산업공학과 연구세미나 개최 : Server-side Sequential Decision-Making Framework for Achieving Long-Term Performance Fairness in Federated Learning
2025.02.06- Date
- 2025-02-12 00:00:00
- Lecturer
- 한석주
- Venue
- Online(Zoom)
📅 세미나 정보
🔸일시 : 2025년 2월 12일(수) 오후 3시
🔸발표자 : 한석주 (Postdoctoral Fellow, Seoul National University)
🔸강연주제: Server-side Sequential Decision-Making Framework for Acㅎhieving Long-Term Performance Fairness in Federated Learning
📑Abstract : In traditional federated learning, a single global model cannot perform equally well for all clients. Therefore, the need to achieve the client-level long-term performance fairness in federated system has been emphasized, which is known to be realized by modifying the static aggregation scheme to an adaptive strategy. The adaptive aggregation can be composed by the central sever in response to the local signals (e.g., local loses) of the participating clients. Our work starts from revealing that existing fairness-aware aggregation strategies can be unified into a well-known online convex optimization framework. In other words, a central server’s sequential decision-making can salvage the performance bias across clients. To enhance the decision-making capability, we propose simple and intuitive improvements for suboptimal designs within existing methods, presenting 𝙰𝙰𝚐𝚐𝙵𝙵 framework for both cross-silo and cross-device settings. Our framework theoretically guarantees sublinear regret upper bounds in both settings, dependent on the number of communication rounds and the number of participating clients. Extensive experiments are demonstrated that the federated system equipped with 𝙰𝙰𝚐𝚐𝙵𝙵 achieves better degree of client-level fairness than existing methods in both practical settings.
⚡연사정보 : Seok-Ju has received his Ph.D. degree (Aug. 2024) in Industrial Engineering at Ulsan National Institute of Science and Technology (UNIST; Republic of Korea), advised by Prof. Dr. Gi-Soo Kim and Prof. Dr. Junghye Lee. He received his B.S. (Feb. 2019) and M.S. (Feb. 2021) degrees in Industrial Engineering also from UNIST. He is now a postdoctoral fellow at the Institute of Engineering Research of Seoul National University (SNU; Republic of Korea), and also has been conscripted into the Republic of Korea Army as a Technical Research Personnel until Aug. 2025. He will join Mathematics and Computer Science Division at Argonne National Laboratory (Illinois, USA) as a postdoctoral researcher from Sep. 2025. His research interests include Federated/Collaborative Machine Learning and Distributed/Online/Private Optimization. Please find his research statement in https://vaseline555.github.io for more details.