IEEE Consumer Electronics Magazine - May/June 2022 - 33

of the network constraints such as overhead,
latency, and memory storage.
TRUSTED HYBRID LEARNING
FRAMEWORK TO SECURE EDGE
COMPUTING
The main components of our trusted hybrid
learning framework are detailed in the following.
Hybrid Learning Security Framework
As illustrated in Figure 2, the proposed framework
is equipped with two main security
engines: feature engineering function and GANbased
cyber-attack detection engine.
Feature Engineering Function (FEF): As shown
in Figure 1, this software function should be
deployed in the 5G core system. It focuses on
determining the attractive feature vectors to be
utilized by the attack detection engine. Feature
determination relies on static and dynamic
approaches. In a static approach, the cyber security
expert feeds the feature engineering engine
with feature vectors related to some identified
attacks. Here, the security expert analyzes the
security alerts generated by the centralized
security monitoring system, Security Operation
Center, and then defines a set of feature vectors
related to the new attack pattern, i.e., zero-day
threats. In a dynamic approach, the feature vectors
are updated dynamically, i.e., without intervention
of the security expert. Here, an
unsupervised machine learning technique based
on game theory is used by the FEF engine to
determine the potential future attack behaviors
and then identify the new features related to
each zero-day attack. The security game is a noncooperative
game between the FEF located at the
5G core level and an attacker (located within the
neighborhood of the security engine). Here, the
action of FEF is to monitor the suspected targets
(located within the same neighborhoods of the
security engines) and the action of an attacker is
to launch an attack against the neighborhood
target. The purpose of this security game is to
determine the state (Nash Equilibrium) at which
the FEF monitors the target instigated by the
attacker at the same time. In this case, the FEF
identifies the attack features prior to the attack.
Readers are referred to the article by Sun et al.12
May/June 2022
Table 1. Comparison Among Edge Security Frameworks.
False
Edge security
frameworks
Baidya et al.5
Hassan et al.6
Cai et al.7
Chen et al.8
Samy et al.9
Subramaniam
and Kaur10
Detection
rate
Low
positive
rate
Low
Low
Medium Medium Medium
Medium Medium Medium
High
High
High
Medium High
Meduim High
Medium High
Computational
overhead
Memory
storage
Medium
High
High
Low
Medium
Medium
for more details about the dynamic determination
of features by using game theory.
GAN-based Cyber-Attack Detection Engine: This
engine is mainly based on a hybrid learning
approach to detect the attacks instigated within
the MEC network as shown in Figure 1. As
explained in the introduction section, the positive
aspect of the hybrid learning approach is to
achieve a high zero-day attack detection rate by
using a robust machine learning algorithm, while
the false positive rate is reduced with the help of
the cyber security expert intervention. The
cyber-attack detection engine is based on the
GAN approach to monitor the feature vectors
delivered by the feature engineering function and
to detect/predict attacks. Our GAN-based cyberattack
detection engine consists of a generator
and two discriminator systems: the ATDD and
the ANDD. GAN-based cyber-attack detection
engine is a closed-loop system. The generator is
Figure 2. Hybrid learning security framework.
33
IEEE Consumer Electronics Magazine - May/June 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - May/June 2022
