Zhiheng Zhang（张智恒）

Zhiheng Zhang（张智恒）

I am an interdisciplinary researcher primarily focusing on causal inference.

Zhiheng Zhang（张智恒）@ CAUSAL lab @ SUFE

欢迎访问张智恒的主页！本实验室名为CAUSAL (Causal Analysis for Underlying Structures And Learning)，致力于发展有理论基础和落地应用的因果推断方法，聚焦于潜在结构的刻画，并将其与现代学习算法和决策系统融合。请查看实验室因果推断入门指南。

张智恒自2025年8月起担任上海财经大学统计与数据科学学院常任轨助理教授。此前，他于清华大学交叉信息研究院（IIIS）获得博士学位。张智恒担任2025年度CCF-滴滴盖亚联合科研基金项目“基于强化学习的长期因果效应序贯决策问题”的联合负责人, AAAI 2025 人工智能会议 Causal Techniques 方向（AICT track）领域主席等。他正在寻找博士生/硕士生/实习生。他的邮箱是zhangzhiheng@mail.shufe.edu.cn。

📢 招生信息 / Recruitment

我正在寻找若干 博士(2026入学)/硕士生（2025、2026入学）。同时也长期招收科研实习生（支持远程线上实习）。

我对博士生的基本期望是：

  • 具备良好的品德和强烈的内驱力，喜欢科研探索（鼓励自主探索感兴趣的研究方向——我会在能力范围内全力指导；在能力范围外帮助建立合作）；
  • 满足以下两项中的任意一项：
    • 扎实的数学基础（尤其是概率统计，专业不限）；
    • 出色的编程能力。
我对硕士生的基本期望是：
  • 对于有志于学界的，我会跟博士生共同带领你进行科研；并支持你的继续深造；
  • 对于有志于业界的，我能够将你推荐到相关方向的前沿研究部门实习。
如果你对上述研究方向感兴趣，欢迎随时与我联系。

📢 论文讨论会 / Reading Group

欢迎对因果推断感兴趣的老师和同学参加我们的论文讨论会，可通过微信公众号CAUSAL-lab-SSDS-SUFE加入。

为理解并构建因果学习系统的核心理论结构，实验室的长期目标可概括为以下三个基础问题：

  1. 观测性研究中：如何系统刻画“模型假设—观测数据—可识别边界”之间的传导机制，从而揭示各种因果假设对可识别性的根本影响？

  2. 实验设计与推断中：如何定量描述“应用场景属性—实验设计与算法结构—统计效率”之间的性能极限，并基于此开发具有最优（或近最优）性质的设计与推断框架？

  3. 在线学习与决策中：如何从数学结构上统一机器学习、经济管理与统计推断等不同领域的优化目标，揭示它们之间的基本兼容性与最优可达边界？

为回答上述问题，实验室的总体研究路线遵循由理论到方法、由方法到实践的逐层递进结构：

  1. 从基础假设的违背出发，构建更具包容性的因果推断框架，例如：探索识别条件的可弱化性、部分识别边界等；

  2. 将这些基础结构与现代统计与机器学习方法融合，发展更高效、更稳定、更具可扩展性的识别与估计技术，例如：最优传输、代理变量与负控方法、共型预测、minimax优化、在线学习等；

  3. 进一步将理论与方法扩展至带有现实约束的任务设置，例如：输入/输出结构复杂、小样本学习、动态/缺失网络结构等；

  4. 最终形成能向现实场景有效辐射的因果推断体系，服务于社交网络分析、博弈论环境、优化决策，并落地于推荐系统、派单机制、市场干预策略、大模型行为建模等。

其中：1 关注输入层面更宽松的结构假设，2 聚焦算法层面精准且高效的识别—估计机制，3-4 主要面向输出层面复杂而贴近现实的决策与推断任务。

为逐步实现，目前实验室正主要围绕以下三个具体方向做研究：

  1. 网络结构下的在线实验设计与推断。

  2. 因果推断中的最优传输（OT）与几何结构.

  3. 面向工业现实约束的因果推断方法论。

他欢迎相关的交流或合作。若感兴趣，可邮件联系或添加他的WeChat。

Welcome to Zhiheng’s Lab! This lab is called CAUSAL (Causal Analysis for Underlying Structures And Learning), focusing on developing principled methods for causal analysis by uncovering underlying structures and integrating them with modern learning and decision-making systems. Please read the onboarding guide of CAUSAL lab.

Zhiheng Zhang (pronunciation: Zhee-hung Jahng) is a tenure-track Assistant Professor in the School of Statistics and Data Science at the Shanghai University of Finance and Economics, starting from 2025.08. Previously, he received his Ph.D. from the Institute for Interdisciplinary Information Sciences, Tsinghua University. Zhiheng Zhang is the co-Principal Investigator of the 2025 CCF-DiDi Gaia Collaborative Research Fund project titled “Sequential Decision-Making for Long-Term Causal Effects Based on Reinforcement Learning” and the area chair of AAAI2025 AICT track. He is looking for PhD/Master/Interns. His email is zhangzhiheng@mail.shufe.edu.cn.

Research Statement

To understand and articulate the core theoretical structure of causal learning systems, the long-term goals of the lab center on the following three foundational questions:

  1. In observational studies: How can we systematically characterize the transmission mechanism linking model assumptions, observed data, and identification boundaries, thereby revealing how different causal assumptions fundamentally shape identifiability?

  2. In experimental design and inference: How can we quantitatively describe the performance limits governing the interplay among application-specific characteristics, design and algorithmic structures, and statistical efficiency, and develop design and inference frameworks that achieve optimal (or near-optimal) performance?

  3. In online learning and decision-making: How can we mathematically unify the optimization objectives arising in machine learning, economic decision-making, and statistical inference, and reveal their intrinsic compatibilities, tensions, and optimal attainable frontiers?

To address these questions, the lab follows a research trajectory that proceeds from theory to methodology and from methodology to practice:

  1. Starting from violations of classical assumptions to construct more general causal inference frameworks, including weakened identification conditions and partial identification boundaries;

  2. Integrating these foundational structures with modern statistical and machine-learning techniques to develop more efficient, stable, and scalable methods for identification and estimation—for example, optimal transport, proxy and negative-control methods, conformal prediction, minimax optimization and online learning;

  3. Extending theory and methodology to settings with realistic constraints, such as complex input/output structures, limited sample regimes, and dynamic or partially observed network environments;

  4. Ultimately building a causal inference ecosystem capable of effectively transferring to real-world applications, including social network analysis, game-theoretic environments, and optimization-driven decision-making, with deployment in recommendation systems, dispatch mechanisms, market interventions, and large-model behavioral analysis.

In this progression, 1 targets more flexible structural assumptions at the input level, 2 focuses on precise and efficient identification–estimation mechanisms at the algorithmic level, and 3-4 address complex, realistic decision and inference tasks at the output level.

To advance this agenda, the lab is currently pursuing three concrete research directions:

  1. Online experimental design and inference under network interference;

  2. Optimal transport and geometric structures in causal inference;

  3. Causal inference methodologies tailored to industrial and real-world constraints.

Open to discussions and collaborations at any time. You can send him an email or add his WeChat.

Selected Honours

2018 Chinese Undergraduate Mathematics Competition Final (CMC), Gold Medal (Top 10 in China)

Academic Service

Reviewer: ICML, NeurIPs, ICLR, AISTATS, UAI, ACM Transactions on Information Systems (TOIS), Transactions of Mobile Computing (TMC), Journal of Machine Research (JMLR), Journal of the Royal Statistical Society, Series B (JRSSB)

Area Chair: AAAI 2025, Artificial Intelligence with Causal Techniques (AICT) track

Teaching

  1. Probability and Statistics (Tsinghua, 30470303-0), Yao Class TA, 2020-2021, Fall
  2. Probability and Statistics (Tsinghua, 30470303-0), Yao Class TA, 2021-2022, Fall
  3. Frontiers of Causal Inference (Tsinghua, 80470282-0): Graduate TA for Tsinghua University, 2022-2023, Spring
  4. Probability and Statistics (Tsinghua, 30470303-0), Yao Class TA, 2023-2024, Fall
  5. Frontiers of Causal Inference (Tsinghua, 80470282-0): Graduate TA for Tsinghua University, 2023-2024, Spring
  6. Advanced Applied Probability (Tsinghua, 40470503-0), Yao Class TA, 2024-2025, Fall
  7. Frontiers of Causal Inference (Tsinghua, 80470282-0): Graduate TA for Tsinghua University, 2024-2025, Spring
  8. 概率论 （上海财经大学，105494），统计与数据科学学院，本科生课，授课教师，2025-2026, Fall
  9. 数据分析与统计建模 （上海财经大学，213488），统计与数据科学学院，研究生课，授课教师，2025-2026，Fall

Invited Talks

TitleVenueDate
Conceptual Mathematics, Article VITsinghua FODS Seminar2022.11
Tight Partial Identification of Causal EffectsAI Time2024.06
 ICML 2024, Vienna2024.06
 Tsinghua Yao Class Graduate Forum2024.12
 Shanghai Jiao Tong University John Hopcroft Center2025.01
 Bear Conference (Goxiong Hui)2025.06
A Systematic ML Framework for CausalityPeking University Causality Seminar2024.11
 13th National Conference on Probability Statistics2024.11
 National University of Defense Technology2025.02
 Xi’an Jiaotong University2025.03
 Shanghai University of Finance and Economics2025.03
 Shandong University2025.03
 National University of Defense Technology2025.03
Partial Identification with Proxy of Latent ConfoundingsUAI 2024, Barcelona2024.06
Robust Causal Inference for Recommender SystemSIGIR 2023, Taiwan2023.07
Adjusting Auxiliary Variables Under Approximate Neighborhood InterferenceTsinghua University2025.04
 Bear Conference (Goxiong Hui)2025.06
Online Experiment Design under Interference Estimation-regret TradeoffShanghai Jiao Tong University John Hopcroft Center2025.01
 10th Statistical Forum2025.04
 NeuIPS 2025, San Diego2025.11
Design-Based Bandits Under Network Interference: Trade-Off Between Regret and Statistical InferenceNeuIPS 2025, San Diego2025.11
Unveiling Environmental Sensitivity of Individual Gains in Influence MaximizationNeuIPS 2025, San Diego2025.11
Active Treatment Effect Estimation via Limited SamplesICML 2025, Vancouver Convention Center2025.7
Topology-Informed Online Experimental Design under Network InterferenceReading group hosted by Professor Chengchun Shi2025.11
 POMSHK20262026.01