Ilyas Fatkhullin

Logo

PhD Student at ETH AI Center and Computer Science Department of ETH Zurich, Switzerland

About

Research

Publications

Contact


About

I am a final year PhD student at ETH Zurich advised by Prof. Niao He. My research focuses on developing theoretically-grounded algorithms for machine learning and optimization, with particular emphasis on large-scale optimization, reinforcement learning, and theoretical foundations. Previously, I had an honor to work with Prof. Boris Polyak on control theory problems and with Prof. Peter Richtárik on federated learning, focusing on communication-efficient distributed training.

My research contributions have appeared in leading venues across machine learning and optimization:
  • • Conferences: NeurIPS, ICML, AISTATS.
  • • Journals: SIAM Journal on Control and Optimization, Journal of Machine Learning Research.
I am currently supported by the ETH AI Center Doctoral Fellowship. During my master's studies in Germany, I was awarded the DAAD scholarship.

Google Scholar Google Scholar


Research Interests

My research focuses on developing theoretical foundations and practical algorithms for machine learning and optimization. Here are my main research areas:

Large-Scale Optimization

Designing scalable optimization algorithms for efficient training of modern machine learning models.

  • • Stochastic optimization
  • • Adaptive and parameter-free algorithms
  • • Communication-efficient distributed training
View related publications →

Reinforcement Learning and Control

Developing principled approaches to decision-making under uncertainty with focus on efficiency, safety, and interpretability.

  • • Sample and computational efficiency
  • • Safe and robust RL algorithms
  • • Multi-agent learning and games
View related publications →

Theoretical Foundations

Uncovering fundamental properties of optimization problems and establishing rigorous theoretical guarantees for algorithmic performance.

  • • Non-convex optimization theory
  • • Hidden convexity and reformulations
  • • Heavy-tailed distributions
  • • Non-Euclidean geometries
View related publications →
My research approach combines theoretical rigor with practical efficiency. Each project aims to develop algorithms that are both theoretically sound and practically efficient, with a particular focus on understanding fundamental properties and establishing rigorous guarantees.