Research map / Probabilistic and scientific modeling / Molecules and proteins
Prediction-Powered Adaptive Inference with Pretrained AI Models for Contextual Bandits
Gabriel Sargent, Will Wei Sun, Zhengwu Zhang, Yufeng Liu
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 |
|---|---|---|---|
| use PPAI with a strong predictive model instead of labeled-only WLS | adaptive inference estimator · Simulated two-armed bandit, T=5000 | avg 90% CI width: 0.111 -> 0.011 | Improved |
| use PPAI with a poor (uninformative) predictive model instead of WLS | adaptive inference estimator · Simulated two-armed bandit, T=5000 | avg 90% CI width: 0.111 -> 0.021 | Improved |
| use a strong rather than poor predictor in PPAI (ablation) | PPAI · Simulated two-armed bandit, T=5000 | avg 90% CI width: 0.021 -> 0.011 | Improved |
| PPAI coverage with poor predictor at T=2500 | PPAI (f poor) · Simulated two-armed bandit, T=2500 | 90% CI coverage: 92% (WLS) -> 88% | Mixed |
| use PPAI with Gemini predictions and unlabeled users instead of WLS | movie recommendation inference · Movies Dataset, beta0 coord 3 | avg 90% CI length: 1.24 -> 0.21 | Improved |
| use PPAI instead of WLS on coverage | movie recommendation inference · Movies Dataset, beta1 coord 1 | 90% CI coverage: 82% -> 99% | Improved |
| use data-driven lambda tuning instead of fixed lambda (ablation) | PPAI · Simulated two-armed bandit (Appendix D) | variance | Improved |
| include the bias-correction term (ablation) | PPAI · Simulated two-armed bandit (Appendix D) | accuracy | Improved |
Closest papers by meaning
- Prediction-Powered Adaptive Shrinkage Estimation (ICML 2025 · similarity 0.88)
- Anytime-valid, Bayes-assisted, Prediction-Powered Inference (NeurIPS 2025 · similarity 0.87)
- FAB-PPI: Frequentist, Assisted by Bayes, Prediction-Powered Inference (ICML 2025 · similarity 0.87)
- Robust Sampling for Active Statistical Inference (NeurIPS 2025 · similarity 0.87)
- Simulation-Based Inference for Adaptive Experiments (NeurIPS 2025 · similarity 0.86)
- Adaptive auditing of AI systems with anytime-valid guarantees (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.
