Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Off-Policy Evaluation and Learning from Logged Bandit Feedback: Error Reduction via Surrogate Policy
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 |
|---|---|---|---|
| replace known logging propensities with ML surrogate policy (NN) | IPS · Yeast | test Hamming loss: 4.571 -> 3.630 | Helped |
| replace known logging propensities with ML surrogate policy (NN) | IPS · Scene | test Hamming loss: 1.342 -> 1.086 | Helped |
| replace known logging propensities with ML surrogate policy (NN) | IPS · TMC | test Hamming loss: 3.023 -> 2.019 | Helped |
| replace known logging propensities with ML surrogate policy (NN) | IPS · LYRL | test Hamming loss: 1.108 -> 0.930 | Helped |
| add ML surrogate policy to POEM (MLPOEM-NN) | POEM · Yeast | test Hamming loss: 4.549 -> 3.477 | Helped |
| add ML surrogate policy to POEM (MLPOEM-NN) | POEM · LYRL | test Hamming loss: 0.981 -> 0.904 | Helped |
| add ML surrogate policy to POEM (MLPOEM-NN) | POEM · Scene | test Hamming loss: 1.143 -> 1.086 | Helped |
| MLPOEM-NN versus Norm-POEM | Norm-POEM · Scene / LYRL | test Hamming loss: Scene 1.045 vs 1.086; LYRL 0.799 vs 0.904 | Mixed |
| use ML surrogate policy instead of assuming uniform logging policy when logging policy unknown | IPS · Yeast | test Hamming loss: 5.893 -> 3.630 | Helped |
| use neural network instead of linear surrogate policy (ablation) | MLIPS · Scene, Yeast, TMC, LYRL | test Hamming loss: Yeast 3.778 -> 3.630; LYRL 1.025 -> 0.930 | No clear effect |
| replace known logging propensities with NN surrogate policy | IPS · Criteo ad placement | test score: 0.551 -> 0.572 | Helped |
| replace logging propensities with surrogate policy for gradient estimation | IPS · LYRL gradient estimation | Euclidean distance to reference gradient | Helped |
Closest papers by meaning
- Log-Sum-Exponential Estimator for Off-Policy Evaluation and Learning (ICML 2025 · similarity 0.90)
- Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning (NeurIPS 2024 · similarity 0.88)
- Cross-Domain Off-Policy Evaluation and Learning for Contextual Bandits (ICLR 2025 · similarity 0.88)
- A General Framework for Off-Policy Learning with Partially-Observed Reward (ICLR 2025 · similarity 0.88)
- Off-policy estimation with adaptively collected data: the power of online learning (NeurIPS 2024 · similarity 0.88)
- MultiScale Contextual Bandits for Long Term Objectives (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.