®®®® SIIA Público

Título del libro: International Conference On Electrical, Computer, Communications And Mechatronics Engineering, Iceccme 2023
Título del capítulo: Adapting Detection in Blockchain-enabled Federated Learning for IoT Networks

Autores UNAM:
DAVID LOPEZ FLORES;
Autores externos:

Idioma:

Año de publicación:
2023
Palabras clave:

Bandwidth; Blockchain; Energy utilization; Internet of things; Learning systems; Pareto principle; Block-chain; Decentralized approach; Federated learning; Machine learning models; Malicious nodes; Multi-objectives optimization; Network resource; Pareto-optimal; Performance; Poisoning attacks; Multiobjective optimization


Resumen:

Blockchain-enabled Federated Learning (FL) is a decentralized approach to coordinate nodes during the training of a machine learning model in order to enhance privacy and save network resources (e.g., bandwidth). However, malicious nodes may try to sabotage the training by carrying out poisoning attacks to hinder the performance of the model. An effective defense to have on the blockchain is a mechanism to monitor the behaviour of the nodes, detect the malicious nodes, and remove them from the training. This work proposes a multiobjective optimization algorithm that adapts local training and model sharing on the blockchain to the wireless and mobile environment of the nodes. The results show an improvement in terms of time, bandwidth, and energy consumption when the optimization of the selection of the proportion of miners responsible for the monitoring and consensus was adapted to take into account multiple factors related to the communication network. © 2023 IEEE.


Entidades citadas de la UNAM: