Ph.D. in Statistics

Ph.D. in Statistics

Postdoc researcher at University of Pennsylvania

Research Papers

CARE: Large Precision Matrix Estimation for Compositional Data

JASA, 2024

This paper introduces a new method called CARE (composition adaptive regularized estimation) for estimating sparse precision matrices in high-dimensional compositional data, providing theoretical guarantees and demonstrating its effectiveness in inferring microbial ecological networks.

Zhang, S., Wang, H.,& Lin, W. (2024). "CARE: Large Precision Matrix Estimation for Compositional Data." Journal of the American Statistical Association, April, 1–13. doi:10.1080/01621459.2024.2335586.

Nonasymptotic theory for two-layer neural networks: Beyond the bias–variance trade-off

Manuscript, 2023

This paper gives a unified statistical guarantee for both underparametrized and overparametrized two-layer ReLU networks, and further reproduces the double descent phenonmenon. Full paper available for download.

Wang, H., & Lin, W. (2023). "Nonasymptotic theory for two-layer neural networks: Beyond the bias–variance trade-off." Manuscript.

Heterogeneous federated learning on arbitrary graphs

Manuscript, 2023

This paper introduces FedADMM, a new federated learning approach for parameter estimation considering heterogeneity in distribution, communication, and accessibility among an exceedingly large number of devices. Full paper available for download.

Wang, H., Zhao, X., & Lin, W. (2023). "Heterogeneous federated learning on arbitrary graphs." Manuscript.

The aggregation–heterogeneity trade-off in federated learning

COLT, 2023

This paper explores the learning limit for heterogeneous federated learning. Full paper available for download.

Zhao, X., Wang, H., & Lin, W. (2023). "The aggregation–heterogeneity trade-off in federated learning." The 36th Annual Conference on Learning Theory. To appear.

Difference-in-Differences meets tree-based methods: Heterogeneous treatment effects estimation with unmeasured confounding

ICML, 2023

This paper combines tree-based methods with Difference-in-Differences to estimate heterogeneous treatment effects in the presence of unmeasured confounding. Full paper available for download.

Tang, C., Wang, H., Li, X., Cui, Q., Zhang, Y. L., Li, L., & Zhou, J. (2022). "Difference-in-Differences meets tree-based methods: Heterogeneous treatment effects estimation with unmeasured confounding." Proceedings of the Fortieth International Conference on Machine Learning, To appear.

Debiased causal tree: heterogeneous treatment effects estimation with unmeasured confounding

NeurIPS, 2022

This paper proposes a new splitting rule for tree-based methods to estimate heterogeneous treatment effects in the presence of unmeasured confounding. Full paper available for download.

Tang, C., Wang, H., Li, X., Cui, Q., Zhang, Y. L., Zhu, F., Zhou, J., & Jiang, L. (2022). "Debiased causal tree: heterogeneous treatment effects estimation with unmeasured confounding." Advances in Neural Information Processing Systems, 35, 5628-5640.