IEEE INTERNET OF THINGS JOURNAL 1

Edge Computing for Internet of Everything:

A Survey
Xiangjie Kong, Senior Member, IEEE, Yuhan Wu, Hui Wang, Feng Xia, Senior Member, IEEE

Abstract—In this era of the Internet of Everything (IoE), billion units by 2025, a significant increase from the 13.8
edge computing has emerged as the critical enabling technology billion units expected in 2021. This indicates that sensors,
to solve a series of issues caused by an increasing amount of actuators, and other intelligent devices will generate massive
interconnected devices and large-scale data transmission. How-
ever, the deficiencies of edge computing paradigm are gradually data at a fast speed, and these massive data need to be further
being magnified in the context of IoE, especially in terms of processed. However, the core network bandwidth is becoming
service migration, security and privacy preservation, deployment the bottleneck when moving all computing tasks to the cloud
issues of edge node. These issues can not be well addressed servers in cloud computing. Moreover, cloud computing is
by conventional approaches. Thanks to the rapid development hard to meet the demands of low cost, high performance,
of upcoming technologies, such as artificial intelligence (AI),
blockchain, and microservices, novel and more effective solutions and ultra-low latency of some time-sensitive applications (e.g.,
have emerged and been applied to solve existing challenges. interactive cloud applications and cooperative autonomous
In addition, edge computing can be deeply integrated with driving) due to its property of centralization. In this context,
technologies in other domains (e.g., AI, blockchain, 6G and edge computing is proposed as a promising means to solve
digital twin) through interdisciplinary intersection and practice, the shortcomings in cloud computing. Edge computing adopts
releasing the potential for mutual benefit. These promising
integrations need to be further explored and researched. In an open platform with core network capabilities, computing
addition, edge computing provides strong support in applications resources, and data storage to provide users with nearest-end
scenarios such as remote working, new physical retail industries, services, which is regarded as the critical enabling technology
and digital advertising, which has greatly changed the way we towards the 5G era. Edge computing is not a substitute for
live, work and study. In this paper, we present an up-to-date cloud computing but a supplement and expansion, which is
survey of edge computing research. In addition to introducing
the definition, model and characteristics of edge computing, we ideal for real-time data analysis and intelligent processing.
discuss a set of key issues in edge computing and novel solutions In recent years, this world has witnessed the dawning of
supported by emerging technologies in IoE era. Furthermore, we the Internet of Everything (IoE) era. People, process, data,
explore the potential and promising trends from the perspective and things are intelligently connected to the Internet and
of technology integration. Finally, new application scenarios and continuously create the value. However, IoE are magnifying
the final form of edge computing are discussed.
the inherent limitations of the existing computing models,
Index Terms—Edge computing, 6G, artificial intelligence, especially edge computing. Some issues of edge computing
blockchain, microservices, digital twin. architecture, including service migration, security and privacy
preservation, deployment issues of edge node, are difficult
I. I NTRODUCTION to be efficiently addressed by using conventional methods.
Thanks to the rapid development of upcoming technologies,

B ENEFITTING from the rapid development of underlying

technologies, the Internet of Things (IoT) has increas-
ingly permeated our lives and become an essential part of
novel solutions have emerged and have the potential to allevi-
ate these challenges. As for service migration, reinforcement
learning (RL) sovles the service migration issue by using
our daily activities. Millions of devices/sensors are contin- Q-learning algorithm to maximize the return of migration
uously generating data and exchanging important messages and reduce the communication cost and migration cost of
through complex network infrastructures that enable machine- user equipment [4], [5]. In terms of security and privacy
to-machine communication [1], [2]. Statista [3] estimated that preservation, federated learning enables users to collabora-
installed bases of IoT devices worldwide will reach 30.9 tively train an algorithm and keep local data samples on
This work was supported in part by the National Natural Science Foun- the device, effectively avoiding privacy by uploading only
dation of China under Grant 62072409, in part by the Zhejiang Provincial parameters such as weights [6], [7]. Blockchain technology
Natural Science Foundation under Grant LR21F020003, and in part by the [8]–[10] maintains users’ changeable keys and ensures security
Fundamental Research Funds for the Provincial Universities of Zhejiang under
Grant RF-B2020001. (Corresponding author: Feng Xia.) and privacy of edge network without metadata disclosure by
X. Kong is with the College of Computer Science and Technology, Zhejiang nodes during edge coordination. Regarding node deployment
University of Technology, Hangzhou 310023, China, and also with the School issues, microservices can integrate the various aspects of IoT
of Software, Dalian University of Technology, Dalian 116620, China (e-mail:
[email protected]). system architecture layers to facilitate distributed software
Y. Wu and H. Wang are with School of Software, Dalian Univer- development, which addresses the challenge concerning the
sity of Technology, Dalian 116620, China (e-mail: [email protected]; deployment issue of application package [11].
[email protected]).
F. Xia is with the Institute of Innovation, Science and Sustainability, Feder- Edge computing research lies at the intersection of the
ation University Australia, Ballarat 3353, Australia (e-mail: [email protected]). computing model and other disciplines, where the existence

©2022 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to p
IEEE INTERNET OF THINGS JOURNAL 2

of many research opportunities has resulted in a highly active promising trends from the perspective of convergencing edge
area. In addition to solving the challenges mentioned above, computing paradigm with other emerging technologies in IoE
the convergence of edge computing and some technologies era.
(some of them mentioned above), including artificial intelli- The rest of this paper is organized as follows. In section II,
gence (AI), blockchain, 6G, and digital twin driven by other we berifly present the overview of edge computing, including
technologies, can fully unleash their potential and benefit definition, five important concepts, and characteristics. Section
from each other. The mutual benefit of integration between III describes several sigificiant challenges of edge computing
different disciplines deserves further exploration and research. and corresponding solutions supported by integrating with
Moreover, new application scenarios have also emerged, es- upcoming technologies. Our work is focused on section IV,
pecially in the COVID-19 pandemic era, gradually changing where we explore the mutual benefits of the integration of
our daily lives, including remote-working, the new physical emerging technologies and edge computing, discuss some new
retail industry, and digital advertising. Although a number application scenarios and finally propose the final development
of efforts have been conducted, the discussion about novel form of edge computing paradigm. Finally, we conclude our
solutions to existing challenges of edge computing and future work in section V.
directions supported by emerging technologies are missing in
other surveys. II. OVERVIEW OF E DGE C OMPUTING
In this work, we summarize existing efforts and previous In this section, a brief overview of several edge computing
work and present our view on future directions of this research concepts, including mobile cloud computing (MCC), mobile
field. We try to provide a state-of-the-art survey of edge edge computing (MEC), fog computing, cloudlet computing
computing, focusing on existing challenges and a series of and the most popular collaborative cloud-edge-end framework
opportunities from the perspective of integration with emerg- in recent years, is presented.
ing technologies in the IoE era. The main contributions of this
paper are outlined as follows:
A. Paradigms of Edge Computing
• We summary several key challenges in edge computing
In a conventional cloud-centric model, data collected by
including service migration, security and privacy preser-
various terminal devices such as photos, videos and the
vation, and deployment issues of edge node. Meanwhile,
surrounding environment must be transferred to the cloud
the novel solutions supported by emerging technologies
center for processing and then the results are sent back
(e.g., deep learning, blockchain, microservices) are dis-
[16]. The increasing amount of terminal devices and large-
cussed.
scale data transmission have posed a significant challenge
• We further point out the promising prospects from
to cloud computing, especially efficiency, energy usage and
the technology integration perspective, including AI,
latency issues. The emergence of edge computing has the
blockchain, 6G and digital twin, and explore their po-
potential to deal with these challenges. Satyanarayanan et al.
tential and mutual benefit.
[17] define edge computing as a new computing model that
• The new application scenarios supported by edge com-
deploys computing and storage resources (such as cloudlets,
puting, especially in the COVID-19 pandemic, including
microdata centers and fog nodes) at the edge of the network
remote-working, new physical retail industry, and digital
closer to mobile devices or sensors. The ”edge” is regarded
advertising are discussed. We further propose the edgeless
as any computing resource and network between cloud and
computing, which is the ultimate form of edge computing.
terminal devices [12]. Generally speaking, the structure of
Related surveys. One of the most important literature edge computing is generally divided into three layers: terminal
surveys of this field is the work by Shi et al. [12]. Their layer, edge layer, and cloud layer. This hierarchy depicts the
work gave a formal definition of edge computing in academia computational capabilities of edge computing elements and
and pointed out the challenges. However, their proposed their properties, as shown in Fig. 1.
solutions to the existing challenges of edge computing are Despite the rapid development of edge computing, the edge
traditional. They did not examine the opportunities brought by computing community has yet to come to an agreement on
emerging technologies. Khan et al. [13] highlighted the core its standardized definitions, architectures, and protocols [18].
applications and importance of edge computing in real life There have been various architectures at the edge, includ-
scenarios. Nevertheless, their work did not adequately discuss ing MCC, MEC, fog computing, and cloudlet computing.
the possibilities provided by upcoming technologies and future Although their concepts overlap and the boundaries are not
trends of edge computing. Varghese et al. [14] discussed the particularly obvious, we still have appropriate characteristics
opportunities and challenges in this area, but the two parts are to distinguish them. Table I and Table II depict the main
discussed separately and are not fully connected. Carvalho et comparisons amongst them, including attribute and feature.
al. [15] provided another up-to-date survey of the field. Their It’s worth noting that when MCC employs cloudlet as part of
work focused on use cases for each edge computing architec- its design, the qualities are identical to those listed in the corre-
ture and future research directions. A discussion of combining sponding column. Moreover, cloud-edge-end collaboration has
edge computing with technologies from other disciplines is emerged as the most popular operational framework for edge
missing therein. We particularly conduct in-depth discussions computing in recent years. In our work, this group of emerging
on novel solutions to existing challenges in this field as well as technologies is uniformly referred to as “edge computing.”
IEEE INTERNET OF THINGS JOURNAL 3

3) Fog Computing: In 2012, Cisco proposed the concept

of fog computing, which comes from the fact that ”fog” is
closer to the ground than clouds. Fog computing can be treated
as the combination of MCC and MEC. Vaquero et al. [27]
defined fog computing as a scenario where a large number
of heterogeneous (wireless and sometimes autonomous) and
decentralized devices communicate and potentially cooper-
ate among them and perform storage and processing tasks
without the intervention of third parties. These tasks can be
utilized to support basic network functions or new services
and applications running in a sandboxed environment. Fog
computing establishes a continuum [28] between data centers
and data sources to provide users with computing, storage, and
network services, transforming the network into a “assembly
line” for data processing rather than a ”data pipeline”. Take
Fig. 1: Three layers of edge computing architecture and the vacuum cleaner as an example. Centralized fog nodes
collaborative cloud-edge-end framework. (or IoT gateways) continuously collect information about
their surroundings from sensors in the home and activate the
vacuum cleaner when garbage is detected. But in the edge
computing solution, the sensors will judge whether there is
In this section, several conceptions of edge computing are garbage respectively, and then send a signal to start the vacuum
introduced. cleaner.
1) Mobile Cloud Computing: The increasing usage of mo- 4) Cloudlet Computing: Cloudlet is a concept proposed by
bile devices has occupied our lives and affected our way of Satyanarayanan et al. [29] in 2009. They defined cloudlet as
life [19], [20]. Entertainment, games, social networks, and trusted, resource-rich computers that provide storage resources
other applications based mobile terminals are becoming more and computing near the mobile users (near or coexisting with
and more abundant. However, there are many restrictions such a wireless access point). The cloudlet originates from the
as computing, storage capacity, and shared wireless medium. integration of mobile computing, IoT and cloud computing,
Cloud computing is a way to host the execution of applications which plays the role of “data center in box” [18]. Cloudlet
by moving mobile devices to an integrated environment other computing consists of three main features: software-only de-
than itself, which can solve those issues in mobile computing. ployment, proximity deployment, and build based on standard
According to [21], MCC, at its simplest, refers to an infras- cloud technologies. Moreover, cloudlet has become a new
tructure where both data storage and data processing happen alternative technology to carry computing tasks on mobile
outside of the mobile device. Mobile cloud applications offload devices. The challenge of insufficient computing resources of
computational power and data storage from mobile phones to mobile devices by supporting cyber foraging [30] in mobile
the cloud, enabling smartphones and other mobile users to computing is effectively solved. By implementing cloudlet dis-
access applications and mobile computing. MCC is becoming covery, virtual machine (VM) provisioning, and VM handoff,
the dominant way to run mobile applications [22]. cloudlet computing also supports application mobility.
2) Mobile Edge Computing: MEC is regarded as a critical 5) Collaborative Cloud-Edge-End Computing: Cloud-
technology and architectural concept for the transition to edge-end computing is a collaborative processing model that
5G [23], [24]. The European Telecommunications Standards take full advantage of both edge computing and cloud com-
Institute (ETSI) [25] defined it as an IT service environment by puting. As shown in Fig. 1, it is a decentralized network
providing cloud computing capabilities at the edge of a mobile with cloud as the center and layered construction, including
network, aiming to extend the concept of edge computing cloud-edge collaboration, edge-edge collaboration and edge-
to Wi-Fi and other non-3GPP access scenarios. MEC is end collaboration. Edge computing and cloud computing
logically independent of other network parts and is suitable complement and cooperate with each other. Edge servers
for supporting high-security applications. Additionally, MEC process data that requires immediate response. The cloud
servers often have a strong computational capacity, making server provides significant computing capacity and the ability
them ideal for analyzing and processing large amounts of to integrate diverse types of information. Real-time interaction
data. MEC also supports the perception of edge applications, between edge nodes and the cloud can help alleviate the data
especially the wireless access part. The MEC node can obtain heterogeneity issues for the cloud [31]. Furthermore, when
real-time network data such as base station ID, available the storage capabilities of the edge nodes are insufficient,
network bandwidth, and information related to the user’s loca- the cloud may store a portion of the data and transfer it to
tion, thereby achieving link-sensing adaptation [26]. Moreover, the client via the network as needed, therefore conserving
MEC technology enables mobile network operators to offer edge storage resources [32]. Cloud-edge cooperation is critical
additional network information and congestion management in a variety of application scenarios, including content dis-
capabilities to third-party developers, increasing the number tribution networks (CDNs), industrial Internet, energy, smart
of applications and services available to consumers. homes, and smart transportation [33]–[36]. In [37], Hong et al.
IEEE INTERNET OF THINGS JOURNAL 4

proposed IIoT–edge–cloud computing model based on multi- placing a greater demand on network capacity. The high net-
hop computation-offloading for resource-intensive applications work bandwidth consumption means high energy exhaustion
(e.g., 3D sensing, AI processing, big data analytics), with the because a significant amount of electric power to deliver data
goal of reducing energy consumption and processing delay. is required [40]. The emergence of edge computing could
For strengthening 5G heterogeneous network (HetNet) secu- help solve this tricky issue by processing more data utilization
rity, Wei et al. [38] proposed a federated learning empowered changes. By building a group of coordination and management
cloud-edge-end cooperation model. The security is guaranteed system of distributed data centers, edge data centers ensure
by equipping node with an attack detection mechanism at the effective resources utilization. Instead of running 24/7 like
end, edge, and the cloud of 5G HetNet. a cloud data center, resources will become dormant when
they are not needed. Additionally, sensors and IoT devices are
being utilized to monitor energy use, assess energy levels in
B. Edge Computing Characteristics
real time, and provide a real-time perspective of consumption,
Edge computing has some notable features compared to enabling organizations to make dynamic modifications to en-
the cloud computing paradigm. Specially, we discuss three ergy supplement and demand. Meanwhile, cooling data centers
main characteristics that makes up for cloud computing in requires a certain amount of energy. Due to the decreased size
the context of processing massive data, including low latency and output of edge data centers, the tiny data center’s overall
computing, more efficient energy consumption, and transferred energy consumption will decrease proportionately.
computing power. 3) Transferred Computing Power: The emergence of intel-
1) Low Latency Computing: The advantage of the cen- ligent devices such as virtual reality glasses and face recogni-
tralized architecture of cloud computing is that it is easy tion applications altered our way of life profoundly. However,
to manage and maintain. However, it is no longer the opti- these IoT devices and applications have some constraints
mal strategy for geographically distributed applications. Some in battery life and computing power while executing some
popular location-based social networking applications (e.g., complex activities [14]. While moving computing tasks to the
Foursquare, Mingle and Google Now) require computing remote cloud has proven to be a useful solution, the challenge
power closer to the data source to optimize system efficiency of execution delays due to data exchange has always been
and user experience [14]. Edge devices generate numerous present. Edge computing aims to offload heavy processing
data streams, and it is impossible to make real-time decisions from mobile devices to network edge infrastructures such as
when analytics is conducted on a remote cloud. For example, tiny cell base stations with computing and storage capacity,
Boeing 787 generates 5 Gigabytes of data every second therefore improving the user experience dramatically. By of-
[12]. The bandwidth between the aircraft and other satellite fering communication capability from Radio Access Network
equipment or base stations can not withstand such a large (RAN) and making computation resources closer to users,
amount of data due to long request-to-response links. Consider it has the potential to significantly reduce latency, prevent
self-driving cars from Google as another example. A variety of network congestion and extend the battery lifetime of terminal
sensors and cameras capture traffic information in real-time, devices.
generating nearly 1 GB of data per second for complex data
processing and driving decisions. Moreover, Americans drive III. C HALLENGES
an average of 600 hours a year, equivalent to 2.16 million Despite the benefits and promising prospect of edge comput-
seconds or about 2 Petabytes of data per vehicle per year ing, there are still many key issues remaining to be addressed.
[12]. It is almost inconceivable that all data must be sent Especially in IoE era, these major challenges are gradually
to the cloud for processing before responding to the results. being amplified. This section discusses several research chal-
The response time will be extraordinarily long and become a lenges followed by partial solutions supported by emerging
bottleneck. Certain time-sensitive applications suffer the same technologies such as deep learning (DL) technology [41],
issue and impose strict limits on latency between edge devices blockchain technology, and microservices.
and cloud servers. Deploying some of the processing power
closer to the user is an effective way to solve the issues in
the above scenarios. The adoption of edge computing has A. Service Migration
the potential to minimize processing latency and network Application mobility is one of the significant factors that
bandwidth requirements. should be taken into account. Although mobility increases the
2) Efficient Energy Consumption: With the rapid devel- flexibility of applications, it brings new challenges. Computing
opment of computing models (e.g., cloud computing), the resources in mobile applications may switch between multiple
network bandwidth consumption of novel applications (e.g., devices as users move [42]. Resource switching requires
video analytics) has increased sharply in recent years. Ac- migrating the currently running service to another device. The
cording to the Cisco Visual Networking Index research [39], major issues concerning service migration are summarized
video streaming services (e.g., YouTube, Hulu, and Netflix) are below:
expected to consume 79 percent of total network bandwidth • How to appropriately select the edge server to migrate
in 2022. High-bandwidth applications, such as online gaming, the service is a significant issue. The service scope of
Ultra-High-Definition (UHD), or 4K video streaming, require multiple edge servers may overlap. Hence, if a user moves
a broadband connection with a speed of at least 5 Mbit/s, into a specific area within the service area of multiple
IEEE INTERNET OF THINGS JOURNAL 5

TABLE I: Attributes amongst four common edge computing models

Attribute Mobile Cloud Computing Mobile Edge Computing Fog Computing Cloudlet Computing
Rich mobile applications to Extend the concept of edge The network service is not
1. Data center in a box.
Key Features be executed on a different computing to Wi-Fi and other only a data pipeline but also
2. Dynamic VM synthesis.
number of mobile devices. non-3GPP access scenarios. a pipeline of data processing.
Context Awareness High High Medium Low
Mobility Management Not specified Yes Not specified Not specified
Computing Power Low on mobile devices High Not specified Not specified
Application Portability High High Not specified Not specified
Latency High Low Not specified Not specified
Location for Computing Special dedicated buildings Base stations and nearby devices Devices along the routing path Nearby cloudlets
Access Mechanisms Bluetooth, Mobile Networks Mobile Networks Bluetooth, Wi-Fi, LAN Wi-Fi, LAN, WAN
Energy Consumption Low Low Low Low on mobile devices
Availability High Average High Moderate
Service Type Local Less global Less global Local
Standardzation Organizations NIST ETSI, 3GPP OpenFog Consortium, IEEE OpenEdge
Power Consumption Low on mobile devices High Low Moderate
Primary Motivator Academia Academia/Industry Academia/Industry Academia
Distance from Users Far Close Relatively close Close
Central cloud with distributed
Architecture Localized/hierarchial Decentralized/hierarchical Localized
mobile devices
Need to be offered on mobile Need to be offered on Need to be offered on Need to be offered on
Security
terminals and along cloud-to-things edge network equipment (RAN, AP) participant nodes participant nodes
Virtualization
Yes Yes Not specified Yes, extends OpenStack
at the Edge

TABLE II: Features amongst four common edge computing models

Feature Mobile Cloud Computing Mobile Edge Computing Fog Computing Cloudlet Computing
Requirements for Infrastructure Yes Yes Yes Yes
Ultra-low Latency No Yes Yes Yes
Distributed Geographically No Yes Yes Yes
Support for Multiple IoT Applications No No Yes Yes
Standardized Yes Yes Yes No
Support for Large-scale Application No Yes No Yes
Support for Real-time Application No Yes Yes Yes
Location Awareness No Yes Yes Yes
Support for Heterogeneity Yes Yes No No

servers, it is supposed to think over choosing which server challenging.

to migrate service. Moreover, the resources available to • The unpredictability of user mobility and the request pat-
edge servers are an essential factor to consider. When terns increase the difficulty of gaining the optimal service
migrating a current user’s service to another edge server, migration approach. If the user equipment travels inside
the edge server receiving the service migration should a specified region and the borders of the two edge servers
have sufficient resources to satisfy the current user’s are near together due to a particular movement mode, the
service requests [43]. user equipment’s movement will have a dynamic effect
• Migration of services requires consideration of both the on the server performance.
duration and cost of the migration. Certain applications • The diversity of applications and heterogeneity of edge
are latency-sensitive, which means that low latency must servers increase the complexity of service migration.
be guaranteed during service migration. Therefore, a Many users and applications must be considered when
appropriate service migration strategy should attempt to migrating services. Therefore, migrating ongoing services
minimize total migration time. Moreover, the application is more complicated. If the current migration strategy is
providers are mainly concerned with the final revenue. used, unreasonable migrations may be repeated.
The migration decision requires a trade-off between mi- The mathematical models, such as markov decision process
gration benefits and migration costs. Designing a mi- (MDP), are proposed to make effective service migration
gration solution that can optimize migration costs is decisions. However, the global optimal solution obtained by
IEEE INTERNET OF THINGS JOURNAL 6

MDP is based on simple theoretical assumptions, which is the terminal equipment can directly communicate with
subject to the complex condition and a large amount of various any MEC node. However, IoT devices may not be aware
parameters. This feature limits the applicability of mathemat- of the existence of the MEC network, which makes
ical models to service migration. Recently, AI technology, symmetric encryption technology unable to be used to
especially DL, has provided a promising solution for service encrypt messages sent by IoT devices. The same is true
migration decisions by considering complex factors, such as for asymmetric key cryptography technology. Secondly,
heterogeneity of node equippments, the dynamic of network the MEC nodes involved in multiple paths cannot be fully
environment, and real-time requirements of users’ rapid move- trusted, so communication between MEC nodes requires
ment. DL technology, especially RL can continuously learn end-to-end security.
from a large amount of historical data, constantly interact with • Service placement is an important research direction in
dynamic environments and respond quickly to changes [5], MEC, which aims to explore an optimal scheme to
[44]. Especially, RL is able to perceive its environment, take improve mobile users’ quality of service (QoS) [48].
corresponding actions, identify the optimal action to maximize Existing service placement strategies, in particular, are
reward in a given scenario. State, action, and reward are based on the degree to which customers value services.
three key elements in whole process. When MEC and RL However, the degree of customer perference may involves
are integrated, the decision issue of service migration can be some sensitive personal information such as history data,
alleviated. Hence, three key elements above mentioned have locations, and customized needs. Hence, implementing
the new representations in MEC network. The first is “state”, an effective privacy preservation scheme is a challenging
which is able to denote the state of the MEC server that user task.
equippments are currently covered. Next, “action” refers to a Various edge services put forward new needs for adequate
list of all available servers to which the current VM can be privacy protection. Aside from designing an efficient strategy
transferred. The final is “reward”, which refers to the mobile of preserving privacy information, it is vital to consider
user chooses an action mentioned above and gets the final how to combine the traditional privacy protection with the
reward. The ultimate goal is to minimize objective function’s characteristics of edge data processing in a diversified service
value and Q-learning algorithm is utilized to maximize the environment. Moreover, if some emerging technologies (e.g.,
reward of migration. Hence, there are more rewards for action federated learning, blockchain) and edge computing are well-
with lower communication costs and migration costs of user integrated, they can offer great potential for addressing the
equipment at the MEC network. aforementioned partial challenges concerning security and
privacy.
B. Security and Privacy Perservation Incorporating federated learning into edge computing.
In traditional machine learning methods, the training data must
Compared to the cloud computing paradigm, edge comput- be centralized on a single machine or in a cloud center. As
ing can avoid the abuse and theft of users’ privacy data on a distributed DL technology, federated learning [49]–[53] en-
long transmission links by processing partial data at the edge ables users to collaboratively train an algorithm while keeping
of the network. However, new security and privacy challenges local data samples on the devices. The various data generated
have emerged due to the access of multi-class and multi-device by the user equipment (e.g., wireless channel quality, battery
devices in edge networks. The challenges can be denoted as: life, and energy consumption) and edge nodes (e.g., computing
• Edge computing devices are usually close to the user. load, wireless communication quality, task queue) are utilized
Therefore, MEC nodes adjacent to the user may collect as raw training data for model input. Federated learning
sensitive information [45], [46], including the user’s avoids the privacy leakage problem caused by uploading these
identity, location information and application usage. Take sensitive data to the cloud center, only submitting learned
the smart home system as an example. The hackers model weights to update [54]. In dealing with the challenges
can easily track electricity usage to identify whether a of service placement in MEC, federated learning allows users
house is unoccupied, which increases the possibility of to send the trained results instead of uploading all the users’
items being stolen in the house. Furthermore, centralized privacy data, such as perference information, to the cloud
control becomes extremely difficult due to the discrete center. User privacy information is well protected in this way.
nature of MEC nodes. Integration with blockchain technology. Blockchain is a
• Traditional approaches of security and privacy protection, distributed ledger technology that does not require centralized
such as certificates and Public Key Infrastructure (PKI) control and is protected by encryption technology [55], [56].
authentication, may not be suitable for being used on It provides a secure, transparent, and non-tamperable platform
edge infrastructures [47]. In dynamic changes of MEC for network data communication, sharing, and transactions.
nodes, nodes must mutually verify the newly formed Furthermore, the blockchain guarantees the automatic exe-
MEC network. In addition, MEC nodes also need to cution of predefined rules and terms by intelligent contract
restrict or reject service requests from malicious and technology, and protects data privacy and account security by
compromised nodes. the asymmetric encryption algorithm.
• The device communication in the MEC network mainly The blockchain here refers to the ability of network partici-
includes the communication between the IoT devices and pants to record in the distributed billing system. The core parts
the MEC nodes and between the MEC nodes. Firstly, of blockchain, such as consensus protocol, ledger topology,
IEEE INTERNET OF THINGS JOURNAL 7

and incentives, will be extended in integrated systems to in edge computing. The operators are able to offer different
accommodate different levels of edge computing systems and services such as local offloading services, edge computer room
combinations [8]. The integration covers the fundamental rental, unified IaaS capabilities when facing various corporate
layers of blockchain and major capabilities of edge computing, customers. Different with large enterprises, operators provide
which provides more secure large-scale data storage and unified planning and deployment of IaaS and PaaS platforms
effective computing without the need for costly encryption for small and medium-sized enterprises, which applies to
overheads. Block chain technology can realize security au- situations where edge nodes are scarce, and the payoff of
thentication, secure data storage and secure computing to management is limited [61]. However, the third-party PaaS
protect the security and privacy of edge network. On one platform and the management of third-party applications in
hand, blockchain technology allows each user to maintain their the edge system need further exploration and improvement.
own changeable keys, which is convenient for users to offer
The fourth is reliability assurance. Protecting the physical
access and manage data without the involvement of any third
environment of edge nodes is also challenging due to the
parties [8]. On the other hand, coordination on the peer-to-
lack of effective measures [62], including data backup, data
peer basis is allowed by blockchain’s pseudonymous property,
recovery, and audit measures. Compared with storing data
and metadata (e.g., source, destination, and content) will not
in a stable cloud computing environment, the attackers may
be disclosed to anyone.
modify or delete the user data on the edge node to destroy
some evidence. Take the traffic supervision scenario as an
C. Deployment Issues example [58], [63]–[65]. The high-precision camera on the
There are still many challenges in the deployment of edge road records the normal and abnormal conditions, trajectory
computing nodes such as business selection, investment re- data, and illegal records of the vehicles. In a traffic accident,
turns, and operating model. By introducing the microservices these data are the critical evidence to find criminal attacks.
technique, the challenge concerning deployment of application The offenders can get away with the law by attacking data on
packages will be effictively alleviated. edge nodes. Similarly, household consumption stored in edge
nodes and personal health information in electronic medical
Business needs and scenario selection is the first issue that
systems may also become the target of attacks. Moreover,
needs to be considered in the deployment scenario, especially
extreme weather conditions (e.g., snowstorms, strong wind)
in the 5G era. Whether it is edge computing for individual
will lead to immeasurable damage of edge data, which is
users in the enhanced Mobile Broadband (eMBB) scenario, or
unacceptable for enterprises and customers. The edge nodes
edge computing for vertical field (e.g., live games, Internet of
also can not provide adequate measures to recover data. When
vehicles [57], [58], smart manufacturing), deployment needs
constructing the entire edge system, it is necessary to leverage
to fully consider the capacity and the feasibility of the business
infrastructures coordination to ensure physical reliability and
scenario.
utilize multiple backup measures to ensure data reliability.
The second is network index and investment returns. The
main stakeholder [59] in the edge ecosystem is classified into The final one is the deployment issue of application pack-
two categories: infrastructure owners and software developers. ages. The virtualization technology, such as container, aims
The first one usually refers to operators and cloud providers at distributing packaged applications as low-overhead virtual
such as Google Cloud and Amazon Cloud. They are responsi- machines (VMs) to edge servers [66]. However, it is not trivial
ble for collecting and storing data, maintaining and managing to decompose monolithic cloud applications into distributed
software and hardware facilities [60]. This business pattern packages and install them into hierarchical IoT system topolo-
is transparent to users, and subscribers only need to pay for gies, especially to meet applications’ specific demands (e.g.,
the service without knowing the technical details. The latter QoS, performance). Therefore, it is necessary to introduce
mainly includes content providers and startups companies. novel programming paradigms that integrate the various as-
They support the deployment of edge servers, create added- pects of IoT system architecture layers to facilitate distributed
value applications and help enrich and expand the innovative software development. Microservices is a promising approach
services. Technology without economic benefits is hard to for modularizing applications and services at the process level.
sustain. In an edge computing system, the maintenance of A single application is decomposed into non-interfering atomic
hardware and software is particularly challenging due to the services in a microservices architecture. Each service unit
geographical dispersion of edge nodes. Whether the cloud that performs a specific task, consumes only a tiny amount
service provider or the content provider pays for maintenance of computing resources in order to software developers can
and management expenses must be considered. Furthermore, quickly build it. These units are operated, updated, and de-
it is also necessary to effectively reduce the cost of users’ ployed independently so that the developer team can carry out
network usage. The closer the computing resources is to the the continuous delivery of functions. The general applications
edge of the network, the better experience for users. However, are divided into several small modules and deployed to the
it will leads to the decrease of access users, the reduction of edge nodes [11]. Each module can compute, store and utilize
network edge revenue and the increase of the total cost. the network resources without affecting other modules. The
The third is the operating model and management. Infras- convergence of distributed IoT and microservices will facilitate
tructure as a service (IaaS), Platform as a service (PaaS), and package deployment optimization for service delivery and
Software as a service (SaaS) in cloud computing also exist address the deployment challenges of application packages.
IEEE INTERNET OF THINGS JOURNAL 8

Moreover, the number of communication rounds required to

train the model is reduced by controlling the number of local
update steps (increasing the amount of calculation for each
user equipment), which shows that federated learning has a
good balance between communication cost and computation
cost in edge network [78]. Lin et al. [7] tries to combine fed-
erated learning with meta learning. By using a small amount
of local data to train a meta model, it is possible to quickly
adapt to the task of the target edge node. Transfer learning
is also a hot research topic [79]–[81], which aims to improve
the target task’s performance by transferring the existing task’s
knowledge to the current target task. As the upstream of the
machine learning model, transfer learning adapts to multi-
source heterogeneous data collected on different edge devices
through decoupling model, and the training data and training
Fig. 2: The overview of edge intelligence. time requirements of the target field have been significantly
reduced. Zhou et al. [82] proposed a CNNEF framework that
effectively detects abnormal activities in the edge computing
IV. P ROMISING D IRECTIONS environment by using embedded features for transfer learning,
This section will discuss promising directions with the which overcomes the limitations of the traditional Convolu-
burgeoning of IoE, including emerging technologies, new tional Neural Network (CNN) model. Abu et al. [83] proposed
application scenarios, and the final development form of edge a family health detection edge computing method based on
computing. transfer learning. Their work only needs to fine tune a small
amount of labeled data to train the model, which can greatly
reduce the health crisis caused by the epidemic. More and
A. Emerging Technologies more methods will be combined with edge computing to make
1) Artificial Intelligence: The concept of AI was proposed the edge more intelligent. Furthermore, we discussed several
in 1956. AI refers to the technology that uses algorithms to other promising research directions about edge intelligence.
make machines imitate human thinking to solve problems.
So far, outstanding achievements have been widely produced • Accelerate AI services through edge computing. At
in some fields (e.g., Computer Vision, Natural Language present, there have been some studies using DL tech-
Processing, and Intelligent Robots). According to Gartner’s nology to optimize mobile edge communication systems.
prediction, by 2022, more than 80% of enterprises will include However, it is also essential to develop special methods
AI components in their IoT projects [67]. In particular, as for optimizing learning computing tasks by combining
the most important branch of AI, DL technology has brought edge computing characteristics with edge devices. For
the vigorous development of AI applications and services. DL a large number of AI tasks with different priorities and
can recognize patterns, detect and analyze abnormal data (e.g., requirements (such as CPU and memory), it is also crucial
population distribution, air quality, temperature, and humidity) to find the right collaboration edge node and allocate
of edge devices, and then send the intelligent analysis results the appropriate resources. Moreover, Game Theory algo-
to decision-making applications [68]–[72]. rithms may be applied to edge intelligence to accelerate
In fact, edge computing and AI are progressively merging, AI services through edge computing.
mutually benefiting from the realization of edge intelligence, • Improve the edge intelligence efficiency of real-time
as shown in Fig.2. Edge computing can provide rich real-time mobile communication systems towards 5G. In the 5G
training data and diverse operating environments for AI models era, communication links with extremely low latency
[73]–[75]. Meanwhile, AI can provide edge computing with and high reliability are required. However, general op-
powerful learning capabilities. Edge intelligence is projected timization and prediction schemes based on DL require
to move as many DL computations as feasible from the quite a long running time to converge to the result,
cloud to the edge, enabling a variety of distributed, low- which is not suitable for mobile edge systems, especially
latency, and dependable intelligent services [76]. Moreover, edge computing tasks that require a rapid response at
edge intends to integrate DL into the edge to provide dynamic, the millisecond level. Edge intelligence should provide
adaptive edge maintenance and management [4]. To deal with various support for different types of services to eliminate
the dynamically changing network environment, emerging the delays caused by caching, networking, and computing
learning methods in DL also bring new opportunities to edge [48].
intelligence [77]. Take federated learning as an example. • Build the incentive and business model of edge intel-
In addition to addressing the privacy challenge of service ligence. The realization of AI services involves three
placement of MEC network in Section III, federated learning parties: mobile operators, service providers (SPs)/content
can also solve the key problem that performance is affected providers (CPs), and mobile users. The user equippments
by unbalanced data and harsh communication environment. (UEs) of a small number of mobile users can provide
IEEE INTERNET OF THINGS JOURNAL 9

ness, consistency, and validity of edge data and regulations

throughout their life cycle. This can effectively solve the
mobility problem between heterogeneous nodes located at
the edge of the network. Moreover, the scattered data in the
edge network is more likely to be lost or stored incorrectly.
The transaction management provided by the blockchain has
potential to address this challenge. For example, the record
of edge data becomes unchangeable once it is stored on the
transaction ledger. In this context, edge devices can perform
large-scale computing or collaboration in an untrusted net-
work environment [89]. Blockchain technology also provides
transaction transparency and data integrity for edge network,
and allows the replication of publicly verified data records
across edge nodes [90]. Moreover, issues such as network
congestion, link failure, and privacy leakage may occur in the
Fig. 3: The common benefits of edge computing and data interaction and service migration between heterogeneous
blockchain. devices and edge servers due to illegal attacks. The solution
to this issue is to add the block mining process to each edge
device, verify the data transaction, and protect and connect
remaining computing resources after meeting their own through the immutable ledger to improve the security of the
computing requirements. However, a large number of edge network.
UEs may depend on the AI computing capabilities of In current cloud-based blockchain services, Microsoft pro-
edge nodes and other user devices. In addition, edge vides blockchain as a service based on the Azure cloud
nodes spanning multiple mobile operators need to meet platform, and cloud-hashing in the UK offers Bitcoin mining
the AI computing requirements of SPS and CPS. There- services to users who purchase services online. The entire
fore, designing a reliable and effective edge intelligence process does not require users to install and deploy any
incentive framework holds great promise. equipments. Other companies such as IBM, Google, and
2) Blockchain Technology: The combination of blockchain Oracle integrate blockchain ledger into their business-level
and edge computing has become an unstoppable trend [84], cloud services. The blockchain transaction economic model
[85]. In addition to addressing security and privacy issues based on edge computing is missing and will be the focus of
mentioned in section III, the convergence also brings some future research.
novel opportunities and directions, as shown in Fig.3. First 3) 6G: The deployment and comprehensive promotion of
consider the benefits of edge computing for blockchain. The 5G have brought the world the ultimate user experience of
transaction time in blockchain is highly related to the perfor- mobile internet, and communication capabilities and service
mance and processing power of the server [86], [87]. Some quality have been improved by leaps and bounds. However,
high-performance processors provided by AMD and NVIDIA the next ten years will transition from Industry 4.0 to Industry
in the edge computing infrastructure can support the whole X.0. 5G cannot cope with application scenarios requiring
transaction. Moreover, in the centralized network architecture, microseconds and Tps levels, such as holographic remote
the data stream that blockchain nodes communicate must pass transmission, remote surgery, and extended reality (XR). As
through the whole network before returning. Combining with the next generation of revolutionary communication technol-
edge computing can eliminate the need for data traversing the ogy, 6G wireless network will support a transmission rate of
core network and solve the problem of network latency in TB per second and a significantly low transmission delay. The
the blockchain. Then consider the benefits of blockchain for speed is 50 times that of 5g network and the delay is half that
edge computing. The edge computing infrastructure is still dis- of 5g network.
persed among telecom operators, which means that application The development of 6G will continue to accelerate the
developers will have to interact with each telecom operator evolution from the IoT to the Intelligent Internet of Things
to ensure that applications can run between consumers in a (IIoT). When data is transferred from cloud servers to edge
country and across borders. The usual solution is to aggregate devices, 6G provides functions such as high-speed security,
operators into an entity, such as cloud service providers such as ultra-high reliability, and ultra-low latency, which greatly
AWS and Azure, with a unified control interface in the public reduces data loss rate and bit error rate and ensures seamless
cloud field, so as to monopolize the cloud service market data connection between edge devices [91]. Consider the the
[88]. Instead, blockchain technology may be utilized to build unmanned aerial vehicle (UAV) system under COVID-19 pan-
a decentralized edge computing marketplace that connects demic. To deal with the epidemic situation, various data such
suppliers of edge infrastructure with those in need, without as the image, digital, CT scanning and other data obtained by
relying on a single point of control. the UAV will be directly sent to the edge server for processing.
The distributed architecture makes edge computing chal- The edge will train the AI model to accelerate the processing
lenging to manage heterogeneous networks. By building dis- and data analysis and then send results to relevant government
tributed control on edge nodes, blockchain ensures the correct- departments. All devices are connected via 4G networks in this
IEEE INTERNET OF THINGS JOURNAL 10

process, and the transmission rate is slow, only 100 Mbps. • Working in Post-epidemic Era. The outbreak of
Even 5G networks with speeds of up to 20 Gbps cannot meet COVID-19 has had a profound impact on the way we live
the needs of future emergencies. For example, the deterioration and the operating model of human society [95]. It will not
of critical patients requires immediate feedback. The social be resolved fast and will continue for a long time. Around
distance and isolated persons in the floating population also 300 million people have been infected with coronavirus
need to be monitored in real-time. In these critical scenarios, worldwide to date, with over 4 million deaths. Not only
6G can provide a microsecond response [92] to deal with human beings, but all walks of life are also experienc-
challenges. Meanwhile, the deployment of edge computing ing unprecedented tests. The weaknesses of the Internet
in the 6G network will improve the system performance, architecture that we rely on every day are completely
realize core network traffic optimization, and provide a novel exposed under epidemic. Enterprises must also respond to
network service. Furthermore, compared with cloud service the ever-changing business world and promote the long-
providers such as Amazon and Google, operators in edge term transition to intelligent edge computing solutions
computing systems will have more advantages in computing to cope with the post-epidemic era. Firstly, significant
and communication resources, which allows them to regain changes have taken place in the way employees work.
their market position and increase the added value of their We call it the “Remote Revolution”. Remote collaboration
services in 6G era. and home office have become the norm. Both consumers
4) Digital Twin: The Digital Twin (DT) originated from and companies are looking for positive ways to meet
the NASA project of the US Department of Defense, which office requirements during the quarantine period. Most
is utilized to maintain and guarantee the health of aerospace of the virtual desktop infrastructure (VDI) hosted in the
vehicles by virtualization technology. It makes full use of vari- cloud allows any device to access it at any time. However,
ous physical models and combines sensor data, algorithms and VDI architecture has some shortcomings, including High
decision analysis to realize real-time virtual mapping of the construction cost and over-reliance on network environ-
physical world on the information platform, thereby reflecting ment. By introducing some implementation schemes of
the entire life cycle of the corresponding physical entity [93], edge computing architecture, such as Intelligent Desktop
[94]. A series of enabling technologies (e.g., AI, blockchain, Virtualization (IDV) and Virtual OS Infrastructure (VOI),
6G) drive digital twins to support some application scenarios employees may more efficiently utilize their devices at
such as smart home, smart city, and smart transportation. home, resulting in cost savings, less hardware needs, and
Fig. 4 shows the applications of digital twins. Digital twin more flexible operation [96]. Different from VDI, which
network (DTN) [31] is a network of many-to-many mappings concentrates all desktop computing resources in the data
formed from many one-to-one DTs, which aims to enable center and sends interfaces to terminal devices, edge com-
the dynamic interaction and coordinated evolution of a large puting takes a more decentralized approach to meeting
number of physical and virtual objects. By linking several DT operational technological requirements while simplifying
nodes in DTN, physical objects and virtual twins are able to administration and deployment functions. Edge comput-
interact, cooperate, exchange information, perform activities, ing solutions enable IT to manage and protect desktop
and establish an information-sharing network. images and devices while maintaining high performance,
Physical to physical (P2P) and physical to virtual (P2V) mobility, and flexibility for users. Thanks to the rapid
communication in DTN network require ultra-low sensor development of edge computing paradigm, organizations
delay, data processing delay, and feedback delay in some may rapidly and inexpensively introduce desktop virtual-
time-sensitive application scenarios such as medication control ization without investing in infrastructure.
and remote surgery. The convergence with edge computing is • Physical Retail Industry. Edge computing shows ex-
empowered to alleviate this issue. The cloud-edge-end frame- traordinary marketing potential in user personalization
work offloads the computing tasks of the twin terminal to the and in-depth positioning and provides users with an un-
edge network, which mainly solves the device’s shortcomings precedented digital product experience. In order to meet
in terms of resource storage, computing performance, and social distancing requirements and comply with epidemic
energy efficiency. Cooperative cloud-edge-end computing is prevention policies, the physical retail industry has been
able to provide DTN with low-latency computation, increased hit like never before. It is undergoing a transformation
computational ability with constrained resources, and con- to survive and compete with the online retail industry.
tinuous update. Processing and analyzing closer to the edge The ways of self-help include building a real-time supply
will reduce the communication delay of mutual mapping in chain, forming a fully automatic manufacturing chain and
physical space and virtual space. Digital twins systems based providing customers with a personalized shopping expe-
on edge deployment will have greater flexibility in defining, rience. By deploying edge computing systems, brick-and-
developing, and utilizing real-time IoT systems. mortar retailers [97] are committed to enabling customers
to seamlessly transition from the physical to the digital
experience via new channel technology. The store may
B. New Application Scenarios offer a virtual reality experience to attract additional con-
The emergence of edge computing paradigm has supported sumers, as well as real-time inventory data presented on
the new applicaiton scenarios, including remote-working, the touch displays. Moreover, if a customer does not have a
physical retail industry, and digital advertising. favorite product, the system will intelligently recommend
IEEE INTERNET OF THINGS JOURNAL 11

Fig. 4: Digital twin framework.

other products based on the historical preferences of this C. Trends

customer. All data will no longer simply pass through While edge computing has facilitated several advances
the edge but will continue to integrate existing data to and displayed amazing promise, it is still a long way from
track target users in order to improve user experience. achieving its final form. Edge computing will finally transform
In addition to providing users with a better shopping into “edgeless computing”. Edgeless indicates that the edge
experience, edge-based systems cases will include smart will collaborate with itself and interact with other devices
shelves with dynamic pricing, digital signage, and more. directly rather than via the cloud. Specifically, the cloud-edge-
Additionally, edge computing technology enhances the end architecture will abandon the cloud level and eliminate
offline experience of the store, which continues to operate the need for computing in the cloud center. The terminal
even when the Internet is down or the main equipment will have the strong computational power and
• Digital Advertising. The data processing model com- the ability to handle large amounts of data. Take the smart
bined with edge computing can accelerate data-driven wearable device (such as Apple watch) as an example. The
advertising decisions and decrease users’ time for ad- smart watch is capable of monitoring the user’s nighttime sleep
vertising, which has made a huge contribution to digital quality. The device gathers different data (e.g., depth and light
advertising such as video advertising [98], [99]. In other sleep duration, heart rate fluctuations, blood oxygen content),
cases, edge computing will make the processing power uploads them to a central server for extensive analysis, and
available on consumer devices grow, and data will be then presents the user’s sleep quality findings and makes
processed on the user device itself. We have seen edge recommendations for future sleep adjustments. However, if
computing used to enhance user privacy. Edge computing the smart watch uses its computational capacity to do a full
can keep the user’s data from leaving the local device, analysis and then directly provides the user with a sleep report,
and the user’s local device can use personal data to make the efficiency will be higher without the involvement of the
advertising decisions, which enables third parties to ac- cloud. The data processing takes place entirely within the
curately place advertisements without directly obtaining terminal device, and it is no longer necessary to go back into
the data. Edge computing enhances the privacy of Internet the cloud. In this context, the cloud server may function as a
advertising by minimizing the amount of data transmitted data repository, and the edge nodes will also have their tiny
by consumer devices. databases, rather than relying on the cloud. The more dispersed
the edge devices are, the more flexible the network is and the
IEEE INTERNET OF THINGS JOURNAL 12

less traffic it carries. The vision that can be handled quickly [14] B. Varghese, N. Wang, S. Barbhuiya, P. Kilpatrick, and D. S. Nikolopou-
anywhere will surely be realized in the future. los, “Challenges and Opportunities in Edge Computing,” in 2016 IEEE
International Conference on Smart Cloud (SmartCloud), 2016, pp. 20–
26.
V. C ONCLUSION [15] G. Carvalho, B. Cabral, V. Pereira, and J. Bernardino, “Edge computing:
current trends, research challenges and future directions,” Computing,
Today, edge computing is one of the most effective solutions vol. 103, no. 5, pp. 993–1023, 2021.
to address some challenges associated with enormous volumes [16] L. Jiao, R. Friedman, X. Fu, S. Secci, Z. Smoreda, and H. Tschofenig,
of data that various industries generate and consume every “Cloud-based computation offloading for mobile devices: State of the
art, challenges and opportunities,” in 2013 Future Network Mobile
day. In this work, we have described the brief overview of Summit, 2013, pp. 1–11.
edge computing; discussed the main existing challenges in [17] M. Satyanarayanan, “The Emergence of Edge Computing,” Computer,
edge computing and novel solutions supported by emerging vol. 50, no. 1, pp. 30–39, 2017.
[18] K. Bilal, O. Khalid, A. Erbad, and S. U. Khan, “Potentials, trends, and
technologies; and explored the possibility of technologies in- prospects in edge technologies: Fog, cloudlet, mobile edge, and micro
tegration, finally discussed several new applications scenarios data centers,” Computer Networks, vol. 130, pp. 94–120, 2018.
and proposed the final form of edge computing. We believe [19] F. Xia, B. Jedari, L. T. Yang, J. Ma, and R. Huang, “A signaling game
for uncertain data delivery in selfish mobile social networks,” IEEE
that the future of edge computing lies in the organic integration Transactions on Computational Social Systems, vol. 3, no. 2, pp. 100–
with emerging technologies. Convergence will transform man- 112, 2016.
ufacturing and services, create actionable business intelligence, [20] F. Xia, A. M. Ahmed, L. T. Yang, J. Ma, and J. J. Rodrigues, “Exploiting
social relationship to enable efficient replica allocation in ad-hoc social
and develop flexible business ecosystems. networks,” IEEE Transactions on Parallel and Distributed Systems,
vol. 25, no. 12, pp. 3167–3176, 2014.
R EFERENCES [21] S. Yi, C. Li, and Q. Li, “A Survey of Fog Computing: Concepts,
Applications and Issues.” Association for Computing Machinery, 2015,
[1] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and p. 37–42.
M. Ayyash, “Internet of Things: A Survey on Enabling Technologies, [22] M. Othman, F. Xia, A. N. Khan et al., “Context-aware mobile cloud
Protocols, and Applications,” IEEE Communications Surveys & Tutori- computing and its challenges,” IEEE Cloud Computing, vol. 2, no. 3,
als, vol. 17, no. 4, pp. 2347–2376, 2015. pp. 42–49, 2015.
[2] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, “A Survey [23] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, “Mobile edge
on Internet of Things: Architecture, Enabling Technologies, Security computing—A key technology towards 5G,” ETSI white paper, vol. 11,
and Privacy, and Applications,” IEEE Internet of Things Journal, vol. 4, no. 11, pp. 1–16, 2015.
no. 5, pp. 1125–1142, 2017. [24] P. Mach and Z. Becvar, “Mobile Edge Computing: A Survey on Archi-
[3] L. S. Vailshery, “Internet of things (iot) and non-iot tecture and Computation Offloading,” IEEE Communications Surveys &
active device connections worldwide from 2010 to 2025,” Tutorials, vol. 19, no. 3, pp. 1628–1656, 2017.
Website, 2021, https://www.statista.com/statistics/1101442/ [25] M. Patel, B. Naughton, C. Chan, N. Sprecher, S. Abeta, A. Neal et al.,
iot-number-of-connected-devices-worldwide/. “Mobile-edge computing introductory technical white paper,” White
[4] X. Wang, Y. Han, V. C. M. Leung, D. Niyato, X. Yan, and X. Chen, paper, mobile-edge computing (MEC) industry initiative, vol. 29, pp.
“Convergence of Edge Computing and Deep Learning: A Comprehen- 854–864, 2014.
sive Survey,” IEEE Communications Surveys & Tutorials, vol. 22, no. 2, [26] Y. Mao, C. You, J. Zhang, K. Huang, and K. B. Letaief, “A Survey
pp. 869–904, 2020. on Mobile Edge Computing: The Communication Perspective,” IEEE
[5] N. C. Luong, D. T. Hoang, S. Gong, D. Niyato, P. Wang, Y.-C. Communications Surveys & Tutorials, vol. 19, no. 4, pp. 2322–2358,
Liang, and D. I. Kim, “Applications of Deep Reinforcement Learning 2017.
in Communications and Networking: A Survey,” IEEE Communications [27] L. M. Vaquero and L. Rodero-Merino, “Finding Your Way in the
Surveys & Tutorials, vol. 21, no. 4, pp. 3133–3174, 2019. Fog: Towards a Comprehensive Definition of Fog Computing,” ACM
[6] D. C. Nguyen, M. Ding, Q.-V. Pham, P. N. Pathirana, L. B. Le, SIGCOMM computer communication Review, vol. 44, no. 5, pp. 27–32,
A. Seneviratne, J. Li, D. Niyato, and H. V. Poor, “Federated Learning 2014.
Meets Blockchain in Edge Computing: Opportunities and Challenges,” [28] M. Chiang, S. Ha, F. Risso, T. Zhang, and I. Chih-Lin, “Clarifying
IEEE Internet of Things Journal, vol. 8, no. 16, pp. 12 806–12 825, 2021. Fog Computing and Networking: 10 Questions and Answers,” IEEE
[7] S. Lin, G. Yang, and J. Zhang, “A Collaborative Learning Framework via Communications Magazine, vol. 55, no. 4, pp. 18–20, 2017.
Federated Meta-Learning,” in 2020 IEEE 40th International Conference [29] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, “The Case for
on Distributed Computing Systems (ICDCS). IEEE, 2020, pp. 289–299. VM-Based Cloudlets in Mobile Computing,” IEEE pervasive Comput-
[8] R. Yang, F. R. Yu, P. Si, Z. Yang, and Y. Zhang, “Integrated Blockchain ing, vol. 8, no. 4, pp. 14–23, 2009.
and Edge Computing Systems: A Survey, Some Research Issues and [30] S. Nithya, M. Sangeetha, K. A. Prethi, K. S. Sahoo, S. K. Panda,
Challenges,” IEEE Communications Surveys & Tutorials, vol. 21, no. 2, and A. H. Gandomi, “SDCF: A Software-Defined Cyber Foraging
pp. 1508–1532, 2019. Framework for Cloudlet Environment,” IEEE Transactions on Network
[9] A. Fitwi, Y. Chen, and S. Zhu, “A Lightweight Blockchain-Based and Service Management, vol. 17, no. 4, pp. 2423–2435, 2020.
Privacy Protection for Smart Surveillance at the Edge,” in 2019 IEEE [31] Y. Wu, K. Zhang, and Y. Zhang, “Digital Twin Networks: A Survey,”
International Conference on Blockchain (Blockchain), 2019, pp. 552– IEEE Internet of Things Journal, vol. 8, no. 18, pp. 13 789–13 804, 2021.
555. [32] Q. Luo, S. Hu, C. Li, G. Li, and W. Shi, “Resource Scheduling in Edge
[10] H. Guo, W. Li, M. Nejad, and C.-C. Shen, “Access Control for Electronic Computing: A Survey,” IEEE Communications Surveys & Tutorials,
Health Records with Hybrid Blockchain-Edge Architecture,” in 2019 vol. 23, no. 4, pp. 2131–2165, 2021.
IEEE International Conference on Blockchain (Blockchain), 2019, pp. [33] F. Xia, J. Wang, X. Kong, D. Zhang, and Z. Wang, “Ranking station im-
44–51. portance with human mobility patterns using subway network datasets,”
[11] F. Guo, B. Tang, M. Tang, H. Zhao, and W. Liang, “Microservice IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 7,
Selection in Edge-Cloud Collaborative Environment: A Deep Reinforce- pp. 2840–2852, 2020.
ment Learning Approach,” in 2021 8th IEEE International Conference [34] C. Chen, Q. Liu, X. Wang, C. Liao, and D. Zhang, “semi-traj2graph:
on Cyber Security and Cloud Computing (CSCloud)/2021 7th IEEE Identifying fine-grained driving style with gps trajectory data via multi-
International Conference on Edge Computing and Scalable Cloud task learning,” IEEE Transactions on Big Data, 2021.
(EdgeCom), 2021, pp. 24–29. [35] C. Chen, S. Jiao, S. Zhang, W. Liu, L. Feng, and Y. Wang, “Tripimputor:
[12] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, “Edge Computing: Vision Real-time imputing taxi trip purpose leveraging multi-sourced urban
and Challenges,” IEEE internet of things journal, vol. 3, no. 5, pp. 637– data,” IEEE Transactions on Intelligent Transportation Systems, vol. 19,
646, 2016. no. 10, pp. 3292–3304, 2018.
[13] W. Z. Khan, E. Ahmed, S. Hakak, I. Yaqoob, and A. Ahmed, “Edge [36] A. Tolba, “A two-level traffic smoothing method for efficient cloud-iot
computing: A survey,” Future Generation Computer Systems, vol. 97, communications,” Peer-to-Peer Networking and Applications, vol. 14,
pp. 219–235, 2019. no. 5, pp. 2743–2756, 2021.
IEEE INTERNET OF THINGS JOURNAL 13

[37] Z. Hong, W. Chen, H. Huang, S. Guo, and Z. Zheng, “Multi-Hop Computing: from standard to actual infrastructure deployment and
Cooperative Computation Offloading for Industrial IoT–Edge–Cloud software development,” ETSI White paper, pp. 1–41, 2019.
Computing Environments,” IEEE Transactions on Parallel and Dis- [60] R. Bruschi, F. Davoli, G. Lamanna, C. Lombardo, S. Mangialardi, and
tributed Systems, vol. 30, no. 12, pp. 2759–2774, 2019. J. F. Pajo, “Enabling Edge Computing Deployment in 4G and Beyond,”
[38] Y. Wei, S. Zhou, S. Leng, S. Maharjan, and Y. Zhang, “Federated in 2020 6th IEEE Conference on Network Softwarization (NetSoft).
Learning Empowered End-Edge-Cloud Cooperation for 5G HetNet IEEE, 2020, pp. 237–241.
Security,” IEEE Network, vol. 35, no. 2, pp. 88–94, 2021. [61] H. Song, R. Dautov, N. Ferry, A. Solberg, and F. Fleurey, “Model-based
[39] C. Team, “Cisco annual internet report (2018–2023) white paper,” fleet deployment of edge computing applications,” in Proceedings of the
Website, 2020, https://www.cisco.com/c/en/us/solutions/collateral/ 23rd ACM/IEEE International Conference on Model Driven Engineering
executive-perspectives/annual-internet-report/white-paper-c11-741490. Languages and Systems, 2020, pp. 132–142.
html. [62] K. Okafor, “Dynamic reliability modeling of cyber-physical edge com-
[40] J. Mocnej, M. Miškuf, P. Papcun, and I. Zolotová, “Impact of edge com- puting network,” International Journal of Computers and Applications,
puting paradigm on energy consumption in IoT,” IFAC-PapersOnLine, vol. 43, no. 7, pp. 612–622, 2021.
vol. 51, no. 6, pp. 162–167, 2018. [63] X. Kong, M. Li, T. Tang, K. Tian, L. Moreira-Matias, and F. Xia, “Shared
[41] X. Su, S. Xue, F. Liu, J. Wu, J. Yang, C. Zhou, W. Hu, C. Paris, subway shuttle bus route planning based on transport data analytics,”
S. Nepal, D. Jin, Q. Z. Sheng, and P. S. Yu, “A comprehensive survey on IEEE Transactions on Automation Science and Engineering, vol. 15,
community detection with deep learning,” IEEE Transactions on Neural no. 4, pp. 1507–1520, 2018.
Networks and Learning Systems, 2022. [64] X. Kong, M. Li, K. Ma, K. Tian, M. Wang, Z. Ning, and F. Xia, “Big
[42] K. Cao, Y. Liu, G. Meng, and Q. Sun, “An Overview on Edge Computing trajectory data: A survey of applications and services,” IEEE Access,
Research,” IEEE access, vol. 8, pp. 85 714–85 728, 2020. vol. 6, pp. 58 295–58 306, 2018.
[43] Y. Siriwardhana, P. Porambage, M. Liyanage, and M. Ylianttila, “A Sur- [65] X. Kong, X. Liu, B. Jedari, M. Li, L. Wan, and F. Xia, “Mobile
vey on Mobile Augmented Reality With 5G Mobile Edge Computing: crowdsourcing in smart cities: Technologies, applications, and future
Architectures, Applications, and Technical Aspects,” IEEE Communica- challenges,” IEEE Internet of Things Journal, vol. 6, no. 5, pp. 8095–
tions Surveys & Tutorials, vol. 23, no. 2, pp. 1160–1192, 2021. 8113, 2019.
[44] X. Kong, B. Zhu, G. Shen, T. C. Workneh, Z. Ji, Y. Chen, and Z. Liu, [66] T. Leppanen, C. Savaglio, L. Loven, T. Jarvenpaa, R. Ehsani, E. Pelto-
“Spatial-temporal-cost combination based taxi driving fraud detection nen, G. Fortino, and J. Riekki, “Edge-Based Microservices Architecture
for collaborative internet of vehicles,” IEEE Transactions on Industrial for Internet of Things: Mobility Analysis Case Study,” in 2019 IEEE
Informatics, vol. 18, no. 5, pp. 3426–3436, 2021. Global Communications Conference (GLOBECOM), 2019, pp. 1–7.
[45] B. H. Husain, S. Askar et al., “Survey on Edge Computing Security,” [67] C. Pemberton, “3 ai trends for enterprise computing,” 2017, https://www.
International Journal of Science and Business, vol. 5, no. 3, pp. 52–60, gartner.com/smarterwithgartner/3-ai-trends-for-enterprise-computing.
2021. [68] Z. Chang, S. Liu, X. Xiong, Z. Cai, and G. Tu, “A Survey of
[46] L. Gao, T. H. Luan, B. Gu, Y. Qu, and Y. Xiang, Privacy-Preserving in Recent Advances in Edge-Computing-Powered Artificial Intelligence of
Edge Computing. Springer, 2021. Things,” IEEE Internet of Things Journal, vol. 8, no. 18, pp. 13 849–
[47] P. Ranaweera, A. D. Jurcut, and M. Liyanage, “Survey on Multi-Access 13 875, 2021.
Edge Computing Security and Privacy,” IEEE Communications Surveys
[69] F. Xia, J. Liu, H. Nie, Y. Fu, L. Wan, and X. Kong, “Random Walks:
& Tutorials, vol. 23, no. 2, pp. 1078–1124, 2021.
A Review of Algorithms and Applications,” IEEE Transactions on
[48] F. Xia, L. Liu, B. Jedari, and S. K. Das, “Pis: A multi-dimensional
Emerging Topics in Computational Intelligence, vol. 4, no. 2, pp. 95–
routing protocol for socially-aware networking,” IEEE Transactions on
107, 2020.
Mobile Computing, vol. 15, no. 11, pp. 2825–2836, 2016.
[70] X. Kong, K. Wang, S. Wang, X. Wang, X. Jiang, Y. Guo, G. Shen,
[49] D. Ye, R. Yu, M. Pan, and Z. Han, “Federated Learning in Vehicular
X. Chen, and Q. Ni, “Real-Time Mask Identification for COVID-19:
Edge Computing: A Selective Model Aggregation Approach,” IEEE
An Edge-Computing-Based Deep Learning Framework,” IEEE Internet
Access, vol. 8, pp. 23 920–23 935, 2020.
of Things Journal, vol. 8, no. 21, pp. 15 929–15 938, 2021.
[50] X. Kong, K. Wang, M. Hou, X. Hao, G. Shen, X. Chen, and F. Xia,
“A Federated Learning-Based License Plate Recognition Scheme for [71] X. Kong, S. Tong, H. Gao, G. Shen, K. Wang, M. Collotta, I. You,
5G-Enabled Internet of Vehicles,” IEEE Transactions on Industrial and S. K. Das, “Mobile edge cooperation optimization for wearable
Informatics, vol. 17, no. 12, pp. 8523–8530, 2021. internet of things: a network representation-based framework,” IEEE
[51] L. U. Khan, W. Saad, Z. Han, E. Hossain, and C. S. Hong, “Federated Transactions on Industrial Informatics, vol. 17, no. 7, pp. 5050–5058,
Learning for Internet of Things: Recent Advances, Taxonomy, and Open 2020.
Challenges,” IEEE Communications Surveys & Tutorials, vol. 23, no. 3, [72] X. Kong, G. Duan, M. Hou, G. Shen, H. Wang, X. Yan, and M. Collotta,
pp. 1759–1799, 2021. “Deep reinforcement learning based energy efficient edge computing for
[52] X. Zhou, W. Liang, J. Ma, Z. Yan, and K. I.-K. Wang, “2d federated internet of vehicles,” IEEE Transactions on Industrial Informatics, 2022.
learning for personalized human activity recognition in cyber-physical- [73] A. K. Ghadiyaram et al., “A Compendium On Edge Computing Where
social systems,” IEEE Transactions on Network Science and Engineer- Ai Can Meet,” Turkish Journal of Computer and Mathematics Education
ing, 2022. (TURCOMAT), vol. 12, no. 10, pp. 4118–4126, 2021.
[53] A. Li, J. Sun, B. Wang, L. Duan, S. Li, Y. Chen, and H. Li, “Lotteryfl: [74] F. Xia, K. Sun, S. Yu, A. Aziz, L. Wan, S. Pan, and H. Liu, “Graph
Empower edge intelligence with personalized and communication- learning: A survey,” IEEE Transactions on Artificial Intelligence, vol. 2,
efficient federated learning,” in 2021 IEEE/ACM Symposium on Edge no. 2, pp. 109–127, 2021.
Computing (SEC). IEEE, 2021, pp. 68–79. [75] X. Kong, Q. Chen, M. Hou, A. Rahim, K. Ma, and F. Xia, “Rmgen:
[54] W. Y. B. Lim, N. C. Luong, D. T. Hoang, Y. Jiao, Y.-C. Liang, Q. Yang, A tri-layer vehicular trajectory data generation model exploring urban
D. Niyato, and C. Miao, “Federated Learning in Mobile Edge Networks: region division and mobility pattern,” IEEE Transactions on Vehicular
A Comprehensive Survey,” IEEE Communications Surveys & Tutorials, Technology, 2022.
vol. 22, no. 3, pp. 2031–2063, 2020. [76] F. Xia and J. Ma, “Building smart communities with cyber-physical
[55] M. Firdaus and K.-H. Rhee, “On Blockchain-Enhanced Secure Data systems,” in Proceedings of 1st international symposium on From digital
Storage and Sharing in Vehicular Edge Computing Networks,” Applied footprints to social and community intelligence, 2011, pp. 1–6.
Sciences, vol. 11, no. 1, p. 414, 2021. [77] M. Hou, J. Ren, D. Zhang, X. Kong, D. Zhang, and F. Xia, “Network
[56] K. S. Bharadwaj, S. Dharanikota, A. Honawad, and K. Chandrasekaran, embedding: Taxonomies, frameworks and applications,” Computer Sci-
“Blockchain Research and Applications: A Systematic Mapping Study,” ence Review, vol. 38, p. 100296, 2020.
IC-BCT 2019, pp. 141–159, 2020. [78] F. Xia, L. Liu, J. Li, A. M. Ahmed, L. T. Yang, and J. Ma, “Beeinfo:
[57] A. Tolba, “Content accessibility preference approach for improving Interest-based forwarding using artificial bee colony for socially aware
service optimality in internet of vehicles,” Computer Networks, vol. 152, networking,” IEEE Transactions on Vehicular Technology, vol. 64, no. 3,
pp. 78–86, 2019. pp. 1188–1200, 2014.
[58] W. Wang, F. Xia, H. Nie, Z. Chen, Z. Gong, X. Kong, and W. Wei, [79] D. Wu, X. Han, Z. Yang, and R. Wang, “Exploiting Transfer Learning for
“Vehicle trajectory clustering based on dynamic representation learning Emotion Recognition Under Cloud-Edge-Client Collaborations,” IEEE
of internet of vehicles,” IEEE Transactions on Intelligent Transportation Journal on Selected Areas in Communications, vol. 39, no. 2, pp. 479–
Systems, vol. 22, no. 6, pp. 3567–3576, 2020. 490, 2020.
[59] D. Sabella, A. Alleman, E. Liao, M. Filippou, Z. Ding, L. G. Baltar, [80] A. Sufian, C. You, and M. Dong, “A Deep Transfer Learning-based
S. Srikanteswara, K. Bhuyan, O. Oyman, G. Schatzberg et al., “Edge Edge Computing Method for Home Health Monitoring,” in 2021 55th
IEEE INTERNET OF THINGS JOURNAL 14

Annual Conference on Information Sciences and Systems (CISS), 2021, Technica Corviniensis-Bulletin of Engineering, vol. 13, no. 2, pp. 41–
pp. 1–6. 45, 2020.
[81] X. Liu, W. Yu, F. Liang, D. Griffith, and N. Golmie, “Toward Deep [99] H. Lee and C.-H. Cho, “Digital advertising: present and future
Transfer Learning in Industrial Internet of Things,” IEEE Internet of prospects,” International Journal of Advertising, vol. 39, no. 3, pp. 332–
Things Journal, vol. 8, no. 15, pp. 12 163–12 175, 2021. 341, 2020.
[82] Y. Zhou, M. Han, J. He, L. Liu, X. Xu, and X. Gao, “Abnormal Activity Xiangjie Kong (M’13-SM’17) received the B.Sc.
Detection in Edge Computing: A Transfer Learning Approach,” in 2020 and Ph.D. degrees from Zhejiang University,
International Conference on Computing, Networking and Communica- Hangzhou, China. He is currently a Professor with
tions (ICNC), 2020, pp. 107–111. College of Computer Science and Technology, Zhe-
[83] A. Sufian, A. Ghosh, A. S. Sadiq, and F. Smarandache, “A Survey on jiang University of Technology. Previously, he was
Deep Transfer Learning to Edge Computing for Mitigating the COVID- an Associate Professor with the School of Software,
19 Pandemic,” Journal of Systems Architecture, vol. 108, p. 101830, Dalian University of Technology, China. He has
2020. published over 160 scientific papers in international
[84] S. Islam, S. Badsha, S. Sengupta, H. La, I. Khalil, and M. Atiquzzaman, journals and conferences. His research interests in-
“Blockchain-Enabled Intelligent Vehicular Edge Computing,” IEEE Net- clude mobile computing, network science, and data
work, vol. 35, no. 3, pp. 125–131, 2021. science. He is a Senior Member of IEEE.
[85] T. Li, Y. Fang, Z. Jian, X. Xie, Y. Lu, and G. Wang, “ATOM:
architectural support and optimization mechanism for smart contract fast
update and execution in blockchain-based iot,” IEEE Internet Things J.,
vol. 9, no. 11, pp. 7959–7971, 2022.
[86] S. Guo, X. Hu, S. Guo, X. Qiu, and F. Qi, “Blockchain meets edge
Yuhan Wu received the B.Sc. degree in network en-
computing: A distributed and trusted authentication system,” IEEE
gineering from the Heilongjiang University, Harbin,
Transactions on Industrial Informatics, vol. 16, no. 3, pp. 1972–1983,
China, in 2019. He is currently pursuing the master’s
2019.
degree with the School of Software, Dalian Uni-
[87] Y. Qu, L. Gao, T. H. Luan, Y. Xiang, S. Yu, B. Li, and G. Zheng,
versity of Technology, Dalian, China. His current
“Decentralized privacy using blockchain-enabled federated learning in
research interests include transfer learning, graph
fog computing,” IEEE Internet of Things Journal, vol. 7, no. 6, pp.
neural networks, and edge computing.
5171–5183, 2020.
[88] Y. Du, Z. Wang, and V. Leung, “Blockchain-Enabled Edge Intelligence
for IoT: Background, Emerging Trends and Open Issues,” Future Inter-
net, vol. 13, no. 2, p. 48, 2021.
[89] V. C. Leung, X. Wang, F. R. Yu, D. Niyato, T. Taleb, and S. Pack, “Guest
Editorial: Special Issue on Blockchain and Edge Computing Techniques
for Emerging IoT Applications,” IEEE Internet of Things Journal, vol. 8,
no. 4, pp. 2082–2086, 2021.
[90] A. Awad Abdellatif, L. Samara, A. Mohamed, A. Erbad, C. F. Chi- Hui Wang received the B.Sc. degree in cyber en-
asserini, M. Guizani, M. D. O’Connor, and J. Laughton, “MEdge-Chain: gineering from the Shandong Jianzhu University ,
Leveraging Edge Computing and Blockchain for Efficient Medical Data Jinan, China, in 2020. He is currently pursuing the
Exchange,” IEEE Internet of Things Journal, vol. 8, no. 21, pp. 15 762– master’s degree with the School of Software, Dalian
15 775, 2021. University of Technology, Dalian, China. His current
[91] J. Cao, W. Feng, N. Ge, and J. Lu, “Delay Characterization of Mobile- research interests include federated learning, graph
Edge Computing for 6G Time-Sensitive Services,” IEEE Internet of neural networks, and edge computing.
Things Journal, vol. 8, no. 5, pp. 3758–3773, 2020.
[92] H. H. H. Mahmoud, A. A. Amer, and T. Ismail, “6G: A compre-
hensive survey on technologies, applications, challenges, and research
problems,” Transactions on Emerging Telecommunications Technologies,
vol. 32, no. 4, p. e4233, 2021.
[93] X. Chen, M. Hou, T. Tang, A. Kaur, and F. Xia, “Digital twin mobility
profiling: A spatio-temporal graph learning approach,” in The 7th IEEE
International Conference on Data Science and Systems (DSS), Haikou, Feng Xia (M’07-SM’12) received the BSc and
China. IEEE, 2021. PhD degrees from Zhejiang University, Hangzhou,
[94] Z. Jiang, Y. Guo, and Z. Wang, “Digital twin to improve the virtual- China. He was Full Professor and Associate Dean
real integration of industrial IoT,” Journal of Industrial Information (Research) in School of Software, Dalian University
Integration, vol. 22, p. 100196, 2021. of Technology, China. He is Associate Professor
[95] P. Zhang, C. Wang, N. Kumar, C. Jiang, Q. Lu, K.-K. R. Choo, and and former Discipline Leader (IT) in Institute of
J. J. Rodrigues, “Artificial Intelligence Technologies for COVID-19-Like Innovation, Science and Sustainability, Federation
Epidemics: Methods and Challenges,” IEEE Network, vol. 35, no. 3, pp. University Australia. Dr. Xia has published 2 books
27–33, 2021. and over 300 scientific papers in international jour-
[96] A. R. Watson, R. Wah, and R. Thamman, “The Value of Remote nals and conferences (such as IEEE TAI, TKDE,
Monitoring for the COVID-19 Pandemic,” Telemedicine and e-Health, TNNLS, TBD, TCSS, TNSE, TETCI, TC, TMC,
vol. 26, no. 9, pp. 1110–1112, 2020. TPDS, TETC, THMS, TVT, TITS, TASE, ACM TKDD, TIST, TWEB,
[97] A. Biswas, A. Jain et al., “Survey on Edge Computing–Key Technology TOMM, WWW, AAAI, SIGIR, CIKM, JCDL, EMNLP, and INFOCOM). His
in Retail Industry,” in Computer Networks and Inventive Communication research interests include data science, artificial intelligence, graph learning,
Technologies. Springer, 2021, pp. 97–106. anomaly detection, and systems engineering. He is a Senior Member of IEEE
[98] P. Verma, “Edge computing: challenges for real time analytics,” Acta and ACM.