The Surrogate Is Not the Reward: Post-Surrogate Primary-Outcome Acquisition in Contextual Bandits
stat.ML, cs.LG
Submitted: 2026-10-05
Updated: 2026-10-05
Terminology
Sources
- Prediction-Guided Active Experiments
- Best Arm Identification with LLM Judges and Limited Human
- Prediction-Powered Active Testing
- Active Multiple-Prediction-Powered Inference
- Doubly Robust Policy Evaluation and Learning
- Stratified Sampling for Model-Assisted Estimation with Surrogate Outcomes
- Surrogate-assisted optimal sampling for risk prediction under measurement constraints
- Pure Exploration Beyond Reward Feedback: The Role of Post-Action Context
- Correlation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing
- Evaluating Surrogates in Individualized Treatment Rules
Related papers
- Behavior of prediction performance metrics with rare events
- Optimal Estimation of Generic Dynamics by Path-Dependent Neural Jump ODEs
- A Posterior-Dynamics Framework for Imaging Inverse Problems with Pretrained Diffusion Priors
- Efficient quadratic entropy with distance sketches
- Embedding-Bias in Conditional Independence Testing
- From Log-Odds to Shapley Values: An Explanatory Geometry for the Weighted Naive Bayes Classifier