Assistant Research Professor
Cornell University
I am an Assistant Research Professor at Cornell University, affiliated with the Department of Economics, the Department of Statistics, and the Center for Data Science for Enterprise and Society. My research develops methods to quantify the information content in data and to use that information efficiently. I also develop methods for automating semiparametric inference and measuring model misspecification. My work draws on influence-function geometry and the theory of positive definite kernels, with applications to causal inference. I received a PhD in Economics and Statistics from MIT.
% Citation keys use first-author surname + four-digit year + short title keyword.
@misc{mukhin2026surrogates,
title = {{Surrogate-powered Causal Inference with Censored Outcomes}},
author = {Yaroslav Mukhin and Tereza Oprea and Arielle Anderer and Christina Lee Yu and Jelena Bradic},
year = {2026},
eprint = {2610.05486},
archivePrefix = {arXiv},
primaryClass = {math.ST},
url = {https://arxiv.org/abs/2610.05486}
}
@inproceedings{mukhin2025kernel,
title = {{Kernel von Mises Formula of the Influence Function}},
author = {Yaroslav Mukhin},
year = {2025},
booktitle = {Advances in Neural Information Processing Systems},
volume = {38},
doi = {10.52202/085713-0324},
url = {https://papers.nips.cc/paper_files/paper/2025/hash/0dde49ec491174a11272c5e1e6013f9f-Abstract-Conference.html}
}
@inproceedings{cho2024kernel,
title = {{Kernel Debiased Plug-in Estimation: Simultaneous, Automated Debiasing without Influence Functions for Many Target Parameters}},
author = {Brian M. Cho and Yaroslav Mukhin and Kyra Gan and Ivana Malenica},
year = {2024},
booktitle = {Proceedings of the 41st International Conference on Machine Learning},
series = {Proceedings of Machine Learning Research},
volume = {235},
pages = {8534--8555},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v235/cho24c.html}
}
@misc{immorlica2026learning,
title = {{Learning from a Mixture of Information Sources}},
author = {Nicole Immorlica and Brendan Lucier and Yaroslav Mukhin and Clayton Thomas and Ruqing Xu},
year = {2026},
eprint = {2609.33214},
archivePrefix = {arXiv},
primaryClass = {econ.TH},
note = {Extended abstract forthcoming at WINE 2026},
url = {https://arxiv.org/abs/2609.33214}
}
@misc{mukhin2021robustness,
title = {{On Robustness of Counterfactuals in Structural Models}},
author = {Yaroslav Mukhin},
year = {2021},
note = {NeurIPS workshop contribution; not a main-conference proceedings paper},
url = {https://sites.google.com/view/robustbayes-neurips21/accepted-papers}
}
@phdthesis{mukhin2019geometric,
title = {{Geometric Methods in Econometrics and Statistics}},
author = {Yaroslav Mukhin},
year = {2019},
school = {Massachusetts Institute of Technology},
type = {PhD thesis in Economics and Statistics},
url = {https://hdl.handle.net/1721.1/124058}
}
@misc{mukhin2018sensitivity,
title = {{Sensitivity of Regular Estimators}},
author = {Yaroslav Mukhin},
year = {2018},
eprint = {1805.08883},
archivePrefix = {arXiv},
primaryClass = {econ.EM},
url = {https://arxiv.org/abs/1805.08883}
}Preprint, .
We condition on a post-treatment event history to recover information lost to censoring, without changing the marginal causal estimand.
The header adapts the observed-data panel of my “Stopped trajectories” slide: censoring partially hides the event history, degrading efficiency and potentially introducing bias. Conditioning on intermediate events partly recovers information lost to censoring and makes censoring ignorability more plausible.
NeurIPS, .
A spectral representation and low-rank RKHS estimator of the influence function, computed without analytic derivations.
ICML, .
Targeted likelihood update along RKHS eignefunction directions yields debiased parameter estimators without requiring influence functions: i.e. TMLE without the IF.
WINE, · Extended abstract forthcoming.
Can data volume substitute for data provenance? We quantify the exchange rate.
NeurIPS workshop, .
Local bounds on counterfactuals via gradient descent.
Preprint, .
A geometric measure of local sensitivity linking perturbation in target and control parameters to influence functions and other functional gradients.
PhD thesis · MIT.
Differential and geometric approaches to statistical efficiency, sensitivity, and gradient flows in econometric models.
With Jelena Bradic and Christophe Gaillac.
With Jelena Bradic and Francesca Molinari.
With Jelena Bradic.
Pareviously, I taught statistics at the Department of Economics at Dartmouth College and at MIT IDSS.