12 June 2026
Defence by Tharuka Chathurani Kasthuri Arachchige
Tharuka Kasthuri Arachchige defended her licentiate
On 8 June 2026, Tharuka Chathurani Kasthuri Arachchige successfully defended her licentiate thesis Heterogeneous Federated Learning: Fairness and Client Behaviour Exploration at the Department of Computer Science, Blekinge Institute of Technology (BTH).
Federated Learning enables collaborative model training without sharing raw data, but real‑world non‑IID client data creates unequal learning outcomes and fairness challenges. The thesis shows that global performance gains can hide disadvantages for certain clients, making it essential to understand heterogeneous client behaviour. It introduces an eccentricity‑based method to detect atypical client contributions and benefit patterns. Two fairness‑oriented frameworks, FeDABoost and DEFFT, are proposed to dynamically weight client updates and group clients by label distribution similarities. Experiments across benchmark datasets demonstrate that modeling heterogeneity improves fairness without sacrificing global accuracy.
The defence was examined by Prof. Veselka Boeva (main supervisor), with additional supervision from Dr. Shahrooz Abghari, Prof. Håkan Grahn, and Prof. Emiliano Casalicchio, all from BTH. The opponent for the seminar was Assoc. Prof. Fredrik D. Johansson from Chalmers University of Sweden.
The Department of Computer Science hosted a reception following the seminar to celebrate this milestone.
Tharuka’s contribution adds valuable insights to the HINTS research environment, strengthening ongoing work on human‑centered and trustworthy intelligent systems.
Find the thesis and abstract here:
