Research

Representative Works

The following works best describe who I am as a researcher. For a list of complete works, see my CV.

Note: * indicates co-first authors

Econometrics & Statistics

1. Distributionally Robust Instrumental Variables Estimation

with Yongchan Kwon. Under Review (2024).

We propose a distributionally robust version of the calssical IV estimation method for inferring causal effects, motivated by common concerns in practice such as instrument validity.

2. Handling Sparse Non-negative Data in Finance

with Agostino Capponi. Working Paper (2024).

Although Poisson pseudo maximum likelihood is robust for modeling data in finance and economics with non-negative dependent variables, there can be better estimators depending on the sparsity and heteroskedasticity. We propose a systematic framework that informs empirical researchers on such choices.

3. Semiparametric Estimation of Treatment Effects in Observational Studies with Heterogeneous Partial Interference

with Ruoxuan Xiong*, Jizhou Liu, and Guido Imbens. Major Revision at Journal of Business and Economic Statistics (2023).

We study augmented inverse propensity weighting (AIPW) estimators for causal effects under partial interference that accomodate heterogeneous interactions among units.

4. Computationally Efficient Estimation of Large Probit Models

with Patrick Ding, Guido Imbens, and Yinyu Ye. Major Revision at Journal of Econometrics (2024).

We leverage methods in approximate Bayesian inference and non-linear optimization to significantly accelerate existing EM algorithms for probit models, enabling the analysis of consumer choice data with a large number of alternatives.

Operations Research, Machine Learning & AI

5. Inferring Dynamic Networks from Marginals with Iterative Proportional Fitting

with Serina Chang*, Frederic Koehler*, Jure Leskovec, and Johan Ugander. 41st International Conference on Machine Learning (ICML) (2024).

We propose the ``biproportional Poisson’’ model, which provides statistical foundations for a widely used message passing algorithm to infer network traffic from marginal information.

6. On Sinkhorn’s Algorithm and Choice Modeling

with Alfred Galichon, Wenzhi Gao, and Johan Ugander. Accepted at Operations Research (2025).

We connect matrix balancing and choice modeling, two topics that are almost 100 years old, and use the connections to obtain new insights on the convergence of Sinkhorn’s algorithm.

7. Optimal Diagonal Preconditioning

with Wenzhi Gao*, Oliver Hinder, Yinyu Ye, and Zhengyuan Zhou. Operations Research (2023).

We provide SDP and interior point methods for finding the optimal diagonal preconditioners of a matrix.

8. A Unified Linear Speedup Analysis of Federated Averaging and Nesterov FedAvg

with Kaixiang Lin*, Zhaojian Li, Jiayu Zhou, and Zhengyuan Zhou. Journal of Artificial Intelligence Research 78: 1143-1200 (2023).

We establish the speedup of a class of popular federated learning algorithms in the number of participating local workers.