Next-Gen AI-Driven Network

Automation
According to Gartner Forecasts, the Worldwide Artificial Intelligence Software Market will reach $62 Billion
by the end of 2022, an increase of 21.3% from 2021. It is also said that market growth will accelerate as
organizations progress in their AI maturity.

Table 1. AI Software Market Forecast by Use Case, 2021-2022, Worldwide (Millions of U.S. Dollars)
2021 2021 2022 2022
Segment Revenue Growth (%) Revenue Growth (%)
Knowledge Management 5,466 17.6 7,189 31.5
Virtual Assistants 6,210 12.0 7,123 14.7
Autonomous Vehicals 5,703 13.7 6,849 20.1
Digital Workplace 3,593 13.7 4,309 20.0
Crowdsourced Data 3,483 13.6 4,171 19.8
Others 27,049 14.1 32,827 21.4
Total 51,503 14.1 62,486 21.3

www.brillio.com/ [email protected]
No longer limited to providing basic phone and internet service, the telecom industry is at the epicenter of
technological growth, led by mobile and 5G broadband services in the Internet of Things (IoT) era. This
growth is expected to continue due to the rapid adoption of artificial intelligence (AI) in the telecom industry.
Projections show that the global AI in telecommunication market size will reach $14.99B by 2027, from
$11.89B in 2020, at a CAGR of 42.6% during 2021-2027.

Introduction
While it is true that we are in the middle of one of the Artificial Intelligence hypes, it is also true that the
combination of unprecedented computation power and data availability, with new variations of
well-seasoned Machine Learning algorithms, is dramatically changing the optimization strategies for large
ICT industries. Especially, the telecommunications industry has always had to deal with complex systems,
stochastic contexts, combinatorial problems, and hard-to-predict users; Machine Learning-aided
optimization was just waiting there to be used by telecoms. In this paper, we introduce some basic Machine
Learning concepts and discuss how they can be used in the context of telecommunications networks,
particularly in wired and wireless networks.

Enterprise AI Strategy & Implementation

In Automated networking, the use of ML technology can help automate routine network operations
involving visualization, analysis, and troubleshooting. It has been a constant industry effort to replace
manual network management processes with various ways of automation and now through ML-based
algorithms which are indeed the de facto future.

The ML technology relies on various data sources like packet-level telemetry, account log messages, KPIs,
and faults data generated by network devices and then analyzed by an ML algorithm. The algorithm then
learns how to recognize normal versus abnormal behavior patterns over time, predicting when the
element/node might be having issues before they occur.

The goal is for networks to adjust and optimize themselves based on real-time traffic flows, congestions,
port failures, high rates of packet drops, configuration changes, software updates, and more without
human intervention; here are some of the most popular AI/ML implementations.

Network automation and AI/ML - It can gain insights through analytics and AI/ML that guide more trusted
automation processes that lower the cost of network operations and provide users with an optimal
connected experience. These technologies help IT automate:

The deployment and management of network policies

The integration of zero-trust security solutions to help ensure network consistency
The identification and classification of devices on the network

Over time, AI will increasingly enable networks to continually learn, self-optimize, and even predict and
rectify service degradations before they occur.

Network Optimization - 5G networks began to roll out in 2019 and are predicted to have more than 1.7 billion
subscribers worldwide – 20% of global connections — by 2025. AI is essential for helping communications
service providers (CSPs) build self-optimizing networks (SONs) to support this growth. These allow operators
to automatically optimize network quality based on traffic information by region and time zone.
Through a network controller and management dashboard, telemetry data from the network can be ingested
and processed through AI/ML engines. These AI engines in the telecom industry use advanced algorithms to
look for patterns within the data, enabling telecoms to both detect and predict network anomalies. As a result
of using AI in telecom, CSPs can proactively fix problems before customers are negatively impacted.

AI and machine reasoning (MR) - Machine reasoning is built from a knowledge base and reasoner (or
reasoning engine) which employs logical techniques such as deduction and induction to generate conclusions.
The knowledge base is fed by various data sources: learnings from previous solutions; codified telecom
domain expertise; product documentation; external data sources; and importantly – insights from all the
machine learning agents in the network.

The reasoner, using the knowledge base, looks at the various predictions, applies logical rules to the
knowledge base, and makes sure that business intent is followed while respecting all service level agreements
and recommending a path towards the goal.

AI/ML and predictive analytics - AI-driven predictive analytics are helping telecoms provide better services
by utilizing data, sophisticated algorithms, and machine learning techniques to predict future results based
on historical data. This means operators can use data-driven insights to monitor the state of equipment and
anticipate failure based on patterns. Implementing AI in telecoms also allows CSPs to proactively fix
problems with communications hardware. In the short term, network automation and intelligence will enable
better root cause analysis and prediction of issues. Long term, these technologies will underpin more
strategic goals, such as creating new customer experiences and dealing efficiently with emerging business
needs.

Use Cases for Network

Common Enterprise Network Optimization &

Network Issues Monitoring Services Security Services Maintenance Services

Wi-FI Authentication Network Traffic SLA Monitoring Network Anomaly & Security Device Performance
Detection Optimization
Incorrect username and Network SLAs are monitored
password through continuous assessment Detect unusual outbound Management of device
of packet latency, jitters, RTT network traffic flow performance degradation
Disabled authentication or and packets dropped leveraging
authorization policy ICMP, TCP, UDP, RTP and HTTP Detect unusual communication Predicting Preventive
protocols with a server Maintenance Service
Unavailability in database
Network Interface Bandwidth Detect unusual outbound email Identify any lags in
LAN Network Fault Utilization activity performance, connectivity
timings
Presence of duplicate IP or MAC Excess bandwidth utiliution of Detect port scanning activity on
addresses network interface leads to a specific host or several hosts
dropped packets and link
Unintented shutdown of port by congestion Detect suspicious login activities
admin through compromised user
Monitoring Dropped Packets credentials or hijack attempts
VlAN misconfiguration
CPU Monitoring Block access to questionable
Failed DOT1x authentication webites and files hosted online
High CPU utilizatlon in network
BGP Flapping devices often leads to dropped Block bandwidth hogging
packets and low network websites (e.g.,streaming or
Excessive update messages are throughput gaming sites)
sent for network reachability in
case of an unstable routing peer Remote Network Monitoring

3
With real AI, the network becomes easier to operate. You can quickly configure, troubleshoot, and protect
your network while minimizing errors and resolving network issues. And when you can fix problems before
they impact users, you are delivering a next-level experience.

AI/ML helps us in detecting network access failures, such as authentication and authorization failures, in
LAN and Wi-Fi systems to determine and remediate the root causes of network access failures. If authen-
tication fails, it may be necessary to check all Syslog data from each network element and then utilize this
Syslog data to identify the exact root cause of this error. Of course, this process can be time-consuming
and may require a great deal of human effort.

Once a root cause is identified, remediation actions can be taken to fix the authentication issue, which
again can require much time and effort by an individual.

Similarly, during the festive seasons like Christmas or the Sales seasons like Thanksgiving, there is always a
spike in traffic that’s much higher than in previous weeks which causes slowness on the Internet due to
Bandwidth Utilization. Hence, AI/ML driven Enterprise strategy helps us in network bandwidth utilization
monitoring as it is essential in finding the root cause of a slow network. It offers a network utilization
monitor for providing comprehensive network traffic and bandwidth analysis based on SNMP monitoring
and the flow of data. Streamlined network bandwidth utilization monitoring allows us to detect, diagnose,
and resolve network performance issues with ease.

It also offers real-time alerts for bandwidth logs, so we can get notified when a device starts to use too
much bandwidth. With customizable notification thresholds, it allows us to set simple or complex condi-
tions to trigger alerts based on simple bandwidth utilization or nested conditions based on the network
topology. Custom alerting also reduces unwanted alerts to the inbox, focusing on important or critical
network issues only.

Methodology
The four-step mechanism is configured to determine the relevant diagnostics given a set of symptoms and
can be configured for automating the process of root cause analysis in a way that can generalize well to
accommodate new types of problems.

1. Data Collection: Collect relevant data from the network and stream the data to a message bus. The
network may be configured to handle a mixture of raw text and quantitative features obtained from
network devices (e.g., L2 switches, wireless controllers, routers, firewalls, authentication servers, and
authorization servers). As such, the network may introduce an ML pipeline that combines Natural Language
Processing (NLP) techniques to extract relevant quantitative metrics, together with supervised learning
models. The first step in this respect may be to rebalance a dataset to properly handle the relative rarity of
the authentication and authorization failures.

4
2. Data Cleaning & Pre-Processing: Syslog messages and datasets, obtained by the data collection
module, may typically be in a complex, unstructured, and noisy format. To get high accuracy and precision,
therefore, a tailored pre-processing step may be used in the present disclosure to make this textual data
more insightful and suitable for model training.

For example, this may be achieved through a combination of techniques or strategies, such as stemming &
lemmatization, OneHot Encoding, CountVectorizer, TF-IDF, Column Transformer, and FastText, which
allows the root-cause analytics systems of the present disclosure to represent unstructured Syslog
messages as fixed-size vectors, which are more suitable for ML algorithms.

3. Model Training & Apply pre-trained ML models: To compute the distance of given symptoms (new
data) to known diagnostics and diagnostic families from the hierarchical knowledge base. The hierarchical
model may combine the regression output of all ML sub-models to predict the most likely root causes. For
example, starting with the prediction of the root sub-model, the models may recursively traverse the
hierarchical model to compute the regression output of each sub-model. By construction, each sub-model
may be increasingly specific. If the accuracy of a sub-model (e.g., internal point or node in the tree) is
below a threshold, the corresponding subtree may be discarded to optimize computational requirements.
If the accuracy of a specific root cause is below another threshold, the system may automatically predict
increasingly generic root causes until the accuracy is sufficient.

5
4. Filter and rank the diagnostics: Hierarchical models may combine the outcome of heuristic or statistical
(non-ML) models. Each node in the hierarchical model is mapped to network automation workflows and
can be triggered for Remediation if the distance calculated by the inference engine is within a certain
trigger threshold.

AI/ML is beneficial in the following:

ML and AI help make enterprise networks more efficient by using data-driven algorithms to identify
patterns within the enterprise infrastructure. For example, ML can be used through anomaly detection—or
looking at changes within your environment over time and determining whether they fit within a normal
pattern or not. Here are eight roles that AI and ML play in network management:

Log analysis - The role of AI and ML in log analysis is simple: it detects, collects, and analyses logs from all
parts of an enterprise environment (e.g., routers, switches, and WAN optimization devices). It then provides
real-time insights into network performance, so you can pinpoint problems faster than ever before.

Logs can be seen as textual data, which means that NLP techniques like Longest Common Substring (LCS),
Bert, or GPT-3 language model can be applied to gather and summarize the same logs in an organized
manner, making it possible to search for specific types of logs.

Advanced analytics - Advanced analytics methods based on machine learning modeling, can make the
computing process smart through intelligent decision-making in a business process. A general structure of a
machine learning-based predictive has a wide range of methods such as regression and classification
analysis, association rule analysis, time-series analysis, behavioral analysis, log analysis, and so on.

One of the most common applications is Anomaly detection which uses time series over Long Short-Term
Memory (LSTM) modeling which is a neural network model. The neural network model is utilized to model a
system log as a natural language sequence. This enables studying log patterns and detecting anomalies
when there is a deviation from the trained model. In addition, workflows are also constructed under the
underlying system log so that developers can analyze the anomaly and perform the necessary RCA for it.

6
Performance monitoring - Nowadays, with software-defined networking (SDN) becoming increasingly
popular, monitoring performance has become steadily more important.

SDN is often used to help automate network tasks and free network administrators from mundane tasks,
like troubleshooting SNMP (simple network management protocol) issues. Now, automated tools can help
monitor traffic flows across an SDN-enabled network.

Security alerts - Automated networking requires an automated security system, which is why many
companies are investing in AI and ML. As a result, organizations can now use these emerging technologies
to automate network security tasks such as malware detection, vulnerability scanning, intrusion prevention
systems (IPS), advanced threat protection (ATP), DDoS mitigation, and more.

With these technologies, enterprises can achieve better uptime while decreasing human error by
leveraging automated networking to create more efficient networks for all users and devices.
Systems can incorporate an assess-centered methodology called STRIDE-AI, an AI/ML component, that
helps in the identification and analysis of threats, Malware prediction, Spoofing, Tampering, DDoS, etc.

Traffic management - Automated tools help manage traffic to optimize performance. With these tools,
information about Internet Protocol (IP) addresses is automatically gathered for analysis. The data is then
integrated with other business or engineering intelligence systems to automate network management tasks.

Intelligent programmable automation controller (IPAC) - A key component to automating networks with AI is
programmable automation controllers (PACs). These devices allow network administrators to automate tasks
using software instead of doing everything manually.

It can be constructed with the help of a Predictive maintenance procedure using AI as follows:

A learning model including a threshold value is generated from current machine data.

The machine is monitored based on the learning model. If the machine status exceeds the threshold
value, a notification is issued.

The machine status is checked. If no error is found, a new threshold value is set.

An error occurs while threshold value setting and monitoring are repeated. Components are replaced.

A new learning model, including the threshold value, is generated based on the previous error level after
components are replaced. Repeating these steps makes status-based maintenance more reliable.

Autonomous scanning and patching - Modern switches employ AI to automate certain tasks, such as
maintenance. These automated tools allow network administrators to shift their focus from a reactive state
to one that is more proactive, allowing them to monitor and identify issues before they become problems.
Some vendors even provide self-healing networks using AI techniques like ML. The goal is to reduce or
eliminate outages by automatically remediating faults as soon as they occur.

7
Automated provisioning - With automated provisioning, you can automate all aspects of network deployment,
from initial configuration to ongoing maintenance. In addition, automated networking complements
continuous service monitoring by measuring performance against SLAs, triggering specific actions based on
threshold breaches or other changes, and proactively alerting administrators. The result is an automated
network that’s always ready for use, even under heavy loads.

The Future of AI in the Telecom Industry

AI in the telecom market is increasingly helping CSPs manage, optimize, and maintain infrastructure and
customer support operations. As tools and applications become more available and sophisticated, the future
of AI/ML in the telecom industry will continue to develop. Employing AI, telecoms can expect to continue
accelerating growth in this highly competitive space.

The below architecture gives us a glimpse of Telecom Data Hub, which provides Network optimization,
predictive maintenance, virtual assistants, RPA, fraud prevention, and new revenue streams are all examples
of telecom AI/ML use cases where the technology has helped deliver added value-for enterprises.

8
ABOUT BRILLIO

At Brillio, our customers are at the heart of everything we do. We were founded on the
philosophy that to be great at something, you need to be unreasonably focused. That’s
why we are relentless about delivering the technology-enabled solutions our customers
need to thrive in today’s digital economy. Simply put, we help our customers accelerate
what matters to their business by leveraging our expertise in agile engineering to bring
human-centric products to market at warp speed. Born in the digital age, we embrace
the four superpowers of technology, enabling our customers to not only improve their
current performance but to rethink their business in entirely new ways. Headquartered
in Silicon Valley, Brillio has exceptional employees worldwide and is trusted by hundreds
of Fortune 2000 organizations across the globe.

https://www.brillio.com/
Contact Us: [email protected]