Compressed Federated Learning with Directional Trend Awareness
Abstract
Federated learning suffers from high communication overhead due to frequent transmission of large local model updates. To address this challenge, we propose a trend-aware projection-based compression method that adaptively selects a compressed update based on directional similarity with the previous update. Simulation results show that the proposed method achieves higher accuracy and lower training loss than other baseline schemes under the same communication cost.
Fields of Study & Subjects
Privacy-Preserving Technologies in Data
Stochastic Gradient Optimization Techniques
Distributed Sensor Networks and Detection Algorithms
Publication details
- Journal / source
- 한국통신학회논문지
- Volume
- 51
- Issue
- 1
- Pages
- 55–59
- Publication date
- 29 January 2026
- Language
- EN