AI Papers Daily

#Reward-Coverage Tradeoff

1 篇

Logging Policy Design for Off-Policy Evaluation

面向离策略评估的日志策略设计

🧭 Eval ★ 深度解读

New York University · Spotify

在已知/未知目标策略与奖励分布的信息条件下，推导最小化IPW估计均方误差的最优日志策略，揭示奖励-覆盖权衡

Eval | 2026-05-14 | cs.IR | score 8/10

Off-Policy EvaluationLogging PolicyInverse Propensity WeightingReward-Coverage TradeoffExperimental Design