New methods improve counterfactual prediction in adaptive online learning

Researchers have introduced Propensity-Weighted Online Conformal Prediction (PW-OCP) and a doubly robust variant (DR-OCP) to address failures in online conformal prediction when predictions influence actions and outcomes. These new methods debias calibration by using inverse-propensity weighting, with DR-OCP further reducing bias by combining outcome-model and propensity errors. Experiments demonstrate that PW-OCP and DR-OCP enhance counterfactual coverage and reduce regret in various decision-making tasks without compromising prediction set sharpness, provided positivity conditions are met. AI

IMPACT Enhances robustness and accuracy in adaptive decision-making systems by improving counterfactual coverage.

RANK_REASON The cluster contains a research paper detailing new methods for online conformal prediction. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• DR-OCP
• Gotit.pub
• Hugging Face
• IArxiv
• Propensity-Weighted Online Conformal Prediction
• PW-OCP
• ScienceCast

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →