Stanford Causal AI Lab

About Us

Our lab develops machine learning and artificial intelligence methodologies for learning causal effects from complex observational and experimental datasets. Our mission is to automate causal inference and make it accessible to decision-makers across various domains. Our methods are motivated by and have been applied in fields such as sustainability, healthcare, operations management, and digital experimentation. The lab's principal investigator has led the development of open-source software widely used in the industry, and the lab continues to support and develop open-source tools that lower barriers to entry in causal machine learning for data scientists.

Supported By

News

New paper posted on Arxiv on A Meta-learner for Heterogeneous Effects in Difference-in-Differences.
New paper on Detecting clinician implicit biases in diagnoses using proximal causal inference. Presented at the 2025 Pacific Symposium on Biocomputing. Associated python library available here.
New papers posted on Arxiv: i) Automatic Doubly Robust Forests, ii) Conditional Influence Functions, iii) Predicting Long Term Sequential Policy Value Using Softer Surrogates.
New paper on Switchback Price Experiments with Forward-Looking Demand
Five papers from our group will be presented at INFORMS 2024! Checkout the papers and schedule !
Three papers accepted at NeurIPS24! Congratulations to Keertana [paper], Jiyuan [paper] and Jikai [paper] for each getting one paper accepted. Congrats to Jikai for getting a spotlight presentation.
Vasilis received the NSF Career Award for his proposal on Automated and Robust Causal Inference for Data-Driven Decisions
New paper on Orthogonal Causal Calibration
New paper on Simultaneous Inference for Local Structural Parameters with Random Forests
New paper on Taking a Moment for Distributional Robustness
New paper on Dynamic Local Average Treatment Effects
New paper on Structure-agnostic Optimality of Doubly Robust Learning for Treatment Effect Estimation
Our new book on Applied Causal Inference Powered by ML and AI, with Victor Chernozhukov, Christian Hansen, Nathan Kallus and Martin Spindler, has been released online.

Applications

Causal AI for Sustainability

Our lab works on understanding causal factors of vulnerability of a region to natural hazards, with the aim of informing appropriate policy interventions.

Causal AI for Healthcare

Our lab works on topics such as detecting implicit biases in medical decisions, understanding the causal effects of latent treatment dimensions from observational patient treamtent trajectories and evaluation of policy changes in Kidney Exchange systems.

Causal AI for Operations

Our lab works on data-driven pricing, customer segmentation, return-on-investment and the impact of GenAI at work.

Causal AI for Digital Experiments

Our lab works on developing data analytic tools for experimentation in digital platforms, such as estimation of heterogeneous effects from large scale A/B tests and recommendation A/B tests.

Foundations

Heterogeneous Effects: Estimation, Inference and Policy Learning

How does the effect of a treatment or intervention vary as a function of characteristics of the treated unit and how should we personalize a treatment policy? Our lab works on ML techniques for the estimation of heterogeneous effects and for policy learning.

Debiased Machine Learning

ML methods typically introduce regularization to balance bias and variance of a learned model. Construction of confidence intervals requires small bias compared to variance. We have developed many techniques for automatic debiasing of many downstream causal quantities of interest.

Source Condition Double Robust Inference on Functionals of Inverse Problems

Automatic Debiased Machine Learning for Dynamic Treatment Effects

Automatic Debiased Machine Learning via Neural Nets for Generalized Linear Regression

Adversarial Estimation of Reisz Representers

Inference on Strongly Identified Functionals of Weakly Identified Functions, COLT23 and JRSSB (R&R)

Debiased Machine Learning without Sample-Splitting for Stable Estimators, NeurIPS22

RieszNet and ForestRiesz: Automatic Debiased Machine Learning with Neural Nets and Random Forests, ICML22

Knowledge Distillation as Semi-Parametric Inference, ICLR21

Orthogonal Machine Learning: Power and Limitations, ICML18

Non-Parametric Instrumental Variables

Instrumental variables (IV) is a powerful technique for parsing causal effects from observational data with unobserved confounding, leveraging natural experimentation. Classical IV methods either make strong parametric assumptions, or are fully non-parametric and suffer from the curse of dimensionality (only applicable with a small number of variables). We have developed generic ML techniques for instrumental variable methods, allowing for flexible causal modeling when having access to instruments.

Regularized DeepIV with Model Selection

Minimax Instrumental Variable Regression and L2 Convergence Guarantees without Identification or Closedness, COLT23

Minimax Estimation of Conditional Moment Models, NeurIPS20

Machine Learning Estimation of Heterogeneous Treatment Effects with Instruments, NeurIPS19

Adaptive Experiments

Adaptive experiments can lead to better decisions by appropriately changing the randomization probabilities based on past observations. Our lab works on novel adaptive experimentation algorithms that can handle high-dimensional settings. Adaptation can also lead to problems when it comes to retrospective analysis and inference. Our lab has introduced many techniques for the construction of confidence intervals for off-policy evaluation from adaptively collected data.

Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity, NeurIPS24

Post Reinforcement Learning Inference, OR (R&R)

Adaptive Instrument Design for Indirect Experiments, ICLR24

Dynamically Aggregating Diverse Information, Econometrica, 2021

Semiparametric Contextual Bandits, ICML18

Accurate Inference for Adaptive Linear Models, ICML18

Low-rank Bandit Methods for High-dimensional Dynamic Pricing, NeurIPS19

Dynamic Treatment Regimes

In many settings, units receive multiple treatments over time in an adaptive and autocorrelated manner. Causal inference in such settings is substantially harder as we need to take into account the intertemporal effects of the treatments. The area typically goes by the name of dynamic treatment regime. Our lab has developed ML based estimation techniques for the evaluation and inference of complex dynamic policies from offline observational data.

Dynamic Local Average Treatment Effects

Inference on Optimal Dynamic Policies via Softmax Approximation

Synthetic Blip Effects: Generalizing Synthetic Controls for the Dynamic Treatment Regime

Automatic Debiased Machine Learning for Dynamic Treatment Effects

Estimating the Long-Term Effects of Novel Treatments, NeurIPS21

Double/Debiased Machine Learning for Dynamic Treatment Effects via g-Estimation, NeurIPS21

Neural Causal Models

Causal inference can be framed as a neural generative model problem. This lends to general purpose algorithms for identification and partial identification via stochastic gradient descent training. Our lab develops methods and provable guarantees for such neural-causal approaches to causal identification and estimation.

Consistency of Neural Causal Partial Identification, NeurIPS24

Partial Identification of Treatment Effects with Implicit Generative Models, NeurIPS22

Causal Representation Learning

When dealing with complex data modalities such as text and images, the raw data do not correspond to the high level causal variables on which we want to perform our causal analysis. Causal representation learning aims to uncover such causal latent factors in an automated manner. Our lab has proven identifiability guarantees and associated algorithms for causal latent factor discovery.

Learning Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity, NeurIPS24

Towards efficient representation identification in supervised learning, CLear22

Sensitivity Analysis

Causal inference is necessarily based on domain assumptions. Sensitivity analysis offers a way to quantify the impact on the causal quantity of the violation of these assumptions. Our lab has developed general purpose sensitivity bound construction procedures with rigorous statistical guarantees, even when ML tecnniques are used in the estimation process.

Long Story Short: Omitted Variable Bias in Causal Machine Learning, ReStat (R&R)

DoWhy: Addressing Challenges in Expressing and Validating Causal Assumptions, ICML21 Workshop

Minimax Lower Bounds for Causal Estimation

The causal inference literature has produced many estimation algorithms that each have their pros and cons. A guiding principle on whether there is room for improvement is minimax optimality frameworks: what is the best achievable estimation quality, when we assume that our data stem from a family of distributions. Our lab has developed such minimax optimal lower bounds that match existing upper bounds of popular algorithms in a structure agnostic minimax optimal framework that is a good fit to estimation methods that use ML techniques as black-boxes.

Structure-agnostic Optimality of Doubly Robust Learning for Treatment Effect Estimation

Robust Statistics and Causal Inference

Real data always contain outliers or small "exceptional" sub-populations. Our lab develops techniques that can deal with such outliers in the context of causal estimation.

Taking a Moment for Distributional Robustness

Robust Generalized Method of Moments: A Finite Sample Viewpoint, NeurIPS22

Causal Inference and Incentives

Experimentation in real world systems, typically involves interacting with human subjects, each having their own incentives and trying to maximize their utility. Our lab develops experimentation and inference procedures that account for such incentives.

Switchback Price Experiments with Forward-Looking Demand

Incentive-Aware Synthetic Control: Accurate Counterfactual Estimation via Incentivized Exploration

Incentivizing Compliance with Algorithmic Instruments, ICML21

Optimal Data Acquisition for Statistical Estimation, EC18

Learning from Experts

Expert demonstration data can augment ML algorithms either by informing adaptive experiments or by informing generative models of the human preferences that their outputs need to align with. Our lab develops learning algorithms that can leverage such datasets and improve learning or quality of models.

Personalized Adaptation via In-Context Preference Learning, Arxiv24

Direct Preference Optimization With Unobserved Preference Heterogeneity, Arxiv24

Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity, NeurIPS24

Teaching

Applied Causal Inference Powered by ML and AI

Fundamentals of modern applied causal inference. Basic principles of causal inference and machine learning and how the two can be combined in practice to deliver causal insights and policy implications in real world datasets, allowing for high-dimensionality and flexible estimation. Lectures will provide foundations of these new methodologies and the course assignments will involve real world data (from social science, tech industry and healthcare applications) and synthetic data analysis based on these methodologies.

Foundations of Causal Machine Learning

Theoretical foundations of modern techniques at the intersection of causal inference and machine learning. Topics include: semi-parametric inference and semi-parametric efficiency, modern statistical learning theory, Neyman orthogonality and debiased machine learning, adaptive non-parametric estimation of conditional moment models, estimation and inference on heterogeneous treatment effects, causal inference and reinforcement learning, off-policy evaluation.

Applied Causal ML Book

Foundations of Causal ML Book

People

Vasilis Syrgkanis
Lab PI, Asst. Professor (MS&E)

Justin Whitehouse
Postdoctoral Researcher

Ravi Sojitra
PhD Student (MS&E)

Hui Lan
PhD Student (ICME)

Keertana Cheedambaram
PhD Student (MS&E)

Jiyuan Tan
PhD Student (MS&E)

Jikai Jin
PhD Student (ICME)

Shiangyi Lin
PhD Rotation Student (ICME)

Ayush Sawarni
PhD Rotation Student (MS&E)

Chenghan Xie
PhD Rotation Student (MS&E)

Calvin Xu
Undergrad Researcher (CS)

Karthik Seetharaman
Undergrad Researcher (Math)

Saanvi Chawla
Undegrad Researcher (CS)

Alumni

Yifan Wu
Visiting PhD Student (23-24)

Publications

2025

A Meta-learner for Heterogeneous Effects in Difference-in-Differences
Hui Lan, Haoge Chang, Eleanor Dillon, Vasilis Syrgkanis, Arxiv25
Detecting clinician implicit biases in diagnoses using proximal causal inference
Kara Liu, Russ Altman, Vasilis Syrgkanis, Pacific Symposium on Biocomputing 2025

2024

Predicting Long Term Sequential Policy Value Using Softer Surrogates
Hyunji Nam, Allen Nie, Ge Gao, Vasilis Syrgkanis, Emma Brunskill, Arxiv24
Conditional Influence Functions
Victor Chernozhukov, Whitney K. Newey, Vasilis Syrgkanis, Arxiv24
Automatic Doubly Robust Forests
Zhaomeng Chen, Junting Duan, Victor Chernozhukov, Vasilis Syrgkanis, Arxiv24
Switchback Price Experiments with Forward-Looking Demand
Yifan Wu, Ramesh Johari, Vasilis Syrgkanis, Gabriel Y. Weintraub, Arxiv24
Personalized Adaptation via In-Context Preference Learning
Allison Lau, Younwoo Choi, Vahid Balazadeh, Keertana Chidambaram, Vasilis Syrgkanis, Rahul G. Krishnan, Arxiv24
Orthogonal Causal Calibration
Justin Whitehouse, Christopher Jung, Vasilis Syrgkanis, Bryan Wilder, Zhiwei Steven Wu, Arxiv24
Dynamic Local Average Treatment Effects
Ravi Sojitra, Vasilis Syrgkanis, Arxiv24
Simultaneous Inference for Local Structural Parameters with Random Forests
David Ritzwoller, Vasilis Syrgkanis, Arxiv24
Structure-agnostic Optimality of Doubly Robust Learning for Treatment Effect Estimation
Jikai Jin, Vasilis Syrgkanis, Arxiv24
Regularized DeepIV with Model Selection
Zihao Li, Hui Lan, Vasilis Syrgkanis, Mengdi Wang, Masatoshi Uehara, Arxiv24
Taking a Moment for Distributional Robustness
Jabari Hastings, Christopher Jung, Charlotte Peale, Vasilis Syrgkanis, Arxiv24
Direct Preference Optimization With Unobserved Preference Heterogeneity
Keertana Chidambaram, Karthik Vinay Seetharaman, Vasilis Syrgkanis, Arxiv24
Learning Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity
Jikai Jin, Vasilis Syrgkanis, NeurIPS24 (Spotlight)
Consistency of Neural Causal Partial Identification
Jiyuan Tan, Jose Blanchet, Vasilis Syrgkanis, NeurIPS24
Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity, NeurIPS24
Vahid Balazadeh, Keertana Chidambaram, Viet Nguyen, Rahul G. Krishnan, Vasilis Syrgkanis, NeurIPS24
Causal Q-Aggregation for CATE Model Selection
Hui Lan, Vasilis Syrgkanis, AISTATS24
Adaptive Instrument Design for Indirect Experiments
Yash Chandak, Shiv Shankar, Vasilis Syrgkanis, Emma Brunskill, ICLR24
Empirical Analysis of Model Selection for Heterogenous Causal Effect Estimation
Divyat Mahajan, Ioannis Mitliagkas, Brady Neal, Vasilis Syrgkanis, ICLR24

2023

Incentive-Aware Synthetic Control: Accurate Counterfactual Estimation via Incentivized Exploration
Daniel Ngo, Keegan Harris, Anish Agarwal, Vasilis Syrgkanis, Zhiwei Steven Wu, Arxiv23
Inference on Optimal Dynamic Policies via Softmax Approximation
Qizhao Chen, Morgane Austern, Vasilis Syrgkanis, Arxiv23
Automatic Debiased Machine Learning for Covariate Shifts
Victor Chernozhukov, Michael Newey, Whitney K Newey, Rahul Singh, Vasilis Srygkanis, Arxiv23
Post Reinforcement Learning Inference
Ruohan Zhan, Vasilis Syrgkanis, Arxiv23, Operations Research (Revise & Resubmit)
Source Condition Double Robust Inference on Functionals of Inverse Problems
Andrew Bennett, Nathan Kallus, Xiaojie Mao, Whitney Newey, Vasilis Syrgkanis, Masatoshi Uehara, Arxiv23
Inference on Strongly Identified Functionals of Weakly Identified Functions
Andrew Bennett, Nathan Kallus, Xiaojie Mao, Whitney Newey, Vasilis Syrgkanis, Masatoshi Uehara, COLT23
Minimax Instrumental Variable Regression and L2 Convergence Guarantees without Identification or Closedness
Andrew Bennett, Nathan Kallus, Xiaojie Mao, Whitney Newey, Vasilis Syrgkanis, Masatoshi Uehara, COLT23

2022

Synthetic Blip Effects: Generalizing Synthetic Controls for the Dynamic Treatment Regime
Anish Agarwal, Vasilis Syrgkanis, Arxiv22
Finding Subgroups with Significant Treatment Effects, CLear22
Jann Spiess, Vasilis Syrgkanis, Victor Yaneng Wang, CLear22
Non-Parametric Inference Adaptive to Intrinsic Dimension
Khashayar Khosravi, Gregory Lewis, Vasilis Syrgkanis, CLear22
Towards efficient representation identification in supervised learning Kartik Ahuja, Divyat Mahajan, Ioannis Mitliagkas, Vasilis Syrgkanis, CLear22
Regularized Orthogonal Machine Learning for Nonlinear Semiparametric Models
Denis Nekipelov, Vira Semenova, Vasilis Syrgkanis, The Econometrics Journal, 2022
Automatic Debiased Machine Learning for Dynamic Treatment Effects
Victor Chernozhukov, Whitney Newey, Rahul Singh, Vasilis Syrgkanis, Arxiv22
Robust Generalized Method of Moments: A Finite Sample Viewpoint, NeurIPS22
Dhruv Rohatgi, Vasilis Syrgkanis, NeurIPS22
Debiased Machine Learning without Sample-Splitting for Stable Estimators
Qizhao Chen, Vasilis Syrgkanis, Morgane Austern, NeurIPS22
Partial Identification of Treatment Effects with Implicit Generative Models
Vahid Balazadeh Meresht, Vasilis Syrgkanis, Rahul G Krishnan, NeurIPS22 (Spotlight)
RieszNet and ForestRiesz: Automatic Debiased Machine Learning with Neural Nets and Random Forests
Victor Chernozhukov, Whitney K. Newey, Victor Quintas-Martinez, Vasilis Syrgkanis, ICML22 (Long Oral)

2021

Long Story Short: Omitted Variable Bias in Causal Machine Learning
Victor Chernozhukov, Carlos Cinelli, Whitney Newey, Amit Sharma, Vasilis Syrgkanis, Arxiv21, Review of Economics and Statistics (Revise & Resubmit)
Automatic Debiased Machine Learning via Riesz Regression
Victor Chernozhukov, Whitney K. Newey, Victor Quintas-Martinez, Vasilis Syrgkanis, Arxiv21
Adversarial Estimation of Reisz Representers
Victor Chernozhukov, Whitney Newey, Rahul Singh, Vasilis Syrgkanis, Arxiv21
Estimating the Long-Term Effects of Novel Treatments
Keith Battocchi, Eleanor Dillon, Maggie Hei, Greg Lewis, Miruna Oprescu, Vasilis Syrgkanis, NeurIPS21
Double/Debiased Machine Learning for Dynamic Treatment Effects via g-Estimation
Greg Lewis, Vasilis Syrgkanis, NeurIPS21
Asymptotics of the Bootstrap via Stability with Applications to Inference with Model Selection
Morgane Austern, Vasilis Syrgkanis, NeurIPS2021
DoWhy: Addressing Challenges in Expressing and Validating Causal Assumptions
Amit Sharma, Vasilis Syrgkanis, Cheng Zhang, Emre Kiciman, ICML21 Workshop on the Neglected Assumptions in Causal Inference
Dynamically Aggregating Diverse Information
Annie Liang, Xiaosheng Mu, Vasilis Syrgkanis, EC2021 and Econometrica, 2021
Incentivizing Compliance with Algorithmic Instruments, ICML21
Daniel Ngo, Logan Stapleton, Vasilis Syrgkanis, Zhiwei Steven Wu, ICML21
Knowledge Distillation as Semi-Parametric Inference
Tri Dao, Govinda Kamath, Vasilis Syrgkanis, Lester Mackey, ICLR21

2020

Estimation and Inference with Trees and Forests in High Dimensions
Vasilis Syrgkanis, Manolis Zampetakis, COLT20
Minimax Estimation of Conditional Moment Models
Nishanth Dikkala, Greg Lewis, Lester Mackey, Vasilis Syrgkanis, NeurIPS20

2019

Machine Learning Estimation of Heterogeneous Treatment Effects with Instruments
Vasilis Syrgkanis, Victor Lei, Miruna Oprescu, Maggie Hei, Keith Battocchi, Greg Lewis, NeurIPS19 (spotlight)
Semi-Parametric Efficient Policy Learning with Continuous Actions
Mert Demirer, Vasilis Syrgkanis, Greg Lewis, Victor Chernozhukov, NeurIPS19
Low-rank Bandit Methods for High-dimensional Dynamic Pricing
Jonas Mueller, Vasilis Syrgkanis, Matt Taddy, NeurIPS19
Orthogonal Statistical Learning
Dylan Foster, Vasilis Syrgkanis, COLT19 (Best Paper Award) and Annals of Statistics 2022
Orthogonal Random Forest for Causal Inference
Miruna Oprescu, Vasilis Syrgkanis, Zhiwei Steven Wu, ICML19

2018

Semiparametric Contextual Bandits
Akshay Krishnamurthy, Zhiwei Steven Wu, Vasilis Syrgkanis, ICML18
Accurate Inference for Adaptive Linear Models
Yash Deshpande, Lester Mackey, Vasilis Syrgkanis, Matt Taddy, ICML18
Optimal Data Acquisition for Statistical Estimation
Yiling Chen, Nicole Immorlica, Brendan Lucier, Vasilis Syrgkanis, Juba Ziani, EC18
Orthogonal Machine Learning: Power and Limitations
Lester Mackey, Vasilis Syrgkanis, Ilias Zadik, ICML18