News

September 2026

• New preprint! Quantile-based loss filtering for robust SGD: a paper on loss-based filtering for finite-sum optimization with a subset of corrupted component functions whose gradients may be highly unreliable — theory for the convex case + experiments on polynomial regression, regularized logistic regression, and regularized hinge loss. Joint work with Jamie Haddock and Anna Ma.
• New preprint! Inference for accelerated sketched Newton methods: We study a sketch-and-project solver with momentum for approximating the Newton direction in an online sketched Newton method: asymptotic normality, faster convergence, and uncertainty quantification via random scaling. Joint work with Xinchen Du, Michał Dereziński, and Sen Na.
• Congratulations, Shambhavi! Shambhavi Suryanarayanan has completed her Ph.D. and will join Oak Ridge National Laboratory as a postdoctoral researcher.
• I have joined the University of Oxford as an Associate Professor in the Numerical Analysis Group at the Mathematical Institute and New College as a Senior Research Fellow!

---

Research

My research focuses on numerical linear algebra and data science. Motivated by applications in machine learning and scientific computing, I develop randomized methods for solving large linear systems and recovering low-rank structure in matrices and tensors. I combine probabilistic analysis and numerical methods to understand how algorithmic choices affect convergence and to establish rigorous error bounds. This also leads me to study how optimization dynamics influence generalization in learning problems. One recurring question is when matrix structure and random sampling can make computation faster or less memory-intensive.

I received my PhD in Mathematics from the University of Michigan in 2018, advised by Roman Vershynin. I then held postdoctoral positions at UCLA (2018–2020) and Lawrence Berkeley National Laboratory (2021). Before joining Oxford in 2026, I was an Assistant Professor in Operations Research and Financial Engineering at Princeton University (2021–2026), affiliated with the Program in Applied and Computational Mathematics and the Center for Statistics and Machine Learning.

My work has been previously supported by NSF DMS-2309685 and NSF DMS-2108479.

I also serve as an Associate Editor for Information and Inference: A Journal of the IMA.

---

Advising

Current Ph.D. students:

• Sofiia Shvaiko; 4th year

Former Ph.D. students:

• Shambhavi Suryanarayanan, Ph.D. 2026

  Next position: Oak Ridge National Laboratory, Postdoctoral Researcher.

• Jackie Lok, Ph.D. 2026

  Next position: National University of Singapore, Peng Tsu Ann Assistant Professor.

• Abraar Chaudhry, Ph.D. 2024, co-advised with A. A. Ahmadi

  Next position: Georgia Tech, Postdoctoral Fellow.

---

Teaching experience

Princeton University (2021–2026)

Probability and Stochastic Processes, Convex and Conic Optimization, Networks.

UCLA (2018–2020)

Probability, Statistics, Stochastic Processes, and Optimization.

University of Michigan (2013–2017)

Calculus and Differential Equations.

Earlier teaching

In Moscow: calculus at math school 57 and algebra at Kolmogorov math and physics high school.

---

Random likes

My first name is the Russian version of Elizabeth. I like all versions of my name: please call me Liza, Lisa, Eliza, or Elizabeth, whatever you like the best. In case you were curious, Elizaveta is pronounced approximately as "Ye (like in yellow) - lee - zuh - VYE - tuh". Standard pronunciation of Liza is "LEE - zuh".

My non-mathy interests include things that I find beautiful or challenging, such as, good stories, arts, oceans, mountains and cities, also finding and making perfect coffee. I also immensely enjoy long conversations with my six-year-old son, currently fascinated by the Jedi, whose curious and very sharp logic is so fun to follow.

“Poirot,” I said. “I have been thinking.”
“An admirable exercise my friend. Continue it.”
(Agatha Christie, Peril at End House)