Publicação

Federated learning framework to decentralize mobility forecasting in smart cities scenarios

Ver documento

Detalhes bibliográficos
Resumo:The new Federated Learning (FL) paradigm has several performance advantages over centralized models. It has lower latency and communication overhead when doing most of the processing on the edge devices, it improves the privacy as data does not travel over the network, it facilitates the handling of heterogeneous data sources and expands scalability. However, the development of FL-based solutions is done through tools usually aimed for specialists as it always requires some programming. To cover this gap, this dissertation proposes a lightweight container-based framework that does not require programming knowledge or experience with machine learning models from its users. This framework, denoted as FedFramework, offers a range of machine learning (ML) algorithms that support the build of prediction engines for edge devices as well as making key algorithms/models available for aggregation and model refinement on the central server. We demonstrate the efficiency of the proposed framework in estimating vehicle mobility in and out of the city, using real data collected by the Aveiro Tech City Living Lab communications infrastructure of the Aveiro city. Moreover, a testbed that integrates the components of the city infrastructure was implemented, where edge devices (Jetsons Nano and Jetson Xavier) are connected to a cloud server. The FedFramework was deployed in this testbed, where its portability, its scalability in devices with few resources, its performance, the impact on the communication between the edge and the server, and the consumption of resources were evaluated.
Autores principais:Valente, Renato Lima
Assunto:Federated learning Distributed forecasting Smart city FL on edge devices
Ano:2023
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade de Aveiro
Idioma:inglês
Origem:RIA - Repositório Institucional da Universidade de Aveiro
Descrição
Resumo:The new Federated Learning (FL) paradigm has several performance advantages over centralized models. It has lower latency and communication overhead when doing most of the processing on the edge devices, it improves the privacy as data does not travel over the network, it facilitates the handling of heterogeneous data sources and expands scalability. However, the development of FL-based solutions is done through tools usually aimed for specialists as it always requires some programming. To cover this gap, this dissertation proposes a lightweight container-based framework that does not require programming knowledge or experience with machine learning models from its users. This framework, denoted as FedFramework, offers a range of machine learning (ML) algorithms that support the build of prediction engines for edge devices as well as making key algorithms/models available for aggregation and model refinement on the central server. We demonstrate the efficiency of the proposed framework in estimating vehicle mobility in and out of the city, using real data collected by the Aveiro Tech City Living Lab communications infrastructure of the Aveiro city. Moreover, a testbed that integrates the components of the city infrastructure was implemented, where edge devices (Jetsons Nano and Jetson Xavier) are connected to a cloud server. The FedFramework was deployed in this testbed, where its portability, its scalability in devices with few resources, its performance, the impact on the communication between the edge and the server, and the consumption of resources were evaluated.