Federated Learning for Heterogeneous Internet of Things Networks

Federated Learning for Heterogeneous Internet of Things Networks

With the growth of the Internet of Things (IoT), the amount of data generated in heterogeneous and resource-constrained environments has become a significant challenge for centralized or traditional Machine Learning (ML) models. These traditional models face challenges such as device heterogeneity, latency, communication costs (power consumption, bandwidth usage, and processing time), data security and privacy issues, and barriers imposed by the formation of data silos. To address these issues, academia and industry have developed and applied technologies such as Federated Learning (FL), which allows local training and sharing of models instead of raw data. Tiny Machine Learning (TinyML) aims to bring intelligence to the network's edge, enabling ML model inference directly on the device. Finally, Tiny Federated Learning (TinyFL) expands the TinyML concept by integrating FL, allowing devices to perform local model training in resource-constrained environments. However, in heterogeneous IoT networks, the different characteristics of devices require customized models to meet their specificities. Consequently, model heterogeneity emerges as a significant challenge for FL-enabled IoT solutions, demanding strategies considering device limitations, performance, and different data formats. In this context, this proposal aims to develop and validate an FL architecture for heterogeneous IoT networks to handle the diversity of IoT devices and explore strategies for aggregating heterogeneous ML models, optimizing computational efficiency and energy consumption. In addition, we will investigate model compression and optimization techniques to ensure scalability in resource-constrained devices. The applied methodology involves a systematic literature review to identify state-of-the-art gaps, develop an experimental environment, and validate the proposed model in simulated and real scenarios. To evaluate the proposal, we designed a scenario with resource-constrained devices. Preliminary results indicate the approach's feasibility, highlighting its potential to address critical challenges in heterogeneous IoT networks. This study seeks to provide both theoretical contributions and practical tools for implementing FL in IoT systems, focusing on improving efficiency and scalability.

Cássio Vinicius Serafim Prazeres

Heitor Soares Ramos Filho (UFMG - PQ-2)

http://lattes.cnpq.br/4978869867640619

Frederico Araujo Durão

http://lattes.cnpq.br/6271096128174325

Vinicius Fernandes Soares Mota (UFES)

http://lattes.cnpq.br/9305955394665920

Maycon Leone Maciel Peixoto

http://lattes.cnpq.br/5003713680310544

24 Mar, 2025

8:30 AM