Research map / Learning theory and algorithms / Prediction and decision losses
Decision-Focused Learning without Decision-Making: Learning Locally Optimized Decision Losses
Sanket Shah, Kai Wang, Bryan Wilder, Andrew Perrault, Milind Tambe
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 DirectedQuadratic LODL instead of 2-stage MSE | predictive model · Web Advertising | normalized DQ: 0.48 -> 0.91 | Improved |
| Train with DirectedQuadratic LODL instead of 2-stage MSE | predictive model · Linear Model | normalized DQ: -0.95 -> 0.96 | Improved |
| Train with DirectedQuadratic LODL instead of 2-stage MSE | predictive model · Portfolio Optimization | normalized DQ: 0.32 -> 0.33 | Improved |
| Replace handcrafted DFL surrogate with learned Quadratic LODL | predictive model · Web Advertising | normalized DQ: 0.85 -> 0.93 | Improved |
| Replace handcrafted DFL surrogate with DirectedQuadratic LODL | predictive model · Linear Model | normalized DQ: 0.83 -> 0.96 | Improved |
| Replace DFL with DirectedQuadratic LODL | predictive model · Portfolio Optimization | normalized DQ: 0.35 -> 0.33 | Hurt |
| Use non-convex NN LODL instead of convex DirectedQuadratic | LODL · Portfolio Optimization | normalized DQ: 0.33 -> -0.11 | Hurt |
| Add directionality to WeightedMSE LODL | LODL · Linear Model | normalized DQ: -0.93 -> 0.96 | Improved |
| Add directionality to WeightedMSE LODL | LODL · Web Advertising | normalized DQ: 0.58 -> 0.53 | No clear effect |
| Add cross-prediction (quadratic) terms to WeightedMSE LODL | LODL · Web Advertising | normalized DQ: 0.58 -> 0.93 | Improved |
| Choose sampling strategy 2-Perturbed vs All-Perturbed (ablation) | NN and DirectedWeightedMSE LODL · Web Advertising | normalized DQ | Mixed |
| Increase number of samples used to fit LODL (ablation) | LODL · Web Advertising | normalized DQ | Improved |
Closest papers by meaning
- Locally Convex Global Loss Network for Decision-Focused Learning (AAAI 2025 · similarity 0.92)
- Solver-Free Decision-Focused Learning for Linear Optimization Problems (NeurIPS 2025 · similarity 0.92)
- Feasibility-Aware Decision-Focused Learning for Predicting Parameters in the Constraints (NeurIPS 2025 · similarity 0.89)
- Decision-Focused Learning with Directional Gradients (NeurIPS 2024 · similarity 0.88)
- DFF: Decision-Focused Fine-Tuning for Smarter Predict-Then-Optimize with Limited Data (AAAI 2025 · similarity 0.88)
- Leaving the Nest: Going beyond Local Loss Functions for Predict-Then-Optimize (AAAI 2024 · 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.
