BROWSE
Detail
Conference Paper
Federated learning for moderating server-side bias
Cited - time in 
Share 
Share
- Authors
- Jihyun Seo, Sunwook Kim
- Issue Date
- 2023-12
- Citation
- International Conference on Ubiquitous Information Technologies and Applications (CUTE) 2023, pp.1-6
- Language
- English
- Type
- Conference Paper
- Abstract
- Deep learning models have shown remarkable performance improvements in many areas of our society, and are actively used in fields such as healthcare, transportation, and robotics. These deep learning models require vast amounts of data and complex computations, leading to a shift from traditional centralized learning methods to distributed learning methods. In particular, a distributed learning paradigm called associative learning offers a number of benefits and has been the subject of active research, but it suffers from biases that can have a devastating impact on performance depending on the distribution of data or clients. In this paper, we review existing research and provide a new perspective on where and how bias can occur in federated learning. We take a different look at server-induced bias, as well as client-induced bias, which has been the main focus of previous work, and present a new architecture designed to address these issues. We also outline future work based on our findings.
- KSP Keywords
- Associative learning, Federated learning, Learning methods, Server side, Side bias, centralized learning, deep learning(DL), deep learning models, distributed learning