Marko Medvedev

PhD student in the Department of Mathematics The University of Chicago Email: [last name] at uchicago dot edu

About Me

I am a fourth-year PhD student at The University of Chicago, advised by Professor Nathan Srebro and Professor Alexander Razborov. I am interested in the theory of machine learning, including its computational and statistical aspects, with a broad interest in the theory of deep learning. My latest work is on understanding weak-to-strong generalization and positive distribution shifts. Before joining UChicago, I obtained my Bachelor of Arts in Mathematics from Princeton University.

Publications

• Positive Distribution Shift as a Framework for Understanding Tractable Learning. Marko Medvedev, Idan Attias, Elisabetta Cornacchia, Theodor Misiakiewicz, Gal Vardi, Nathan Srebro. Submitted.

• Shift is Good: Mismatched Data Mixing Improves Test Performance. Marko Medvedev, Kaifeng Lyu, Zhiyuan Li, Nathan Srebro. Submitted.

• Weak-to-Strong Generalization Even in Radnom Feature Networks, Provably. Marko Medvedev, Kaifeng Lyu, Dingli Yu, Sanjeev Arora, Zhiyuan Li, Nathan Srebro. ICML 2025.

• Overfitting Behaviour of Gaussian Kernel Ridgeless Regression: Varying Bandwidth or Dimensionality. Marko Medvedev, Gal Vardi, Nathan Srebro. NeurIPS 2024.

• Refinement of Chebotarev's density theorem in SL_2(\Z). Marko Medvedev. Senior Thesis at Princeton University.

Talks

• Weak-to-Strong Generalization Even in Random Feature Networks, Provably. Deep Learning Theory Workshop at The Simons Institute for the Theory of Computation, February 2025.