没有合适的资源?快使用搜索试试~ 我知道了~
互联和自动化车辆的联合学习:现有方法和挑战的调查-Federated Learning for Connected and Au
需积分: 5 0 下载量 114 浏览量
2024-04-29
11:01:28
上传
评论
收藏 3.01MB PDF 举报
温馨提示
试读
19页
互联和自动化车辆的联合学习:现有方法和挑战的调查-Federated Learning for Connected and Au
资源推荐
资源详情
资源评论
IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 9, NO. 1, JANUARY 2024 119
Federated Learning for Connected and Automated
Vehicles: A Survey of Existing Approaches
and Challenges
Vishnu Pandi Chellapandi , Member, IEEE, Liangqi Yuan , Graduate Student Member, IEEE,
Christopher G. Brinton
, Senior Member, IEEE,StanislawH.
˙
Zak , Life Member, IEEE,
and Ziran Wang
, Member, IEEE
Abstract—Machine learning (ML) is widely used for key tasks in
Connected and Automated Vehicles (CAV), including perception,
planning, and control. However, its reliance on vehicular data for
model training presents significant challenges related to in-vehicle
user privacy and communication overhead generated by massive
data volumes. Federated learning (FL) is a decentralized ML ap-
proach that enables multiple vehicles to collaboratively develop
models, broadening learning from various driving environments,
enhancing overall performance, and simultaneously securing local
vehicle data privacy and security. This survey paper presents a
review of the advancements made in the application of FL for CAV
(FL4CAV). First, centralized and decentralized frameworks of FL
are analyzed, highlighting their key characteristics and method-
ologies. Second, diverse data sources, models, and data security
techniques relevant to FL in CAVs are reviewed, emphasizing their
significance in ensuring privacy and confidentiality. Third, specific
applications of FL are explored, providing insight into the base
models and datasets employed for each application. Finally, existing
challenges for FL4CAV are listed and potential directions for future
investigation to further enhance the effectiveness and efficiency of
FL in the context of CAV are discussed.
Index Terms—Federated learning, connected and automated
vehicles, distributed computing, privacy protection, data security.
I. INTRODUCTION
C
ONNECTED and automated vehicles (CAV) are the key
to future intelligent transportation systems (ITS) that en-
compass both ground and air transportation [1] , [2], [3], [4], [5],
[6], [7], [8], [9]. With the advent of Big Data, the Internet of
Things (IoT), edge computing, and intelligent systems, CAVs
have the potential to improve the overall transportation system
by reducing traffic accidents, congestion, and pollution [10],
[11], [12], [13]. CAVs integrate both Vehicle-to-Vehicle (V2V)
and Vehicle-to-Infrastructure (V2I) communication capabilities,
facilitating an enhanced perception of the environment beyond
Manuscript received 3 October 2023; revised 31 October 2023; accepted 10
November 2023. Date of publication 14 November 2023; date of current version
23 February 2024. (Corresponding author: Ziran Wang.)
The authors are with the College of Engineering, Purdue University, West
Lafayette, IN 47907 USA (e-mail: cvp@purdue.edu; liangqiy@purdue.edu;
cgb@purdue.edu; zak@purdue.edu; ryanwang11@hotmail.com).
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TIV.2023.3332675.
Digital Object Identifier 10.1109/TIV.2023.3332675
the direct line of sight [14], [15], [16]. This involves interac-
tion with other vehicles, traffic s ignals, pedestrians, and other
elements of the transportation ecosystem. Furthermore, CAVs
are designed to assume control of driving tasks by the human
operator under certain conditions, using a variety of sensors
and sophisticated machine learning (ML) algorithms to achieve
autonomous operation.
Currently, CAVs are generating a tremendous amount of raw
data, between 20 and 40 TB per day, per vehicle . The various
sources of these data include engine components, electronic con-
trol units (ECU), perception sensors, and vehicle-to-everything
(V2X) communications. This large amount of data is sent to
other vehicles, roadside infrastructures, or the cloud, continu-
ously or periodically for monitoring, prognostics, diagnostics,
and connectivity features [17]. This flow of data has driven
the flourishing deployment and application of ML in CAVs,
including areas such as Advanced Driver-Assistance Systems
(ADAS) [18], automated driving [19],ITS[20], and sustainable
development [21].
A. Motivation
Due to the large amount of data required to train ML models,
concerns have been raised about data security in terms of the
legitimacy of data collection, data misuse, and privacy breaches.
Data collected by various sensors in CAVs, are also considered
private and are subject to stringent privacy protection regulations
in different regions. One such example is the General Data
Protection Regulation (GDPR) in the European Union [22],
which imposes strict requirements and guidelines on the han-
dling and processing of personal data to ensure i ndividuals’
privacy rights are protected. Even with the development of
advanced ML techniques and vehicle connectivity, it has not
been feasible to have a secure framework to collect data from
every vehicle and train an ML model. These limitations led to
the development of a new ML paradigm known as Federated
Learning (FL) [23], [24]. The term Federated Learning (FL) has
been coined by Google [25]. FL was initially used for mobile
keyboard prediction in Gboard [26] to allow multiple mobile
phones to cooperatively and securely train an ML model. FL has
been extensively applied in various fields such as industry [27],
[28], [29], energy [30], [31], healthcare [32], [33], and more.
2379-8858 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Zhengzhou University. Downloaded on April 27,2024 at 09:15:34 UTC from IEEE Xplore. Restrictions apply.
120 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 9, NO. 1, JANUARY 2024
Fig. 1. Roadmap of this survey paper.
In FL, edge devices/clients only send the gradients or the
learnable parameters to cloud servers rather than sending mas-
sive local datasets i n a centralized learning framework. Cloud
servers perform a secure aggregation of the received gradi-
ents/weights and update the global model parameters that are
transmitted back to clients/edge devices [34]. This procedure,
known as a communication round, continues iteratively until the
convergence criteria are met in the global model optimization.
The key advantage of FL is reducing the strain on the network
while also preserving the privacy of the local data. FL is a
potential candidate that can utilize the data available from each
CAV and develop a robust ML model.
Despite the benefits of V2X communications among CAVs,
the invasion of privacy, accuracy, effectiveness, and commu-
nication resources is an essential concern to be addressed. FL
frameworks have received attention for their natural ability to
preserve privacy by transmitting only model data between the
server and its clients without including local vehicle data. In
particular, the model data packets are smaller than the user
data, thus saving the consumption of communication resources.
Similarly, FL frameworks distribute training tasks to each client,
and the server does not perform training but only aggregates,
which can reduce the computational demand on the server and
improves training efficiency. Recently, there have also been
efforts to train a decentralized FL that allows multiple vehi-
cles to collaboratively train a model without needing a central
server [35], [36]. In our first survey of FL for CAV (FL4CAV)
presented in [37], we emphasized applications and explored
foundational challenges in the subject. Building upon that con-
ference version, this extended journal paper further delves into
the underlying methodologies, provides a more comprehensive
review of recent developments, and introduces novel insights and
evaluations, thereby presenting a more exhaustive and nuanced
understanding of the field.
B. Paper Organization
In this paper, we provide a survey of FL4CAV, including
deployment of various FL frameworks on CAVs, data modal-
ities and security, diverse applications, and key challenges. The
organization of this survey is shown in Fig. 1. The following
topics are covered in this survey:
r
A systematic review of FL algorithms is conducted, specif-
ically focusing on their deployment in CAVs. Additionally,
we examine the integration of ML models within the FL
framework for CAV applications.
r
Data modalities and data security considerations in CAVs
are summarized, highlighting the diverse range of multi-
modal data generated by various sensors.
r
Critical applications of FL4CAV are explored, such as
driver monitoring, steering wheel angle prediction, vehi-
cle trajectory prediction, object detection, motion control
application, traffic flow prediction, and V2X communica-
tions.
r
Current challenges and future research directions of
FL4CAV are highlighted, such as performance, safety,
fairness, applicability, and scalability. A comparison of our
survey with other related surveys can be found in Table I.
The remainder of this survey is organized as follows. In
Section II, we describe the two main FL frameworks along
with their algorithms. In Section III, we discuss various data
modalities, ML methods used in FL4CAV applications, and FL
data security in CAVs. Section IV reviews various applications
of FL in CAVs. The multi-modal data, algorithms, and datasets
used in the relevant literature are also summarized. Challenges
and potential research areas are discussed in Section V.In
Section VI, we present conclusions of this survey.
II. F
EDERATED LEARNING METHODS
In this section, we describe the FL frameworks in terms of two
categories: centralized FL and decentralized FL. An illustration
of the categories is shown in Fig. 2. In addition, we provide an
overview of the ML techniques that are commonly used as base
models on local devices during the FL process. The steps of this
process can be described as:
1) Global Model Distribution: The edge server disseminates
the global model parameters to K vehicles.
2) Model Update Using Local Data: Each vehicle indepen-
dently trains the ML model using its own local data.
Authorized licensed use limited to: Zhengzhou University. Downloaded on April 27,2024 at 09:15:34 UTC from IEEE Xplore. Restrictions apply.
CHELLAPANDI et al.: FEDERATED LEARNING FOR CAVS: A SURVEY OF EXISTING APPROACHES AND CHALLENGES 121
TABLE I
C
OMPARISON OF RELATED SURVEYS OF FEDERATED LEARNING FOR CONNECTED AND AUTOMATED VEHICLES
TABLE II
C
OMPARISON OF MACHINE LEARNING APPROACHES IN CONNECTED AND AUTOMATED VEHICLES
Fig. 2. Illustration o f (a) centralized and (b) decentralized federated learning for connected and automated vehicles.
Authorized licensed use limited to: Zhengzhou University. Downloaded on April 27,2024 at 09:15:34 UTC from IEEE Xplore. Restrictions apply.
剩余18页未读,继续阅读
资源评论
yuan_0012
- 粉丝: 1
- 资源: 6
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功