Muthusamy, Siranjeevi (2024) Federated Deep Learning for Privacy Preserving Collaborative Data Sharing and Fault Recovery in Connected Vehicles. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (632kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (264kB) | Preview |
Abstract
In smart connected internet-of-vehicles (IoVs), locally generated sensory data contains the most important and knowledgeful vehicle-related information captured from onboard units like speedometers, dashboard cameras, and location trackers can offer deeper insights into its status. Through IoV with autonomous driving, this data can be used to optimize and update over-the-air (OTA) driving models from vectors on a cloud or supercomputer that can improve traffic handling dynamically. The current method of using centralized servers for processing vehicle data is limited in its ability to deliver real-time analysis which causes long-lasting vehicle downtimes because it does not consume the common localized information located throughout the IoV network. This research has been influenced to understand if federated learning (FL) can enable privacy-preserving, collaborative use of data from connected intelligent vehicles, while also being robust against vehicle faults. To realize this objective, we propose an FL framework that utilizes deep learning (DL) algorithms such as feed-forward neural networks (FNNs) and convolutional neural networks (CNN), to obtain vital information from vehicular data to collaboratively train models that manage vehicle operations, without the need to share raw vehicle data. The proposed methodology concatenates local model training and global aggregation of the models of every vehicle belonging to the IoV network to overcome security threats, and communication overhead from sharing huge volumes of raw data between vehicles in a high-precision driving scenario. This ensures improved throughput with reduced uplink communications by offering enhanced privacy on large-scale datasets. The proposed FL framework is tested for accuracy, and effectiveness through a case study experiment on the MNIST dataset to diagnose faults and recovery times in the vehicle operation which will not only reduce downtime of the vehicle but will also provide an exquisite driving experience backed with the extracted intelligence from decentralized FL computing models
Item Type: | Thesis (Masters) |
---|---|
Supervisors: | Name Email Samarawickrama, Yasantha UNSPECIFIED |
Uncontrolled Keywords: | Federated Learning; Deep Learning; Internet-Of-Vehicles; Collaborative Data Sharing; Fault Diagnosis |
Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing |
Divisions: | School of Computing > Master of Science in Cloud Computing |
Depositing User: | Ciara O'Brien |
Date Deposited: | 04 Jul 2025 08:41 |
Last Modified: | 04 Jul 2025 08:41 |
URI: | https://norma.ncirl.ie/id/eprint/8038 |
Actions (login required)
![]() |
View Item |