Bio

Tijana Zrnic is an Assistant Professor at Stanford University, jointly appointed between Statistics, Management Science & Engineering, and, by courtesy, Computer Science. She works on foundational questions in machine learning, statistics, and data-driven decision-making. Example topics of interest include AI-assisted statistical inference and data collection, performative prediction, and studying selection bias.

Academic Appointments

Professional Education

  • PhD, University of California, Berkeley, Electrical Engineering and Computer Sciences (2023)

Contact

  • Academic
    tzrnic@stanford.edu
    University - Faculty Department: Statistics Position: Asst Professor
    University - Faculty Department: Management Science and Engineering Position: Assistant Professor

Additional Info

  • Mail Code: 4026

Links

2025-26 Courses

All Publications

  • A FLEXIBLE DEFENSE AGAINST THE WINNER'S CURSE ANNALS OF STATISTICS Zrnic, T., Fithian, W. 2025; 53 (6): 2516-2535

    View details for DOI 10.1214/25-AOS2553

    View details for Web of Science ID 001659559100009

  • Can Unconfident LLM Annotations Be Used for Confident Conclusions? Gligoric, K., Zrnic, T., Lee, C., Candes, E. J., Jurafsky, D. edited by Ritter, A., Chiruzzo, L., Wang, L. ASSOC COMPUTATIONAL LINGUISTICS-ACL. 2025: 3514-3533

    View details for Web of Science ID 001598852000179

  • Cross-prediction-powered inference. Proceedings of the National Academy of Sciences of the United States of America Zrnic, T., Candès, E. J. 2024; 121 (15): e2322083121

    Abstract

    While reliable data-driven decision-making hinges on high-quality labeled data, the acquisition of quality labels often involves laborious human annotations or slow and expensive scientific measurements. Machine learning is becoming an appealing alternative as sophisticated predictive techniques are being used to quickly and cheaply produce large amounts of predicted labels; e.g., predicted protein structures are used to supplement experimentally derived structures, predictions of socioeconomic indicators from satellite imagery are used to supplement accurate survey data, and so on. Since predictions are imperfect and potentially biased, this practice brings into question the validity of downstream inferences. We introduce cross-prediction: a method for valid inference powered by machine learning. With a small labeled dataset and a large unlabeled dataset, cross-prediction imputes the missing labels via machine learning and applies a form of debiasing to remedy the prediction inaccuracies. The resulting inferences achieve the desired error probability and are more powerful than those that only leverage the labeled data. Closely related is the recent proposal of prediction-powered inference [A. N. Angelopoulos, S. Bates, C. Fannjiang, M. I. Jordan, T. Zrnic, Science 382, 669-674 (2023)], which assumes that a good pretrained model is already available. We show that cross-prediction is consistently more powerful than an adaptation of prediction-powered inference in which a fraction of the labeled data is split off and used to train the model. Finally, we observe that cross-prediction gives more stable conclusions than its competitors; its CIs typically have significantly lower variability.

    View details for DOI 10.1073/pnas.2322083121

    View details for PubMedID 38568975