I am broadly interested in score-based generative AI, particularly diffusion models, and score matching. I have also been interested in developing provably optimal statistical and machine learning methods for signal and anomaly detection, particularly in high-dimensional settings of extremely low signal-to-noise ratio where prediction is impossible.

Papers

---

Diffusion models are optimal for hypothesis testing
S. Kotekal
In submission (Available upon request)

Sharp optimality of simple, plug-in estimation of the Fisher information of a smoothed density [openreview]
S. Kotekal
International Conference on Machine Learning (ICML), 2025, to appear

From optimal score matching to optimal sampling [arXiv]
Z. Dou, S. Kotekal, Z. Xu, and H. Zhou

Locally sharp goodness-of-fit testing in sup norm for high-dimensional counts [arXiv]
S. Kotekal, J. Chhor, and C. Gao
Bernoulli, to appear

Variance estimation in compound decision theory under boundedness [proc] [openreview]
S. Kotekal
Advances in Neural Information Processing Systems (NeurIPS), 2024

Optimal estimation of the null distribution in large-scale inference [arXiv] [doi]
S. Kotekal and C. Gao
IEEE Transactions on Information Theory, 2025

Sparsity meets correlation in Gaussian sequence model [arXiv] [doi]
S. Kotekal and C. Gao
Annals of Statistics, 2025

Optimal heteroskedasticity testing in nonparametric regression [arXiv] [doi]
S. Kotekal and S. Kundu
Annals of Statistics, 2025

Minimax signal detection in sparse additive models [arXiv] [doi]
S. Kotekal and C. Gao
IEEE Transactions on Information Theory, 2024

Minimax rates for sparse signal detection under correlation [arXiv] [doi]
S. Kotekal and C. Gao
Information and Inference, 2023

Statistical limits of sparse mixture detection [arXiv] [doi]
S. Kotekal
Electronic Journal of Statistics, 2022

Recurrent interactions can explain the variance in single trial responses [doi]
S. Kotekal and J. N. MacLean
PLOS Computational Biology, 2020