IEEE Consumer Electronics Magazine - January/February 2023 - 114

Security, Trust and Privacy Solutions for Intelligent Internet of Vehicular Things
Edge computing: The feature extraction and
training process for AI-based intrusion detection
need a large-scale quantity of computation.
Due to the restrictions of resource,
power, and communication latency, onboard
devices and cloud platforms will not be able
to store and process the massive data generated
by such computing-intensive tasks.19 To
enhance the computing capacity and decrease
the latency,20 the intrusion detection tasks
may be offloaded to the nearby mobile edge
servers or devices.21 Intelligent caching and
computational offloading (such as power-efficient/energy-efficient
data offloading) schemes
to effectively decide where to perform intrusion
detection tasks are also worth investigating.
Furthermore, the design of multilayer IDS
based on cloud-edge collaboration is also a
topic worth exploring.
Federated learning (FL): AI-based IDSs for IoV
inevitably need to dynamically update the detection
model in the process of carrying out detection
tasks. The current methods are to aggregate
large amounts of data distributed in different
mobile devices or nodes (such as vehicles, onboard
devices) and upload the data directly to
the central server of the cloud platform for training,
which will result in serious communication
overhead and delay. FL allows multiparticipant
to collaboratively train and share the model
without directly accessing the original data.
Each participant uploads the model trained by
the local dataset to the central server for global
model aggregation. It can not only efficiently utilize
bandwidth, but also reduce response
latency.22,23 Therefore, FL is very suitable for
designing IoV IDS.
Transfer learning (TL): With the development
of intelligent IoV, many new attack methods with
different feature distributions will emerge. The
detection accuracy of the original model will
decrease significantly. Thus, collecting the
labeled data from these new attacks to retrain
the ML models is very necessary. However, this
work will take a lot of labor and time. The model
may fail to detect these attacks accurately
because it cannot get enough labeled data within
a short time. TL is an effective solution for little
training data problems. By reusing the trained
model for similar problems, the existing
114
problems and the original problems will be
mapped to the same feature space, so as to
quickly train an effective detection model, and
get better intrusion detection results.24
Active defense: There are a large number of
IDSs for IoV. However, intrusion detection is
only a means to discover the security threats in
IoV. In order to ensure the IoV real security, it is
necessary to develop defensive measures and
respond immediately when an attack has been
detected. The active defense system is able to
make an automatic response by canceling
users, terminating processes, closing systems,
disconnecting connections, and other ways to
prevent or reduce intrusion hazards. Therefore,
theactivedefense forIoV is ameaningfulfuture
research direction. Although the automated
response actions are effective in mitigating the
damage causedby attacks, theremay be some negative
risks. For example, the isolation measures
may result in partial interruption of communication
links and even affect the normal operation of
the network. Thus, how to use AI technology to
make optimal active defense strategies will also be
a considerable research subject.2
Privacy protection: For intrusion detection algorithms,
especially AI schemes, such as ML, there
are huge risks and vulnerabilities of privacy disclosure
in the input data and model parameters.
Therefore, intrusion detection for IoV also needs
to meet the needs of privacy protection.10 At present,
privacy protection has three main research
directions: differential privacy method based on
noisedisturbance, secure multiparty computation,
and homomorphic encryption. The FL framework
mentioned previously can also be applied in collaboration
with these technologies. The privacy
protection mechanismhas a certain impact on efficiency,
including the impact on computational
overhead and communication costs. In practical
application scenarios, especially in IoV intrusion
detection tasks, how to balance detection performance
and meet usability conditions (such as
lightweight and low latency) while satisfying privacy,
remains a great challenge.
In addition, the intelligent service data privacy,
location privacy of connected vehicles, user identity
privacy, and other information in the IoV also
need a very strong security guarantee. Therefore,
privacy protection technologies such as the ones
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2023

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2023
