| Full text | |
| Author(s): |
de Souza, Lucas Airam C.
;
Camilo, Gustavo F.
;
Campista, Miguel Elias M.
;
Costa, Luis Henrique M. K.
Total Authors: 4
|
| Document type: | Journal article |
| Source: | 2024 IEEE LATIN-AMERICAN CONFERENCE ON COMMUNICATIONS, LATINCOM; v. N/A, p. 6-pg., 2024-01-01. |
| Abstract | |
Federated Learning (FL) depends on the data distribution among the clients. FL struggles in scenarios with heterogeneous data collected by different clients, which reduces the model's accuracy. Hence, we propose a client clustering system to mitigate the convergence obstacles of federated learning in non-Independent and Identically Distributed (IID) scenarios. Our proposal identifies the clusters using a testing neural network model sent to clients as a probe. After a few training epochs, each client returns to the server a vector containing the last-layer weights of the obtained model. Thus, we save communication and computation resources compared with other clustering proposals that apply iterative methods or use the entire neural network model. We also evaluate three clustering algorithms: K-Means, DBSCAN, and OPTICS. The DBSCAN algorithm demonstrates better results, correctly identifying the clients' clusters in IID and non-IID data distributions. Finally, the results show that our system has a better classification performance than FedAVG, increasing its accuracy by approximately 16% on non-IID scenarios. (AU) | |
| FAPESP's process: | 15/24494-8 - Communications and processing of big data in cloud and fog computing |
| Grantee: | Nelson Luis Saldanha da Fonseca |
| Support Opportunities: | Research Projects - Thematic Grants |
| FAPESP's process: | 18/23292-0 - ACCRUE-SFI project: advanced collaborative research infrastructure for secure future internet |
| Grantee: | Otto Carlos Muniz Bandeira Duarte |
| Support Opportunities: | Regular Research Grants |
| FAPESP's process: | 15/24485-9 - Future internet for smart cities |
| Grantee: | Fabio Kon |
| Support Opportunities: | Research Projects - Thematic Grants |
| FAPESP's process: | 14/50937-1 - INCT 2014: on the Internet of the Future |
| Grantee: | Fabio Kon |
| Support Opportunities: | Research Projects - Thematic Grants |