Page 200 - Kaleidoscope Academic Conference Proceedings 2021
P. 200
2021 ITU Kaleidoscope Academic Conference
the current situation of the network and its expected future to system completely. Intelligent algorithms have also shown
make the best decision for the best outcome. Computational to be beneficial for implementing this technique. With the
methods need to have complete data on the system’s purpose of providing security in the network, deep neural
variables; instead, intelligent methods have the ability to network and principal component analysis algorithms are
consider all variables and find the equilibrium point where utilized and the goal is to learn the behavior of users and
a KPI is met while having the maximum revenue possible [41]. devices in order to detect malicious behavior and intrusion
Deep Reinforcement Learning (DRL) is a perfect candidate for in the network [16].
this problem, considering all variables in the network and
learning its behavior to determine the best decision for 5. CHALLENGES AND OPEN ISSUES
reaching the desired outcome. Deep reinforcement
techniques operate as follows: Due to the daily advances of technology and industry, there
are still a number of challenges and research topics that need
• When a new request arrives, the system takes an action to to be addressed to ensure the proper functioning of the sliced
maximize the long-term award. This action can be either 5G networks. Communication networks are undergoing a
accepting or rejecting the request. major evolutionary transformation to meet the needs of a
• After taking the action, the algorithm evaluates the action large number of users and connected devices, and to enable
and the award by interacting with the system and then the the operation of newly introduced services in an adaptive
information from the evaluation is used to retrain the manner. Establishing isolation between network slices is
network. very important and it is possible to achieve complete
• When noticing a change in system behavior, the isolation with today's technologies. But with complete
algorithm should be retrained to adapt to the new isolation, network efficiency decreases significantly, not
behavior [17]. being able to share the limited resources of ours. Therefore,
one of the challenges in this area is exploiting intelligent
4.3.2 Fault management techniques for finding the attainable degree of isolation for
Fault management is another paramount function in slice each slice based on its use cases. As mentioned, the isolation
automation. Fault management tasks are as mentioned below: of the slices is of great importance. Not only because of
resource sharing, but also for the security of slices. In a
• Analyzing the system activities and classifying into two poorly isolated slices network, damage and attack in one
groups of normal and flawed slice will harm other slices too, causing a huge disaster in the
• Recognizing usual and unusual user behavior and traffic network’s operation. Therefore, the challenge here is to find
• Detecting the precise location of error a solution considering the necessity of slice isolation
• Trying to fix the flaws alongside the need of resource sharing. These issues and
challenges need to be addressed in order to provide a good
Principal component analysis, independent component
analysis, logistic regression and Bayesian networks are efficient model for 5G networks and beyond.
algorithms that are used for fault management in networks.
All these techniques try to detect faults in the network’s 6. CONCLUSION
operation and predict unusual behavior in the future.
Furthermore, unsupervised learning techniques are useful In this paper, we presented the 5G network slicing concept
for intrusion detection and spoofing attacks [16]. and its standards, use cases and architecture. Moreover, we
discussed the critical role of AI and machine learning
4.3.3 Security techniques for automation of network slicing functions and
mentioned some of the intelligent algorithm used for each
Security is a major issue in 5G network slicing. For having a function. With the help of AI-based solutions, we can
secure network, there are 3 subfunctions that need to be address complex problems emerging in different slicing
carried out and the system can achieve an approving level of functions such as designing, resource management and fault
security with the help of artificial intelligence in these management. Considering the high potential of artificial
subfunctions. These main security system subfunctions are: intelligent techniques, we can conclude that the challenges
• Analyzing the traffic, service requests and status of slice mentioned above, maintaining a balance between isolation
• Spotting security vulnerabilities and detecting attacks in degree of slices and resource sharing, can be addressed by
the slice exploiting ML and DL techniques. Moreover, since beyond
• Taking the proper action against threats and attacks 5G networks have higher data rates and a vast number of
new emerged use cases, these challenges become more
Artificial intelligence has a worthwhile impact in analyzing severe. Therefore, the learning algorithms-based solutions
the network traffic as well as detecting attacks and making will get deeper and more complex.
the right move against the attacks and to handle the
vulnerabilities in the system. However, the main role of REFERENCES
artificial intelligence in security is in analyzing the traffic,
service requests and status of each slice. One of the effective [1] B. NGMN Alliance, R. El Hattachi, and J. Erfanian,
actions against attacks is quarantining the contaminated slice. “NGMN 5G White paper", 2015. “5G PPP Architecture
In this way, we get to restrict the attack and its following Working Group View on 5G Architecture View on 5G
damage to other slices without having to shut down the
– 138 –