Research map / Learning theory and algorithms / Prediction and decision losses
Learning Predictions for Algorithms with Predictions
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 |
|---|---|---|---|
| Apply online learning to a convex upper bound on algorithm cost that depends on prediction quality | Algorithms with predictions across matching, page migration, scheduling, and ski-rental · Adversarial online instances | regret: Table 1 gives regret bounds of \mathcal O\left(n\sqrt T\right), \mathcal O\left(\sqrt{mT\log(mT)}\right), and problem-sp | Helped |
| Use online-to-batch conversion of the online learning procedure | Minimum-weight bipartite matching · I.i.d. cost instances | sample complexity: a \tilde{\mathcal O}(n) lower sample complexity | Helped |
| Relax discrete request predictions to probability vectors and optimize the convex maximum-over-expectations bound | Online page migration · Request sequences | competitive ratio: \mathcal O\left(\frac{\log n}{\gamma D}\right) additive error over [cite] in the case of a perfect predictor | Helped |
| Learn a linear map from instance features to predictions | Algorithms with predictions · Feature-dependent instances | learning guarantees | Helped |
| Use KT-OCO without a bounded predictor domain | Online job scheduling · Adversarial machine-weight and feature pairs | regret: \|[0m\*A\|_F\sqrt{T\log(1+24T^2\|\*A\|_F^2)}+1 | Helped |
| Tune the robustness-consistency parameter with an exponential forecaster | Preferential round-robin job scheduling · Online job scheduling with predictions | regret: 9B\left(1+\sqrt{\frac T2\log T}\right) | Helped |
| Jointly learn the buy threshold and trade-off parameter | Discrete ski-rental · Ski-rental with integer days | regret: 6N\sqrt{T\log(BNT)} | Helped |
| Use a dispersion assumption to learn discontinuous threshold losses | Continuous ski-rental · β-dispersed ski-days | regret: \tilde{\mathcal O}\left(\sqrt{T\log(NT)}+(N+B)^2T^{1-\beta}\right) | Helped |
| Replace the log-concave data assumption with β-dispersion | Continuous ski-rental · I.i.d. ski-days | assumption strength | Helped |
| Replace the maximum in the prediction-error objective with a simple sum | Online page migration · Request sequences | worst-case bound: \mathcal O\left(\frac n{\gamma D}\right) | Hurt |
Closest papers by meaning
- Optimal Robustness-Consistency Trade-offs for Learning-Augmented Online Algorithms (NeurIPS 2020 · similarity 0.91)
- Learning-Augmented Algorithms with Explicit Predictors (NeurIPS 2024 · similarity 0.90)
- Algorithms with Calibrated Machine Learning Predictions (ICML 2025 · similarity 0.89)
- Online Algorithms with Uncertainty-Quantified Predictions (ICML 2024 · similarity 0.88)
- Online Classification with Predictions (NeurIPS 2024 · similarity 0.86)
- The Stability of Online Algorithms in Performative Prediction (NeurIPS 2026 · 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.