Ilya Tolstikhin

About

I am a Senior Research Scientist at Google DeepMind in Zurich, which I joined in 2018, initially as part of Google Brain. From 2014 to 2018, I was a postdoctoral researcher at the Max Planck Institute for Intelligent Systems in Tübingen, Germany, where I led the statistical learning theory team.

I received my PhD in 2014 from the Dorodnicyn Computing Center of the Russian Academy of Sciences and a Diploma (equivalent to an MSc) in 2010 from Lomonosov Moscow State University.

Research

My recent work, part of the AlphaProof project, focuses on improving the reasoning of large language models (LLMs) and their application to formal mathematics. To support this research, my colleagues and I developed the OneTwo Python library, designed to simplify interactions with LLMs.

Previously, I've worked on pre-training large-scale computer vision architectures [MLP-Mixer] and unsupervised generative modeling for natural images [Wasserstein Auto-Encoders].

In my academic career, I focused on statistical learning theory, particularly data-dependent generalization error and excess risk bounds and related mathematical tools from probability and combinatorics. I remain passionate about mathematics and enjoy learning about it in my free time.