School of Electrical and Electronics Engineering, Pusan National University, Korea
We strive for fundamental understanding of data and intelligence.
Our goal is to design effective and efficient algorithm for intelligent systems.
Recent work focuses on optimizing federated learning systems with attention to personalization, robustness to data and system heterogeneity, and support for multi-domain datasets.
On-device AI · Federated Learning · Domain Adaptation · Distributed Systems · Convergence of Machine Learning and Communications
When heterogeneous devices collaborate in a federated learning setting, challenges arise in data distribution, model convergence, and communication efficiency. We address these challenges by proposing novel algorithms and frameworks that improve the performance and robustness of federated learning systems in heterogeneous environments.
One of the critical challenges in federated learning is the presence of multiple data domains across clients. We propose a personalized federated learning framework that adapts to the unique characteristics of each domain while leveraging shared knowledge across them. This approach enhances model performance and robustness in multi-domain scenarios and further provides personalized models tailored to individual clients.
To enable collaboration without exchanging local data, frequent communication between participants is essential. We investigate the interplay between federated learning systems and communication networks, focusing on the fundamental problem of jointly optimizing both. Our approach develops integrated strategies that account for the dynamics of federated learning and the characteristics of communication networks, enabling efficient training of high-quality models.
Jaeyoung Song is an assistant professor at School of Electrical and Electronics Engineering, Pusan National University, Korea. He obtained his B.S. degrees in Physics from KAIST in 2012, and the Ph.D. degree in Electrical Engineering from KAIST in 2019. He was a postdoctoral researcher at department of communication systems, EURECOM, France in 2020. He worked as a senior researcher at Agency for Defense Development, Korea from 2021 to 2022. Since 2022, he joined the faculty at Pusan National University. His current research interest includes multi-domain federated learning and on-device artificial intelligence.
M.S. Candidate
is interested in efficient personalized federated learning and client sampling for clustered federated learning.
M.S. Candidate
is interested in overcoming system and data heterogeneity in personalized federated learning systems.
M.S. Candidate
is interested in reducing global bias for improving performance of personalized federated learning.
M.S. 2025
studied resource allocation for wireless federated learning.
M.S. 2025
studied event-triggering decentralized federated learning.
Email : jsong@pusan.ac.kr
Address : Room #9508, The 9th Engineering Building, Pusan National University, Busandaehak-ro 63beon-gil 2, Geumjeong-gu, Busan, 46241