Page 94 - Proceedings of the 2018 ITU Kaleidoscope

‎ 2018 ITU Kaleidoscope Academic Conference‎

Network Machine learning techniques Purposes
functions
Planning and Support vector machine - Classification of service requirements
design Gradient boosting decision tree - Forecasting trend, user behavior
Spectral clustering - Configuration of parameters
Reinforcement learning
Operation and K-mean clustering - Clustering cells, users, devices
management Deep neural network - Routing, forwarding, traffic control
Reinforcement learning - Decision making for dynamic resource
control, policy formulation
- Reconfiguration of parameters
Monitoring Spectral clustering - Clustering of syslog data
K-mean clustering - Classification of operation modes
Support vector machine - Forecasting resource utilization trend
Deep neural network
Fault detection Principal component analysis - Classification of operation data
Independent component analysis - Detection of network anomaly
Logistic regression - Predicting unusual behavior
Bayesian networks
Security Deep neural network - Clustering users and devices
Principal component analysis - Detecting malicious behavior
- Intrusion detection
Table 1. Network functions and relevant machine learning techniques.

sum of the expected values of the rewards of all future and unsupervised learning, such as spectral clustering, can
steps. be used for classifying a new service in one of eMBB,
eMTC, and URLLC categories on the basis of requirements
In between supervised and unsupervised learning, there is in terms of bandwidth, latency, bit-error rate, etc. Similarly,
semi-supervised learning, which is given a few labeled reinforcement learning can be applied for reasoning to
example data but can perform on a large collection of determine the appropriate values of parameters for the
unlabeled data. Deep learning, which is based on artificial optimal network setup. It would help in designing network
neural network, belongs to a broader family of machine slices that would be suitable for continuously evolving new
learning methods. Its learning can be supervised, semi- services and use cases by provisioning optimal amount of
supervised or unsupervised. Unlike the other machine resources.
learning techniques that require highly tuned and many
rules to solve specific problems, deep learning techniques Operation and management involve tasks for efficient use
can successfully deal with huge volumes of data to learn of resources while optimally satisfying the service and user
and recognize abstract patterns by using vast, virtual neural requirements all the time. They require understanding the
networks [12]. variations in system states, learn uncertainties,
(re)configure the network, forecast immediate challenges,
Machine learning techniques are being considered useful and suggest appropriate solutions timely. The resource
for various functions of network services; mainly, planning allocation and management take into account node
and design, operation, control and management, monitoring, computation capacity, link bandwidth, radio spectrum, and
fault detection, and security. They enable machine to energy currently available and in use. They require
recognize patterns and anomalies that human may not clustering of cells (for frequency allocation and power
notice or take unacceptably long time. Table 1 shows the management), users and devices for intelligent mobility
list of relevant machine learning techniques for the management or establishing device to device (D2D)
automation of these networking functions. networks for optimal management of available spectrum
and energy in mobile devices. Unsupervised learning
techniques such as K-mean clustering are suitable for the
Planning and design involve decision making based on the clustering. Similarly, deep learning has been shown to be
information provided about service requirements and effective in control of heterogeneous network traffic [12] to
expected user behavior. The design process can exploit achieve high throughput of packet processing.
machine learning techniques for data acquisition (extracting Reinforcement learning is applicable in decision-making
relevant data), processing data for knowledge discovery, for the reconfiguration of network parameters and dynamic
and using the knowledge for reasoning and decision adjustment of resources such as channel selection. Smart
making [11]. Supervised learning techniques, such as reconfiguration of parameters for faster adaptation of
support vector machine and gradient boosting decision tree,

– 78 –