Privacy-Preserved Federated Learning for Autonomous Driving

In recent years, the privacy issue in Vehicular Edge Computing (VEC) has gained a lot of concern. The privacy problem is even more severe in autonomous driving business than the other businesses in VEC such as ordinary navigation. Federated learning (FL), which is a privacy-preserved strategy propos...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on intelligent transportation systems 2022-07, Vol.23 (7), p.8423-8434
Main Authors: Li, Yijing, Tao, Xiaofeng, Zhang, Xuefei, Liu, Junjie, Xu, Jin
Format: Article
Language:English
Subjects:
Algorithms
Autonomous driving
Autonomous vehicles
Blockchains
Collaboration
Cryptography
Data privacy
Edge computing
Federated learning
homomorphic encryption
Privacy
Servers
Training
Vehicles
vehicular edge computing
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In recent years, the privacy issue in Vehicular Edge Computing (VEC) has gained a lot of concern. The privacy problem is even more severe in autonomous driving business than the other businesses in VEC such as ordinary navigation. Federated learning (FL), which is a privacy-preserved strategy proposed by Google, has become a hot trend to solve the privacy problem in many fields including VEC. Therefore, we introduce FL into autonomous driving to preserve vehicular privacy by keeping original data in a local vehicle and sharing the training model parameter only with the help of MEC server. Moreover, different from the common assumption of honest MEC server and honest vehicle in former studies, we take the malicious MEC servers and malicious vehicles into account. First, we consider honest-but-curious MEC server and malicious vehicles and propose a traceable identity-based privacy preserving scheme to protect the vehicular message privacy where improved Dijk-Gentry-Halevi-Vaikutanathan (DGHV) algorithm is proposed and a blockchain-based Reputation-based Incentive Autonomous Driving Mechanism (RIADM) is adopted. Further, when the case comes to the non-credibility of both parties where semi-honest MEC server and malicious vehicles are considered, we propose an anonymous identity-based privacy preserving scheme to protect the identity privacy of vehicles with Zero-Knowledge Proof (ZKP). Based on the simulation of virtual autonomous driving based on real-world road images, it is verified that our proposes scheme can reduce 73.7 % training loss of autonomous driving, increase the accuracy to around 5.55 % while keeps effective privacy of message and identity under the threat of dishonest MEC server and vehicles.
ISSN:1524-9050
1558-0016
DOI:10.1109/TITS.2021.3081560