Jianqing Fan
Princeton University, NJ·Mathematics & Statistics
Huge datasets now drive decisions in medicine, business and artificial intelligence. Yet the statistical methods behind deep learning, decision-making algorithms and large-scale data analysis are still poorly understood.
- Fields
Mathematics & Statistics
Computer Science
Neuroscience
- Topics
statistical learning
machine learning
deep learning
- Grants
- 3 known grants, through Jul 2029
What they work on
- Use a low-dimensional simplex structure found in the eigenvectors of noisy data to develop methods for fitting hierarchical topic models and dynamic network models.
- Advance random matrix theory to derive sharp entry-by-entry error bounds and uncertainty estimates for leading eigenvectors when data noise is highly uneven.
- Build new text datasets and language models that predict influential citations and detect flaws in financial statements.
- Explain theoretically why deep learning works by studying over-parametrization, minimum-norm interpolation and low-rank matrix recovery, and how neural networks interact with heavy-tailed data.
- Develop semiparametric and nonparametric models for contextual dynamic pricing, and explain deep reinforcement learning's success through hierarchical composition models and new dimension-reduction methods for policy learning.
- Develop statistical methods for big data with Markovian dependence, missing or censored responses, highly correlated measurements and data distributed across machines.
- Build scalable clustering, marker-gene, Hodges-Lehmann and factor-adjusted methods for single-cell RNA sequencing data to identify blood vessel cell subgroups and their crosstalk with immune cells after spinal cord injury.
More about the research
These datasets are often noisy, very uneven and sparse, meaning most entries carry little information. The lab develops statistical theory and methods that can pull reliable signals out of such data. Its projects cover three areas: hidden structure in text and network data, the statistics behind learning and decision-making, and tools for analyzing single-cell gene data in spinal cord injury.
Why it matters
This research could make machine learning and big-data analysis more reliable and efficient. The benefits could reach fields from finance to e-commerce pricing, where language models may help spot flaws in financial statements. In spinal cord injury, new statistical tools for single-cell data could reveal how damaged blood vessels and immune cells interact. That could point to new treatment targets for patients facing paralysis. Public software releases would let other scientists apply the methods.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Collaborative Research: FRGMS: Exploring hidden components in complex data: Geometry, uncertain quantifications, and dynamics
NSF · Aug 2026 – Jul 2029
Interface of Statistical Learning and Optimal Decisions
NSF · Jul 2022 – Jun 2027
DMS/NIGMS 2: Collaborative Research: Developing Statistical Learning Methods for Revealing the Molecular Signatures of Microvascular Changes in Neural Injury
NSF · Jul 2021 – Dec 2026
Closest labs
Zheng Ke
Harvard University, MA·Mathematics & Statistics
Use a low-dimensional simplex structure found in the spectral domain of the data to develop new methods for fitting hierarchical topic models and dynamic network models.
Minge Xie
Rutgers University-New Brunswick, NJ·Mathematics & Statistics
Develop PCSI (Predictability-Computability-Stability Inference) for binary and multi-class classification, screening many algorithms by predictive performance and using bootstrap resampling to improve uncertainty estimates, robustness and accuracy.
Armeen Taeb
University of Washington, WA·Mathematics & Statistics
Develop a framework that organizes complex model classes (partitions, permutations, directed acyclic graphs, subspaces) as partially ordered sets, and use it to define generalized false positive error.
The school
Princeton University: 26 labs on PI Pages, $559M a year on research, 403 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.