Photo: Andrei Smoliakov
Machine learning (ML) algorithms power intelligent systems across different domains, where centralized ML-based data processing may not be possible due to large data volumes, privacy considerations, and latency constraints. Such limitations may be addressed with distributed learning, where devices collaboratively train a common ML model without sharing raw data. However, differences in the computational and communication capabilities, hardware configurations, and data-generation patterns of devices may substantially impede the performance of collaborative learning. In his doctoral dissertation, MSc Nikita Zeulin proposes new distributed learning frameworks that explicitly account for the unequal capabilities of devices. His work explores several approaches to addressing this problem, including computational load balancing, resource-efficient learning methods, and hardware-agnostic ML model designs.
The doctoral dissertation titled Distributed Learning Methods for Heterogeneous Systems of Resource-Constrained Devices by MSc Nikita Zeulin will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University on 2 October 2026. The dissertation is in the field of machine learning.
The Opponent will be Professor Cenk Toker from Hacettepe University, Turkey. The Custos will be Professor Sergey Andreev from Tampere University.