【主讲】张任宇，香港中文大学商学院，副教授
【主题】基于深度学习大型在线平台多实验因果推断
【时间】2022 年 10 月 7 日(周五)10:30am-12:00pm
【地点】Zoom 会议:814 6475 8150 密码: 1007
【语言】英文
【主办】管科系
摘要: 大型在线平台每天都会启动数千个随机实验(也称为A/B测试)来迭代其业务策 略。因此，平台的每个用户可能会同时被大量 A/B 测试命中。这就引发了如下两 个对学术研究与平台运营实践都非常重要的问题:(a)如何估计和推断平台上 多个实验组合的整体效果?(b) 在无法观察到所有实验组合的情况下，如何找到 最佳实验组合(i.e., best-arm identification)?我们结合深度学习 (Deep Learning) 和双重机器学习 (Double Machine Learning) 的开发一套新的统计分析 框架来估计平台每个用户受到任何实验组合的异质性处理效果 (Heterogeneous Treatment Effect)。我们提出的神经网络架构兼顾了可解释性和灵活性。我们的框 架(称作 debiased deep learning，DeDL)利用 Neyman 正交性产生了一致且渐近 正态的估计量，从而进行有效实验效果推断与最佳实验组合识别。我们与大型短 视频平台(平台 O)合作，部署了我们的框架分析平台 O 上 3 个独立 A/B 测试。 与基于线性回归和深度学习的基准方法相比，我们的 DeDL 方法可以更准确地估 计和推断任意实验组合的效果，并正确识别最佳实验组合。我们通过随机仿真数 据进一步验证 DeDL 框架在 model misspecification 下的稳健性。从应用的角度， 我们第一次在文献中通过大规模实验数据来验证理论上优雅的 DML 方法在因果 推理方面的实践价值。
关键词:深度学习;双重机器学习;因果推断;A/B 测试;在线平台实验
Bio: Prof. Philip Renyu Zhang joined The Chinese University of Hong Kong (CUHK) Business School as a visiting scholar in the Department of Decision Sciences and Managerial Economics in September 2021 and has been an Associate Professor since September 2022. Prior to joining CUHK Business School, he was an Assistant Professor of Operations Management at New York University (NYU) Shanghai and an NYU Global Network Assistant Professor since August 2016. Prof. Zhang holds a PhD degree in Business Administration (Operations Management) at Olin Business School, Washington University in St. Louis, and a Bachelor’s degree in Mathematics from Peking University.
Prof. Zhang’s research interests are developing data science methodologies (data- driven optimisation, causal inference, and machine learning) to evaluate and optimise the operations strategies in the contexts of online platforms and marketplaces, sharing economy, and social networks, especially their recommendation, advertising, pricing, and matching policies. His research works have appeared inManagement Science, Operations Research, and Manufacturing & Service Operations Management, and have been recognised by various research awards of the INFORMS and POMS communities. His reseach projects have been funded by NSFC, SMEC, STCSM, and HK RGC.
