Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
End-to-End Learning for Stochastic Optimization: A Bayesian Perspective
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 decision map directly on task loss using constraint-aware layers instead of predict-then-optimize regression | Economic dispatch · Myopic observation, 10 minute frequency | average test cost: 281.436(1.652) -> 68.251(3.73) | Helped |
| use myopic context rather than incomplete myopic context | E2E-CAL · Economic dispatch, 10 minute frequency | average test cost: 72.616(0.003) -> 68.251(3.73) | Helped |
| use historical context rather than myopic context | E2E-CAL · Economic dispatch, 30 minute frequency | average test cost: 66.566(3.068) -> 77.06(5.585) | Hurt |
| use optimization layers with the Softplus prediction output | Economic dispatch · Myopic observation, 10 minute frequency | average test cost: 68.251(3.73) -> 71.33(1.87) | Hurt |
| replace the Softplus output activation with ReLU (ablation) | E2E-OPL · Economic dispatch, Myopic observation, 10 minute frequency | average test cost: 71.326(1.872) -> 72.601(0.0) | Hurt |
| use Lag-1 instead of E2E-CAL with myopic information | Economic dispatch · 10 minute frequency | average test cost: 68.251(3.73) -> 64.959 | Helped |
| use Lag-1 instead of E2E-CAL with myopic information | Economic dispatch · 30 minute frequency | average test cost: 66.566(3.068) -> 67.705 | Hurt |
| use standard E2E training with optimization layers | Economic dispatch · Myopic, Myopic Incomp., and Historical observations | average test cost | Mixed |
| use a Bayesian decision map trained under a prior that matches deployment | Bayesian newsvendor strategies · Newsvendor | out-of-sample profit | Helped |
| use a Bayesian strategy under a misspecified prior | BAY and NN_BAY · Newsvendor with distribution shift | out-of-sample profit | Hurt |
| use ERM rather than the other newsvendor strategies | Newsvendor · Out-of-sample profit lower tail | risk of negative profit: almost 10 times higher | Hurt |
Closest papers by meaning
- The Perils of Learning Before Optimizing (AAAI 2022 · similarity 0.90)
- Task-based End-to-end Model Learning in Stochastic Optimization (NeurIPS 2017 · similarity 0.89)
- NOVAS: Non-convex Optimization via Adaptive Stochastic Search for End-to-end Learning and Control (ICLR 2021 · similarity 0.87)
- Learning and Policy Search in Stochastic Dynamical Systems with Bayesian Neural Networks (ICLR 2017 · similarity 0.87)
- A Generalization Result for Convergence in Learning-to-Optimize (ICML 2025 · similarity 0.86)
- Decision-Focused Learning with Directional Gradients (NeurIPS 2024 · similarity 0.85)
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.