Best-Arm Identification with Generative Proxy

📄 arXiv: 2607.06879v1 📥 PDF

作者: Tianyi Ma, Hanzhang Qin, Ruihao Zhu, Jierui Zuo

分类: cs.LG, stat.ML

发布日期: 2026-07-08


💡 一句话要点

提出PROBE算法以解决最佳臂识别中的样本效率问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 最佳臂识别 样本效率 控制变量 异方差识别 普通最小二乘法 机器学习 在线学习

📋 核心要点

  1. 现有的最佳臂识别方法在面对昂贵的奖励观察时,样本效率较低,难以有效利用廉价的代理评分。
  2. 论文提出的PROBE算法通过控制变量调整,将问题转化为异方差识别,利用普通最小二乘法维护残差方差的上界。
  3. 实验表明,PROBE在合成实例和实际应用中表现出显著的样本节省,验证了理论的有效性和实用性。

📝 摘要(中文)

最佳臂识别是数据驱动决策中的经典模型,但在许多应用中,每次奖励观察的成本较高。本文研究了一种固定置信度的最佳臂识别方法,其中每次昂贵的奖励拉取都与一个廉价但相关的代理评分配对。通过控制变量调整,将该模型转化为异方差识别问题,提出了PROBE算法,该算法通过普通最小二乘法直接维护残差方差的上界,确保了在未知相关性的情况下证书的有效性。实验结果表明,PROBE的样本节省与奖励-代理相关性的强度成正比,验证了理论预测的准确性。

🔬 方法详解

问题定义:本文旨在解决最佳臂识别中的样本效率问题,现有方法在使用昂贵的奖励观察时,无法有效利用廉价的代理评分,导致样本浪费。

核心思路:论文的核心思路是通过控制变量调整,将最佳臂识别转化为异方差识别问题,并提出PROBE算法,利用普通最小二乘法直接维护残差方差的上界,从而提高样本效率。

技术框架:整体架构包括两个主要阶段:第一阶段是离线估计代理的边际均值,第二阶段是在线学习奖励与代理之间的相关性。PROBE算法在在线学习过程中,动态调整残差方差的上界。

关键创新:最重要的技术创新在于将控制变量调整引入异方差识别问题,并通过普通最小二乘法确保证书的有效性,这与现有方法在处理未知相关性时的保守性形成鲜明对比。

关键设计:在算法设计中,关键参数包括代理评分的边际均值和残差方差的上界,损失函数采用普通最小二乘法的形式,确保在未知相关性下仍能保持算法的有效性。算法还在$(ε,δ)$-PAC设置下进行了扩展。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示,PROBE算法在合成数据和实际的汽车贷款定价重放中,样本节省与奖励-代理相关性的强度成正比,验证了理论预测的准确性。具体而言,PROBE在样本复杂度上相较于基线方法有显著提升,达到了已知相关性下的最优样本复杂度。

🎯 应用场景

该研究的潜在应用领域包括在线广告投放、推荐系统和金融决策等场景,能够有效降低决策成本,提高样本利用效率。随着机器学习和大语言模型的普及,PROBE算法在实际应用中具有重要的价值,能够帮助决策者在高成本环境中做出更优选择。

📄 摘要(原文)

Best-arm identification is a canonical model for data-driven decision-making, but in many applications each reward observation is costly. Motivated by the growing availability of cheap predictions from machine learning and large language models, we study fixed-confidence best-arm identification in which each costly reward pull is paired with a cheap but correlated proxy score. The marginal mean of the proxy can be estimated offline and is treated as known, whereas its correlation $ρ$ with the reward, which governs how much the proxy helps, is unknown and must be learned online in pair with real rewards. We show that a control-variate adjustment turns this model into a heteroscedastic identification problem whose oracle sample complexity improves by residual variance $1-ρ^2$. The central difficulty is that the correlation must be learned from the same costly samples that identification consumes online, and that a plug-in estimate of the residual variance is anti-conservative and can compromise correctness. We propose PROBE (PRoxy OLS for Best-arm Exploration), a phase-elimination algorithm that directly maintains an upper certificate on the residual variance with an ordinary least squares fit, whose exact chi-square law keeps the certificate valid regardless of the unknown correlation. We prove that PROBE is $δ$-PAC and attains the known-correlation oracle sample complexity up to a constant multiplicative factor and a constant additive calibration cost. The guarantee extends to the $(ε,δ)$-PAC setting under minimal changes to the algorithm. Numerical experiments on synthetic instances and on an auto-loan pricing replay with large language model and tabular proxies confirm that the sample savings of PROBE scale with the strength of the reward-proxy correlation, exactly as the theory predicts.