet | MILP predict+optimize · 0-1 knapsack, capacity 100 | mean post-hoc regret: 9.45 -> 1.26 | Improved |
| train with 2S instead of CombOptNet | MILP predict+optimize · 0-1 knapsack, capacity 250 | mean post-hoc regret: 20.42 -> 0.07 | Improved |
| train with 2S instead of best classical (CART) | predict+optimize · 0-1 knapsack, capacity 100 | mean post-hoc regret: 8.67 -> 1.26 | Improved |
| increase knapsack capacity from 100 to 250 (ablation) | 2S · 0-1 knapsack, penalty 0.21 | mean post-hoc regret: 1.26 -> 0.07 | Improved |
| train with 2S instead of NN | predict+optimize · Nurse scheduling, penalty 0.25 | mean post-hoc regret: 4.27 -> 3.94 | Improved |
| train with 2S instead of NN at large penalty | predict+optimize · Nurse scheduling, penalty 8.0 | mean post-hoc regret: 124.54 -> 46.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CombOptNet — CombOptNet: Fit the Right NP-Hard Problem by Learning Integer Programming Constraints (ICML 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-Stage Predict+Optimize for (Mixed Integer) Linear Programs (NeurIPS 2024 · similarity 0.92)
- Predicting Lagrangian Multipliers for Mixed Integer Linear Programs (ICML 2024 · similarity 0.87)
- RoME: Domain-Robust Mixture-of-Experts for MILP Solution Prediction across Domains (NeurIPS 2025 · similarity 0.87)
- Contrastive Predict-and-Search for Mixed Integer Linear Programs (ICML 2024 · similarity 0.86)
- Apollo-MILP: An Alternating Prediction-Correction Neural Solving Framework for Mixed-Integer Linear Programming (ICLR 2025 · similarity 0.86)
- Learning Generalized Linear Programming Value Functions (NeurIPS 2024 · 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.
