Ring-Based Decentralized Federated Learning with Cosine Similarity Grouping
Wu‐Chun Chung, Yan‐Hui Lin, Jau-Ai Luo · 2024
Federated Learning (FL) is an emerging methodology in machine learning but facing significant challenges. The most pressing issue is the problem of Non- Independently and Identically Distributed (Non-IID) data, which is caused by non-uniform data distribution among clients. To address the Non-IID challenge, this paper proposes a Decentralized Federated Learning (DFL) framework named RDFL-CS. RDFL-CS adopts a ring-based topology in DFL and incorporates cosine similarity to determine the ring position based on data distribution. Furthermore, RDFL-CS organizes participating clients into groups according to the value of cosine similarity to handle Non-IID data more effectively. The key contributions of this paper include the design of RDFL-CS and experimental results of its effectiveness in achieving high accuracy for the Non-IID data scenario.