Page 89 - Proceedings of the 2018 ITU Kaleidoscope

CONSIDERATION ON AUTOMATION OF 5G NETWORK SLICING WITH
MACHINE LEARNING

Ved P. Kafle, Yusuke Fukushima, Pedro Martinez-Julia, and Takaya Miyazawa

National Institute of Information and Communications Technology, Tokyo, Japan

ABSTRACT resources by using network function virtualization (NFV)
Machine learning has the capability to provide simpler and software defined networking (SDN) technologies. A
solutions to complex problems by analyzing a huge volume network slice is configured and controlled dynamically by
of data in a short time, learning for adapting its software, which is also called network softwarization [2].
functionality to dynamically changing environments, and
predicting near future events with reasonably good A huge volume of data of complex nature need to be
accuracy. The 5G communication networks are getting analyzed to carry out smart decision for the design,
complex due to emergence of unprecedentedly huge construction, deployment, operation, administration and
number of new connected devices and new types of services. management of a network slice so that it can effectively
Moreover, the requirements of creating virtual network satisfy the quality of service (QoS) requirements of the
slices suitable to provide optimal services for diverse users service intended to be delivered through it, despite time-
and applications are posing challenges to the efficient varying workloads and network conditions. It is difficult
management of network resources, processing information for a human to create and operate network slices manually
about a huge volume of traffic, staying robust against all by processing the large volumes of data in a short time.
potential security threats, and adaptively adjustment of Therefore, it is being necessary to automate these tasks.
network functionality for time-varying workload. In this Machine learning techniques are enabler for the automation
paper, we introduce about the envisioned 5G network of network slicing functions. Machine learning has the
slicing and elaborate the necessity of automation of capability of sensing (e.g., anomaly detection), mining (e.g.,
network functions for the design, construction, deployment, service classification), prediction (e.g., forecasting user or
operation, control and management of network slices. We traffic trend), and reasoning (e.g., configuration of system
then revisit the machine learning techniques that can be parameters for adaptation) [3]. Namely, it has capabilities
applied for the automation of network functions. We also to analyze a huge volume of data in a very short time, learn
discuss the status of artificial intelligence and machine to adjust the system to time-varying environments, make
learning related activities being progressed in standards prediction of future events with reasonably good accuracy,
development organizations and industrial forums. and prescribe proactive solutions.

Keywords— Machine learning, artificial intelligence,
5G network, slicing, standardization Machine learning has been considered for the automation
of various functions of network operation and management,
such as resource management, on-demand and adaptive
1. INTRODUCTION
network configuration, service creation and orchestration,
The 5G networks are expected to enable ultra-high-speed fault detection, security, mobility management, user
data transmission (about 10Gbps) that would be about 1000 experience enhancement, and dynamic adjustment of policy
times the speed of current LTE networks, connect massive [3]. Recent advancements in big data, cloud computing,
number of devices that would be 10-100 times the number cyber physical systems (CPS), and Internet of Thing (IoT)
of existing mobile phones, ultra-low latency (about 1ms) have become enabler for the realization of artificial
that would be 5 times lower than the latency of LTE intelligence (AI) and machine learning based network
networks, and highly energy efficient with 10 times longer control and management technologies [4].
battery life [1]. They should be possessed with the
capability to satisfy the diverse requirements of various In this paper, we describe the envisioned 5G network
services for the fully connected smart society, such as slicing and elaborate the necessity of automation of
enhanced mobile broadband (eMBB), massive machine type network functions for the design, construction, deployment,
communication (mMTC), and ultra-reliable and low latency operation, control and management of network slices. We
communication (URLLC). Since each of these services then provide an overview of machine learning techniques
requires different types of network capabilities (e.g., eMBB that can be applied to the automation of network functions.
services require very high bandwidth and mMTC services We also discuss the status of artificial intelligence and
require ultra-dense connectivity), they cannot be provided machine learning related activities being progressed in
effectively over a single homogeneous network. Therefore, various standards development organizations (SDO) and
5G networks are being enabled with the capability of the industrial forums. The main contribution of this paper is to
on-demand constructions of network slices with sufficient highlight the essential functions of network slicing and

978-92-61-26921-0/CFP1868P-ART @ 2018 ITU – 73 – Kaleidoscope