Sustainable Engineering

Systems and Environment

MANU 1381
Managing Information Technology
Artificial Intelligence
Cloud Computing
1
 Acknowledgement of country
Womin djeka (Welcome)
We would like to acknowledge Wurundjeri people of the
Kulin Nations as the Traditional Owners of the land on
which the University stands. We respectfully recognise
Elders past, present and future.

2
http://kes2021.kesinternational.org/index.php

3
http://kes2020.kesinternational.org/index.php

4
Sessions

MANU1381 5
6
7
8
9
10
11
 12
https://www.businessinsider.com.au/5g-help-retailers-shopping-augmented-reality-2019-8?r=US&IR=T
13
 How to Walk to Save Joints

Photos courtesy http://www.businessinsider.com/this-bionic-exoskeleton-can-give-you-superhuman-strength-2013-7/?r=AU&IR=T

14
15
16
17
 What is Intelligence
• A property of mind that include many related abilities:
• Reasoning, Planning,
• Problems solving, Abstract thinking,
• Comprehend ideas, Language, Learning…
• Intelligence also include qualities as:
• Creativity, Personality, Character, Knowledge...
• How could we recognise whether something inhuman
have intelligence

18
 Intellegere
• Latin verb "intellegere", means "to understand“
• Deeper understanding of the relationships of the
world around us
• Capability for metaphysical manipulation of
abstract objects and relationships once their
understanding is achieved.

19
 Artificial Intelligence
• What type of mater and organization is
necessary? C, Si, Ge
• Is it possible for a machine made of other
material than C to have intelligence comparable
to our?
• Could non-organic machine had problem-
solving capabilities like a human's
• Could a machine have consciousness and
emotions? 20
 https://www.britannica.com/science/periodic-table
What type of mater and organization is
necessary? C, Si, Ge

Carbon, silicon and germanium

have four valence electrons each.

21
 Electron Energy Bands

Conductor
Electron Energy

Band

Valence
Band

Metal Semiconductor Insulator

22
What is Artificial Intelligence - AI
• Study of human intelligence so that it could be
artificially simulated, or generated,
• The search for a way to map intelligence into
machine, hardware and software, and enable such
system to formalize thinking,
• A branch of computer science, dealing with machine
intelligent behaviour,
• The implementation and study of systems that
show an autonomous, intelligent, behaviour…
23
 Artificial Intelligence
Artificial Intelligence – imitating humans

Machine Learning / computer learns from data:

training a model using datasets

Deep Learning: machine learning using network of

neurons

MANU1381 24
What is a Neural Network (NN)?
• A neural network is a massively parallel distributed multiprocessing machine
that has a natural propensity for:
• storing experimental knowledge, and
• making it available for use.
• It is similar to the brain in two ways:
• Knowledge is acquired by the network through a learning process.
• Knowledge is stored using interneuron connections strengths known as
synaptic weights.

Biological vs
Artificial NN
https://www.quora.com/What-
is-the-differences-between-
artificial-neural-network-
computer-science-and-
biological-neural-network)
A Few Neurons Together
• Axon of one neuron almost touches
dendrites of another neuron
• The small gap in between is the
Synapses.
• Synapses can impose excitation
(active) or inhibition (inactive) on the
receptive neuron.
Looking at the Neuron on the Right
1. Input signals come from other
neighbouring neurons.

Input 1

Input 2

Input 3
Looking at the Neuron on the Right
2. The synapses impose excitation or
inhibition of the signal. This can be
thought of as multiplying with a weight
of 0 or 1.
Input 1 W1
W2 x
Input 2 x
W3
x

Input 3
Looking at the Neuron on the Right
3. The cell body sums the incoming signals
temporally).

∑
Looking at the Neuron on the Right
4. When sufficient input is received
(more than threshold), the neuron fires,
i.e. generate a spike, which is
transmitted to the axon.

∑
∑≥
threshold
Looking at the Neuron on the Right
5. If input is less than threshold, then no
firing occurs.

∑
∑<
threshold
 Early Model of Neuron
Very similar to biological neuron.

Threshold function:
1 if weighted sum > T
x1 0 if weighted sum < T
wk1
x2 wk2
T
x3 wk3
Inputs Output yk
wk4
Neuron
x4 wkm no. k
…

xm
 Improved Model of Neuron
Having threshold function not symmetrical about 0 makes some calculations
difficult. Therefore, a bias is added.
Threshold function:
1 if (weighted sum + bias) > 0
x1 0 if (weighted sum + bias) < 0
wk1 bk (bias)
x2 wk2
vk 0
x3 wk3
Inputs Output yk
wk4
Neuron
x4 wkm no. k
…

xm
 Further Improvements
Various activation functions were also proposed in place of the threshold
function.

x1
wk1 bk (bias)
x2 wk2 Activation
Function
wk3 vk φ(•)
Inputs x3
Output yk
wk4
Neuron
x4 wkm no. k
…

xm
 Bias as Input x Weight
With the last formulation, the model of a neuron looks as follows:

Fixed input x0=1

Wk0=bk (bias)
x1 wk1 Activation
Function
wk2 vk φ(•)
x2
Output yk
Inputs wk3
Neuron
x3 wkm no. k
…

xm
 Neural Networks (NNs)
• Artificial NN consist of layers of
– Linear and
– Nonlinear
Algebraic functions that transform a given input
xi(i=1, 2,…,n) to an output.
• During the neuron, k, learning stage, the parameters of the
linear functions, called weights, w, are defined by processing
input sets and producing output sets

𝑦 𝐴𝑐𝑡𝑖𝑣𝑎𝑡𝑖𝑜𝑛 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑤 𝑥

• Parameters of the nonlinear functions are chosen based on

practical considerations. 36
Activation Functions
• Various activation functions have been proposed for neural networks.
• Threshold function (hard-limiter):
• Note: McCulloch-Pits model (1943) of neuron used this form of function.
φ(v)

1
v
0
• Piecewise linear function:
• Linear combiner within certain range, then saturated to 0 or 1.
• If gradient of linear region is very high, it would reduce to threshold
function.
φ(v)

1
v
0
Activation Functions
1. Gaussian function:
• Continuous & Differentiable everywhere
• Single maximum at v = µ.
• Used in Radial Basis Function networks.

φ(v)
1 1 1 𝑣 𝜇
𝜑 𝑣 exp
2𝜋𝜎 2 𝜎
v
µ

2. Linear:
𝜑 𝑣 𝑣
• This is mostly used in the output layer of a multilayer network (later).
 Example
• Solution:
vj = (1 x -10) + (10 x 0.7)
+ (-20 x 0.1) + (6 x 1)
+ (-2 x -0.4)
= 1.8
Fixed input x0=1
Wj0= -10
10 0.7 Activation
Function
0.1 vj φ(•)
-20
Output yk
Inputs 1
Neuron
6 -0.4 no. j
…

Case 1, Linear: Output = 1.8

-2
Case 2: Hard Lim: Output = 1
Exercise x0
1
x1
2
x2
2
x3
1
x4
2

Wk0 Wk1 Wk2 wk3 wk4

0.5 0.8 0.1 ‐0.1 ‐0.5

Activation
Function
Output yk

Case 1, Linear: Output = ??

Case 2: Hard Lim: Output = ??

Linear Hard-Limiter
Solution x0
1
x1
2
x2
2
x3
1
x4
2

Wk0 Wk1 Wk2 wk3 wk4

0.5 0.8 0.1 ‐0.1 ‐0.5

Activation
Function
Output yk

Case 1, Linear: Output = 1.2

Case 2: Hard Lim: Output = 1

Linear Hard-Limiter
Network Architectures
• One single neuron is only able to solve some very simple problems.
• To solve for more complex problems, networks with large number of neurons
are required.
• Multilayer feedforward neural networks: Connections only from left to right.

Hidden Layer 2

Output Layer
…

Hidden Layer 1
Input Layer
Network Architectures
• The number of hidden layer is a design parameter. It can be 0, 1, 2…
• The network on the page before has 2 hidden layers.
• On this page, the example below has no hidden layers.
• It is called “Single-layer Feedforward Network”. The layer refers to output
layer.

Output Layer
Input Layer
 Deep Neural Networks
• The state-of-the-art method for object recognition is the application of Deep
Neural Networks.
• It gives very good accuracy and has won many competitions / challenges in
object recognition.
• Why is it called deep?
• Shallow networks, e.g. MLP’s usually have only a few layers. Adding more
layers to MLP’s does not increase its capability to approximate functions.
• Deep neural networks, on the other hand, has many layers (can be
several tens or hundreds) where each layer has its own specific task, e.g.
in convolutional neural network:
• First layer finds the edges within an image
• Second layer combines the edges to get corners etc.
• Third layer combines the corners (etc.) to get nose, eyes etc.
• Fourth layer finally recognises that it is a face.
• Multilayer Perceptron (MLP) is a class of feedforward Artificial Neural
Network (ANN)
 A Simplified Neural Network
Structure for Image Classification

45
A Simplified Neural Network Structure
for Traffic State Prediction

46
 Conventional AI
• This approach is distinguished by formalism,
statistical analysis, definitions and proof.
• Machine learning is primarily associated with
conventional AI.
• It includes
– Expert systems,
– Case based reasoning,
– Semantics.
47
 Computational Intelligence - CI
• CI is known for its informal, non-statistical and trial-and-error
approach.
• Learning is generally an iterative process of improvement
based on empirical data.
CI categories:
– Neural networks,
– Fuzzy systems,
– Evolutionary computation.
• Research overlaps with a-life, cognitive science, cybernetics &
robotics, hybrid intelligent systems design and study.

48
 AI Application in Computer Vision
Hardware: high-speed processors, graphical
processing units (GPU)s, in the cloud and on the edge.
Large, already-available datasets, used to train NNs:
• YouTube-8M (Google), Trecvid, UCF-101, Kinetics
MathWorks Neural Network Toolbox™ and Python
Algorithms: YOLO, Single Shot Detector (SSD),
Faster R-CNN, Histogram of Oriented Gradients
(HOG) , YOLOv5, released in 2020, which has
balance between real-time performance and detection49
accuracy.
 Speed up Development
• Application of support packages like AlexNet
• Transfer Learning is used for the retraining of
predesigned CNN for using our own images, or
videos
• Transfer Learning in AI is the equivalent process
of Requalification or Retraining of someone to
perform other tasks in the workplace

50
 Artificial Intelligence Topics
• Planning, Decision making and Problem Solving
• Uncertainty, Probability, Games,
• Deduction in Logic, Fuzzy Logic
• Machine Learning,
• Knowledge Engineering: Databases, Expert Systems,
Diagnostic Systems
• Web, Google, Learning Hub, Online Exam?,
• Natural Language : Markov models…
• Artificial Life
• Neural Networks
51
 Neural Networks (NNs)
• Non-recurrent NNs
– No feedback
• Recurrent NNs
– Feedback

52
 AI Areas
• Machine Vision:
• Interpreting visual data. Face recognition…
• Speech Recognition:
• Voice communication / Siri,…
• Text to Speech (ACROBAT Reader), OCR
• Advanced Logics: Advanced logic systems.
• Reinforcement Learning
• Robotics: Software and hardware. Projects. 53
 Algorithms and Circuits
• Boolean Intelligence
• Everything can be express using binary Alphabet
B={0,1} => Binary brain can be built
• 1-Bit Learner
• N-Bit Learning Circuit
• Circuits encoded in Java (Neuron),
• Other Object Oriented languages
54
 Intelligent Robots
• If we could create robots with intelligence
comparable to ours, should we do that?
• Many ethical questions may arouse
• Machine that can learn
• Machine that can reproduce itself
• Can such a machine be more intelligent that
us???
55
 Garry Kasparov World Champion
vs. Deep Blue, IBM
• Garry Kasparov playing chess
against computer Deep Blue.
• Deep Blue was the first
machine to win a chess match
against a world champion.
• Photo courtesy of IBM.
• 1996: Garry Deep Blue 4:2
• 1997: Garry Deep Blue 2.5:3.5

http://www.thetech.org/robotics/universal/breakout_p11_ibm.html

56
 Research
• Software Development for Autonomous and
Social Robotics Systems
• https://link.springer.com/chapter/10.1007/978-
3-319-92231-7_16

MANU1381 57
 Cloud Computing Stack
Client
User Interface Machine Interface
Application
Components Services
Platform
Processing Networking Storage
Infrastructure
Server
58
 Level of Service Offerings
• Software as a Service (SaaS)
– Offered as a hosted set of software (running on a platform and
infrastructure), that customer do not own but pay for utilization
on consumption basis.
• Platform as a Service (PaaS)
– Provided hardware plus a certain amount of application
software, as integration into a common set of programming
functions, or databases. It is a foundation which can be used to
build own applications. PaaS is an application development and
deployment platform delivered as a service to developers over
the Web.
• Infrastructure as a service (IaaS)
59
 Infrastructure as a Service
• Offering: Compute power, storage, and networking
infrastructure.
• Unit of deployment: Virtual Machine Image
• Pricing structure:
– Compute usage per hour, data transfer in/out per GB, IO
requests per million, storage per GB, data transfer in/out to
storage per GB, data storage requests per thousand.
– All charges per billing period.
• Customer: Software owner that would like an
application hosted on the Internet for their end users.
60
 CISCO ICT Example
• Networking, Internetworking hardware and
software, Internet and CISCO Routers
• Deliver a Universal Curriculum to Thousands of
Students Worldwide
• 1997 problem: Customers were buying networking
equipment but there were no of qualified, skilled
ICT personnel who could install, support, and
maintain networks
 2012, Dr Milan Simic MANU1381 61
 CISCO Networking Academy
• ICT education program
• Managing worldwide growth while
• Maintaining the high level of quality
• Migrating to a cloud computing model

MANU1381 62
 CISCO Cloud Computing
• Cloud is a term for the delivery of computing as
a service rather than a product
• Shared resources, software, and information are
provided as utility over the network
• Provide storage and processing power to users
• Networking Academy utilize cloud computing
to provide streamlined method for deploying
ICT education curriculum
MANU1381 63
 Cloud and Hybrid Collaboration
• Hybrid services combine cloud-based software as
a service with premises-based software and with
hosted services (Cisco HCS).
https://www.cisco.com/c/en_au/solutions/collaboration/cloud-collaboration/index.html

• Cisco Unified Communications Manager Cloud

https://blogs.cisco.com/collaboration/finally-a-clear-path-to-the-cloud-that-will-accelerate-
workforce-transformation

64
 Edge Computing
• A distributed computing and data storage topology
• Information processing is located close to the edge,
locally, not on the cloud
• Close to the source and the use of information
• Computing does no rely on central cloud location
• Important for real time data processing
• Edge computing is great support for IoT

65
What is edge computing and why does it matter? | Network World
 Edge Computing

Cloud

Edge Node Internet of

Corporate Data
Local Processing Things
Centre Devices
And storage

66
Systems, Applications & Products
in Data Processing (SAP)
• European multinational software corporation
• SAP is an enterprise software to manage
business operations and customer relations
• SAP is a common system for multiple tasks

MANU1381 67
 Data Storage Architecture
• SAP is Founded June 1972
• Punch cards replaced with Common Database
• Based on mainframe, Centralised
• Mid 1990 shift to client/server architecture
• 2004, 2006… Internet based
• 2014 Cloud based
• 2020 and later SAP IoT, AI, Machine learning…

MANU1381 68
SAP Enterprise Resource
Planning (ERP) Modules

https://solutiondots.com/blog/sap-erp-modules/
MANU1381 69
 Functional SAP ERP Modules
1. Human Resource Management (SAP HRM), also known as Human
Resource (HR)
2. Production Planning (SAP PP)
3. Material Management (SAP MM)
4. Financial Supply Chain Management (SAP FSCM)
5. Sales and Distribution (SAP SD)
6. Project System (SAP PS)
7. Financial Accounting and Controlling (SAP FICO)
8. Plant Maintenance (SAP PM)
9. Quality Management (SAP QM)

https://solutiondots.com/blog/sap-erp-modules/
MANU1381 70
 Technical SAP ERP Modules
• SAP Basis
• SAP Security
• Advanced Business Application Programming (SAP ABAP)
• High Performance Analytic Appliance (SAP HANA)
• SAP NetWeaver
• Information Systems Management (SAP IS)
• Customer Relationship Management (SAP CRM Technical module)
• Exchange Infrastructure (SAP XI)
• SAP Solution Manager

https://solutiondots.com/blog/sap-erp-modules/
MANU1381 71
SAP Business Suite Applications
1. Enterprise Resource Planning SAP (ERP SAP)
2. SAP Customer Relationship Management (SAP CRM)
3. SAP Supply Chain Management (SAP SCM)
4. SAP Supplier Relationship Management (SAP SRM)
5. SAP Product Life cycle
6. SAP HANA (High Performance Analytic Appliance)
7. Government, Risk and Compliance (SAP GRC)
8. SAP Strategic Enterprise Management (SAP SEM)
9. SAP Compliance Management for SOA

https://solutiondots.com/blog/sap-erp-modules/
MANU1381 72
74
Technology acceptance framework

75
KES Journal
• International Journal of Knowledge-Based and
Intelligent Engineering Systems
• http://www.kesinternational.org/journal/
• http://www.kesinternational.org/journal/topics.php
• Emerging Intelligent Technologies:
• Artificial Intelligence and the Internet, Information Agents on the Internet,
E-commerce/E-business and E-learning, Intelligent Information Retrieval,
Intelligent Web Mining & Applications, Intelligent User Interfaces,
Bioinformatics using Intelligent & Machine Learning Techniques, Intelligent
Tutoring Systems, Virtual Reality & Multi-Media Intelligent Information
Systems, Artificial Life, Ant and Swarm Algorithms.

MANU1381 76
Thank you,
Questions

MIET1381 77