Research focus
I study optimization for deep learning in the OMLS group at the University of Basel, supervised by Prof. Aurelien Lucchi.
- Training dynamics: understanding how neural networks learn, including the edge of stability.
- Stochastic optimization: investigating the behavior of optimization methods under noisy updates.
- Distributed optimization: learning efficiently when computation and data are spread across devices.
Selected paper
Cohort Squeeze
Beyond a Single Communication Round per Cohort in Cross-Device Federated Learning
Kai Yi, Timur Kharisov, Igor Sokolov, and Peter Richtárik · 2024
Can we get more out of each cohort of clients in federated learning? We investigate reusing a cohort across multiple communication rounds to reduce communication costs.
Read on arXiv · FL@FM, NeurIPS 2024 presentation
Talk archive
What has Edge of Stability to do with the Convergence Theory?
Timur Kharisov, Foivos Alimisis, and Aurelien Lucchi · August 2026