跳到正文
返回论文列表
stat.ML提交于 已译

带 Kaplan--Meier 生存估计的多臂老虎机的方差自适应置信上界

Variance-Adaptive Upper Confidence Bounds for Multi-Armed Bandits with Kaplan--Meier Survival Estimation

Yangfan Cui · Andrew Papanicolaou

中文摘要

研究随机多臂老虎机问题:每个臂的奖励是预先指定时间点上的生存概率,观测结果是右删失的时间-事件变量。标准老虎机算法在此设置下面临延迟且不完整的观测。此前针对 Kaplan--Meier(KM)估计器的适配依赖 Hoeffding 型界,忽略了连续生存过程固有的异方差性。为弥补这一差距,提出 KM-UCB,一种将连续时间生存分析与离散序贯决策相结合的方差自适应策略。利用鞅论和连续影响函数,推导出以 Greenwood 方差公式为参数的固定队列经验 Bernstein 界。对于实时分配,保留 Greenwood 指数,并使用来自已完成记录的生存界来复核推荐臂。在追赶式报告下,该策略具有对数期望遗憾,其探索界中包含臂特定的方差项。在双层延迟模拟和乳腺癌试验数据的回顾性回放中评估原始策略,并在单独的配对模拟中评估修正。

关键要点

  1. 01问题:生存概率奖励下观测右删失且延迟,现有 KM 适配用 Hoeffding 界忽略异方差。
  2. 02方法:KM-UCB 结合鞅论与连续影响函数,导出基于 Greenwood 方差的经验 Bernstein 界。
  3. 03实时分配保留 Greenwood 指数,用已完成记录的生存界复核推荐臂。
  4. 04结果:追赶式报告下期望遗憾为对数级,探索界含臂特定方差项。
  5. 05评估:双层延迟模拟、乳腺癌试验数据回放及配对模拟验证修正。

解读

尚无解读。

原始英文摘要

arXiv:2611.24988v1 Announce Type: new Abstract: We study the stochastic multi-armed bandit problem in which the reward associated with each arm is the survival probability at a pre-specified time horizon, and the observed outcome is a right-censored time-to-event variable. Standard bandit algorithms face delayed and partial observations in this setting. Prior adaptations for the Kaplan--Meier (KM) estimator rely on Hoeffding-type bounds that ignore the inherent heteroscedasticity of continuous survival processes. To bridge this gap, we propose KM-UCB, a variance-adaptive policy that unites continuous-time survival analysis with discrete sequential decision-making. By leveraging martingale theory and continuous influence functions, we derive a fixed-cohort empirical Bernstein bound parameterized by Greenwood's variance formula. For realtime allocation, we retain the Greenwood index and use a survival bound from completed records to review the recommended arm. Under catch-up reporting, the policy has logarithmic expected regret with an arm-specific variance term in the exploration bound. We evaluate the original policy in a dual-layer delayed simulation and a retrospective replay of breast cancer trial data, and evaluate the correction in a separate paired simulation.

同方向论文 · stat.ML

查看全部 →