Research

I am broadly interested in developing theory and statistically valid methods for problems motivated by applications, particularly in machine learning. My current research interest spans time series, learning theory, and stochastic optimization, with an emphasis on statistical inference under complex dependence. More recently, I have been exploring the theoretical foundations of generative models, including diffusion models.

Preprints

  • Fast localization of anomalous patches in spatial data under dependence.
    S. Bonnerjee, S. Karmakar, G. Michailidis
    Under review at Biometrika.
  • Fast segmentation of watermarked texts from large language models through epidemic change-points framework.
    🗣️
    S. Bonnerjee, S. Roy, S. Karmakar
    Major Revision at Journal of American Statistical Association, 2026
    🏆 IMS New Researcher Travel Award, 2026
    🏆 ASA-SLDS Paper Award, 2026
  • Identifying stability regions of SGD with constant learning rates.
    O. Goldreich, S. Bonnerjee, Q. Lei, J. Li, W. B. Wu
    Under review.
  • Testing synchronization of change-points for multiple time series.
    S. Bonnerjee, S. Karmakar, M. Cheng, W. B. Wu
    Major Revision at Biometrika, 2025
    🏆 IMS Hannan Graduate Student Travel Award, 2025

Journal Publications

  • How Private is Your Attention? Bridging Privacy with In-Context Learning.
    🗣️
    S. Bonnerjee, Z. W. Yeon, A. Asch, S. Nandy, P. Ghosal
    Accepted at Transactions on Machine Learning Research., 2026
  • Gaussian Approximation For Non-stationary Time Series with Optimal Rate and Explicit Construction.
    S. Bonnerjee, S. Karmakar, W. B. Wu
    Annals of Statistics, 52(5): 2293–2317, 2024
  • A Generalized Epidemiological Model with Dynamic and Asymptomatic Population.
    A. Ghatak, S. S. Patel, S. Bonnerjee, S. Roy
    Statistical Methods in Medical Research, 31(11): 2137–2163, 2022

Conference Publications

  • Stability beyond bounded differences: sharp generalization bounds under finite Lp moments.
    Q. Lei, S. Bonnerjee, Y. Han, W. B. Wu
    ICML 2026, Poster
  • Sharp asymptotic theory for Q-learning with LD2Z learning rate and its generalization.
    S. Bonnerjee, Z. Lou, W. B. Wu
    ICLR 2026, Poster
  • Sharp Gaussian approximations for Decentralized Federated Learning.
    S. Bonnerjee, S. Karmakar, W. B. Wu
    NeurIPS 2025, Spotlight (Top 3%)
  • Asymptotic theory of SGD with a general learning-rate.
    O. Goldreich, Z. Wei, S. Bonnerjee, J. Li, W. B. Wu
    NeurIPS 2025, Poster