PeopleofSuperintelligence

Researcher

Pragya Sur

Pragya Sur is a statistician whose work explains how high-dimensional data changes the behavior of machine-learning methods. She completed a PhD in Statistics at Stanford in 2019, won a dissertation award for high-dimensional logistic-regression theory, and coauthored research on the method with Emmanuel Candès. Her research also addresses high-dimensional boosting and model collapse when training data includes synthetic outputs. She led the Institute of Mathematical Statistics New Researchers Group from 2022 to 2024. As of October 2026, she is an Assistant Professor of Statistics at Harvard, where she studies machine-learning theory and mathematical foundations of AI.

Works
33
Citations
587
h-index
8

Publication numbers from OpenAlex, as of 27 October 2026.

Timeline

  1. 2019

    A Stanford statistics doctorate

    In 2019, Sur completes a PhD in Statistics at Stanford and receives the Theodore W. Anderson Theory of Statistics Dissertation Award. She receives it for original large-sample maximum-likelihood results for logistic regression with many covariates.

  2. 2019

    High-dimensional logistic regression

    In 2019, Sur and Emmanuel Candès publish a theory of high-dimensional logistic regression. It shows that maximum-likelihood estimates can be biased and more variable than classical theory predicts when feature and sample counts grow in a fixed ratio, and characterizes the distribution of likelihood-ratio tests.

  3. 2020

    A theory of boosting

    In 2020, Sur coauthors a high-dimensional asymptotic theory of boosting on separable data with Tengyuan Liang. Their analysis studies test error and optimization behavior for boosting in overparameterized settings.

  4. 2022–2024

    IMS New Researchers Group president

    She serves as president of the Institute of Mathematical Statistics New Researchers Group from 2022 to 2024. The group supports collaboration and career development for young statisticians and probabilists.

  5. 2025

    Preventing model collapse

    In 2025, she coauthors a preprint on preventing model collapse with Anvit Garg and Sohom Bhattacharya. They derive optimal mixing ratios for real and synthetic labels in overparameterized linear regression; for minimum-norm interpolation, the optimal real-data share converges to the reciprocal of the golden ratio.

  6. As of October 2026

    Assistant Professor at Harvard

    As of October 2026, she is an Assistant Professor of Statistics at Harvard, following a postdoctoral fellowship at Harvard's Center for Research on Computation and Society. Her research spans high-dimensional statistics, machine-learning theory, and mathematical foundations of AI. In Spring 2026, she teaches Harvard's STAT 195, Statistical Machine Learning, a course on theory and methods in statistical machine learning.

Career

  1. NowHarvard UniversityAssistant Professor of Statistics
  2. 2022–2024Institute of Mathematical StatisticsPresident, New Researchers Group
  3. Harvard UniversityPostdoctoral Fellow, Center for Research on Computation and Society

Education

  1. 2019Stanford UniversityPhD in Statistics

From the biography.

Checked career sources: simons.berkeley.edu.

Sources and credits

Photo: We have no published photo of Pragya Sur yet. Send us a free-licensed photo.

  1. Pragya Sur simons.berkeley.edu. Checked 27 October 2026.
  2. 2019 Statistics Department Dissertation Awards statistics.stanford.edu. Checked 27 October 2026.
  3. A modern maximum-likelihood theory for high-dimensional logistic regression pmc.ncbi.nlm.nih.gov. Checked 27 October 2026.
  4. A precise high-dimensional asymptotic theory for boosting and minimum-ℓ1-norm interpolated classifiers arxiv.org. Checked 27 October 2026.
  5. IMS-NRG - About Us imsnrg.com. Checked 27 October 2026.
  6. Preventing Model Collapse Under Overparametrization: Optimal Mixing Ratios for Interpolation Learning and Ridge Regression arxiv.org. Checked 27 October 2026.
  7. Harvard Academic Positions | Postdoctoral Fellow in Statistics academicpositions.harvard.edu. Checked 27 October 2026.
  8. STAT 195: Statistical Machine Learning qrd.college.harvard.edu. Checked 27 October 2026.