Research map / Learning theory and algorithms / Prediction and decision losses
Melding the Data-Decisions Pipeline: Decision-Focused Learning for Combinatorial Optimization
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 |
|---|---|---|---|
| Train the neural network directly on decision quality instead of a task-agnostic prediction loss | 1-layer neural network · Budget allocation | solution quality: 32.13 $\pm$ 2.47 -> 49.18 $\pm$ 0.24 (k=5); 45.63 $\pm$ 3.76 -> 72.62 $\pm$ 0.33 (k=10); 61.88 $\pm$ 4.10 -> 98.95 $\pm$ | Helped |
| Train the neural network directly on decision quality instead of a task-agnostic prediction loss | 2-layer neural network · Budget allocation | solution quality: 9.69 $\pm$ 0.05 -> 44.35 $\pm$ 0.56 (k=5); 18.93 $\pm$ 0.10 -> 67.64 $\pm$ 0.62 (k=10); 36.16 $\pm$ 0.18 -> 93.59 $\pm$ | Helped |
| Use decision-focused training rather than two-stage training | 2-layer neural network · Bipartite matching | solution quality: 3.49 $\pm$ 0.32 -> 6.15 $\pm$ 0.38 | Helped |
| Use decision-focused training rather than two-stage training | 1-layer neural network · Bipartite matching | solution quality: 2.99 $\pm$ 0.76 -> 2.50 $\pm$ 0.56 | Hurt |
| Use decision-focused training rather than two-stage training | 1-layer neural network · Diverse recommendation | solution quality: 4.08 $\pm$ 0.16 -> 15.81 $\pm$ 0.50 (k=5); 8.42 $\pm$ 0.29 -> 29.81 $\pm$ 0.85 (k=10); 19.16 $\pm$ 0.57 -> 52.43 $\pm$ 1 | Helped |
| Use decision-focused training rather than two-stage training | 2-layer neural network · Diverse recommendation | solution quality: 11.63 $\pm$ 0.43 -> 13.34 $\pm$ 0.77 (k=5); 22.79 $\pm$ 0.66 -> 26.32 $\pm$ 1.38 (k=10); 42.37 $\pm$ 1.02 -> 47.79 $\pm$ | Helped |
| Use a 1-layer rather than a 2-layer neural network | Decision-focused learning · Budget allocation | solution quality: 44.35 $\pm$ 0.56 -> 49.18 $\pm$ 0.24 (k=5); 67.64 $\pm$ 0.62 -> 72.62 $\pm$ 0.33 (k=10); 93.59 $\pm$ 0.77 -> 98.95 $\pm$ | Helped |
| Use a 2-layer rather than a 1-layer neural network | Decision-focused learning · Bipartite matching | solution quality: 2.50 $\pm$ 0.56 -> 6.15 $\pm$ 0.38 | Helped |
| Standard prediction accuracy is not a reliable proxy for decision utility | Decision-focused and two-stage models · Budget allocation, bipartite matching, and diverse recommendation | solution quality | Mixed |
| Replace decision-focused training with two-stage training | 1-layer neural network · Budget allocation, MSE | MSE: 0.8673e-02 $\pm$ 1.83e-04 -> 0.0501e-02 $\pm$ 2.67e-06 | Helped |
Closest papers by meaning
- Decision-Focused Learning: Through the Lens of Learning to Rank (ICML 2022 · similarity 0.93)
- Solver-Free Decision-Focused Learning for Linear Optimization Problems (NeurIPS 2025 · similarity 0.91)
- What Data Enables Optimal Decisions? An Exact Characterization for Linear Optimization (NeurIPS 2025 · similarity 0.88)
- Leaving the Nest: Going beyond Local Loss Functions for Predict-Then-Optimize (AAAI 2024 · similarity 0.87)
- Probing Neural Combinatorial Optimization Models (NeurIPS 2025 · similarity 0.87)
- Feasibility-Aware Decision-Focused Learning for Predicting Parameters in the Constraints (NeurIPS 2025 · similarity 0.87)
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.