Recent Trends in Research - Special Reference to Data Science, Machine Learning and Explainable

Artificial Intelligence

A Session for Academicians and Professionals

Bhagirath Kumar Lader

Chief Manager (Business Information System)
GAIL (India) Limited
https://www.linkedin.com/in/bhagirathl

ai79bh@gmail.com August 20, 2021

§ Artificial Intelligence
§ Machine Learning
§ Deep Learning
§ AI Use Cases
§ Responsible AI
§ Explainable AI
§ AI Research Trends
§ Conclusion
 Disclaimer
• The information shared during the presentation is taken from public sources and some materials
may be copyrighted by respective authors/publishers. Accordingly, information is being shared for
educational purpose only and not for any commercial purpose.

• The views and opinions expressed during the session are my own and do not reflect the views or
opinions of my employer GAIL (India) Limited in any manner, whatsoever.

• All reference to data or datasets are public in nature and no confidential information related to any
entity will be shared during the session.
 Artificial Intelligence

A Primer for Academicians & Professionals

 Artificial Intelligence

Basic Ideas

5 |
 Cognitive Ability reasoning

learning from problem

experience solving

Cognitive ability is defined as a

general mental capability involving

Intelligence is the measure of

cognitive capabilities
complex idea
planning
comprehension

abstract
thinking

6 |
 • Prominent philosophers such as
• Aristotle
• St. Thomas Aquinas
• William of Ockham

Artificial • René Descartes

• Thomas Hobbes, and
Intelligence • Gottfried W. Leibniz
have asked the questions:

ü What are basic cognitive operations?

ü What necessary conditions should a (formal) language fulfill in order to be an

adequate tool for describing the world in a precise and unambiguous way?

ü Can reasoning be automatized?

ü Is it possible to construct an artificial intelligence system?

7 |
 Artificial Intelligence
• Artificial Intelligence or Machine Intelligence is a set of related
technologies that seems to emulate human thinking and action in
machines or agents

• AI is about idea of building machines or agents which are capable of

thinking and acting like humans

• AI systems:
• learn from experience i.e. data / feedback
• arrive at their own conclusions
• appear to understand complex real-world case or scenario
• participate in natural-language dialogues with people
• have cognitive capabilities i.e. learning and problem solving

8 |
 Some Definitions of AI – Four Categories

Computational Intelligence
is the study of the design of
intelligence agents.

(Poole et. al.)

Artificial Intelligence – A Modern Approach

Russel & Norvig

9 |
 Acting Humanly – The
Turing Test Approach • Turing (1950): Computing Machinery and
Intelligence
• Operational definition of intelligence
• Operation test for AI
• Can Machines Behave Intelligently?

The Imitation Game

• Capabilities for passing (Total) Turing Test?
• Natural Language Processing
• Knowledge Representation
• Automated Reasoning
• Machine Learning
• Computer Vision
• Robotics
10 |
 Enigma
Video

https://www.youtube.com/watch?v=j2jRs4EAvWM
11 |
 Rationality

A system is rational if it
does the “right thing,”
given what it knows.

12 |
 Rational Agents

• Abstractly, an AI agent is a function from percept

histories to actions:

𝑓 ∶ 𝑃∗ → 𝐴
Artificial
Intelligence is
• Percept history is the history of all that an agent has
perceived till date. about
• The agent function is based on the condition-action rule. A
designing
condition-action rule is a rule that maps a state i.e. Rational Agents
condition to an action. If the condition is true, then the
action is taken, else not.
• For any given class of environments and tasks, we seek the
agent (or class of agents) with the best performance.
Goals of Artificial Intelligence

The Goals of AI is driven by two groups:

Industry, whether related to goods or services, is concerned primarily with creating
expert systems
• Expert Systems should demonstrate intelligent behavior, regardless of their
resemblance or non-resemblance to human intelligence.
• The systems which exhibit intelligent behavior, learn, demonstrate, explain, and
advice its users.

Academia of Cognitive Science, rising out of psychology, linguistics, philosophy,

biology, social sciences and computer science which is interested in AI for its
ability to:
§ Model Human Intelligence with aim to some day replicating or surpassing
human level intelligence.
§ Creating systems that understand, think, learn, and behave like humans.

14 | https://www.cs.swarthmore.edu/~eroberts/cs91/projects/ethics-of-ai/sec2.html
 History of
Artificial YEAR MILESTONE / INNOVATION

Intelligence 1923 Karel Čapek play named “Rossum's Universal Robots” (RUR)
opens in London, first use of the word "robot" in English.
• Source:
TutorialPoint
1943 Foundations for neural networks laid by McCulloch and Pitts

1945 Isaac Asimov, a Columbia University alumni, coined the

term Robotics.

Alan Turing introduced Turing Test for evaluation of

1950 intelligence and published Computing Machinery and
Intelligence. Claude Shannon published Detailed Analysis of
Chess Playing as a search.

The first AI system, called Logic Theorist, was designed by

1955 Allen Newell and Herbert Simon and implemented by J.
15 | Clifford Shaw at CMU
 Year Milestone / Innovation

History of Artificial John McCarthy coined the term Artificial

Intelligence 1956
Intelligence.

1958
John McCarthy invents LISP programming
• Source: language for AI.
TutorialPoint
Computers can understand natural language well
1964
enough to solve algebra word problems correctly.

1965
Joseph Weizenbaum at MIT built ELIZA, the first
chatbot
Shakey, a robot, by Stanford Scientists equipped
1969 with locomotion, perception, and problem
solving.
Edinburgh University built Freddy, the Famous
1973 Scottish Robot, capable of using vision to locate
and assemble models.

16 |
 Year Milestone / Innovation

The first computer-controlled autonomous

1979
vehicle: Stanford Cart

History of AI 1985
Harold Cohen created and demonstrated the
drawing program, Aaron.
• Source:
TutorialPoint
Major advances in all areas of AI −

• Case-based reasoning
• Multi-agent planning
1990 • Scheduling
• Data mining, Web Crawler
• natural language understanding and
translation
• Vision, Virtual Reality
• Games

17 |
 History of Artificial • Source:

Intelligence TutorialPoint and Nature Journal

Year Milestone / Innovation

The Deep Blue Chess Program beats the then
1997 world chess champion, Garry Kasparov.
MIT displays Kismet, a robot with a face that
expresses emotions. The robot Nomad explores
2000 remote regions of Antarctica and locates
meteorites.

2011 IBM Watson AI beats human in Jeopardy

2016 DeepMind’s AlphaGo AI beats human in Go

Facebook and CMU’s AI poker bot is first to beat
2019 professionals at multiplayer game
DeepMind’s AI makes gigantic leap in solving
18 | 2020
protein structures
 Symbolic Artificial Intelligence View
• In computer science, a symbolic
language is a language that
uses characters or symbols to
• Mind is something akin to a digital computer processing a represent concepts, such
symbolic language.
as mathematical operations and the
entities (or operands) on which
• The field of AI, since its inception, has been conceived these operations are performed.
mainly as the development of models using symbol
manipulation. • Modern programming languages use
• The computation in such models is based on explicit symbols to represent concepts
representations that contain symbols organized in some and/or data and are therefore,
specific ways
examples of symbolic languages.
• The aggregate information are constructed from
constituent symbols and syntactic combinations of these
symbols.

Also called Good Old-Fashioned AI (GOFAI)

A physical symbol system has the necessary and sufficient means for general intelligent action.

-Newell and Simon, 1976

19 |
 Connectionists View

Inspired by A large number of Due to their

human brain simple and massively parallel
uniform processing nature, such Edward Thorndike’s (1874-1949) work on
elements models are good at animal behaviour and the learning process
(neurons) flexible and robust led to the theory of connectionism, which
interconnected processing states that behavioural responses to
with extensive
specific stimuli are established through a
links
process of trial and error that affects neural
connections between the stimuli and the
Image Credit: http://www.mysearch.org.uk/website1/html/106.Connectionist.html most satisfying responses.
20 |
 Weak AI or Narrow AI

• AI system that is designed and

trained for a particular task which it
can do at par or better than a
human.

• Virtual personal assistants, such as

Apple's Siri or Google Assistant, are
a form of weak AI.

21 |
 Strong Artificial
Intelligence
• Strong Artificial Intelligence is a theoretical
form of machine intelligence that is equal to
human intelligence.
• In some sense, strong AI would share same
characteristics as general AI.

Artificial General Intelligence (AGI)

An artificial intelligence reaches the general state

when it can perform any intellectual task with the
same accuracy level as a human would. The term was
introduced around 1998 and came in popular
discussion after 2002.

22 |
 AI Technological Representations

Machine Robotics Search

Learning

Fuzzy Logic Expert Systems NLP

23 |
 Machine Learning

A Primer for Academicians & Professionals

24 |
 AI

ML

DL

Hierarchy
25 |
 AI

ML

DL

Hierarchy

26 |
Machine • Machine learning is an application of artificial intelligence (AI) that provides
systems the ability to automatically learn and improve from experience

Learning without being explicitly programmed. Machine learning focuses on the

development of computer programs that can access data and use it learn
for themselves.
• The term was pioneered by Arthur Samuel of IBM in 1959.

Machine learning is the training of a model from data that generalizes a

A computer program is •said to learn from experience E with respect to some class of
decision against a performance measure.
tasks T and performance measure P if its performance at tasks in T, as measured
by P, improves with experience E.
(Tom M Mitchell)

27 |
 Machine Learning Terminologies

1. Dataset
2. Records 8. Loss Function
3. Features
9. Optimization Method
4. Labels
10.Regularization
5. Training Set
6. Validation Set 11.Hypothesis Set
7. Test Set
28 |
 Types of Machine Learning

Supervised Unsupervised Reinforcement Transfer

Learning Learning Learning Learning

29 |
 Supervised Learning
Algorithms that are designed to learn by examples.

• It is like having a teacher supervise the whole process.

• Training data will consist of inputs paired with the correct outputs
(data points with corresponding labels).

• Supervised learning classified into two categories of algorithms:

• Classification: A classification problem is when the output

variable is a category, such as “Red” or “Blue” or “spam” and
“not spam”.

• Prediction/Regression: A regression problem is when the

output variable is a real value, such as “Rupees” or “weight”.
30 |
Random Forest
Support Vector Machine
Naïve Bayes Linear Regression
Classifier

Some Supervised Learning

Algorithms

K-Nearest Neighbour

Logistic
Regressio
31 | n
 • Unsupervised Learning is a class of Machine Learning algorithms to find

Unsupervised Learning
the patterns in data without any label.

• The most common unsupervised learning method is cluster analysis,

which is used for exploratory data analysis to find hidden patterns or
grouping in data.

32 |
Unsupervised Learning – Common Algorithms

Principal
K-Means Clustering Component
Analysis

33 | Apriori Algorithm for Association Generative Adversarial Networks

 Reinforcement Learning

• Reinforcement learning is about taking

suitable action to maximize the
expected reward in a particular
situation
• The reinforcement agent decides what
to do to perform the given task
• In the absence of training dataset, it is
bound to learn from its experience

34 |
 Deep Learning

A Primer for Academicians & Professionals

35 |
 Connectionism
• Connectionism is an approach in the fields of cognitive
science that hopes to explain mental phenomena
using artificial neural networks.

• Learning occurs by modifying connection strengths

based on experience and behavioral associations
(connections) could be predicted by application of the
two laws.

• The law of effect stated that those behavioral

responses that were most closely followed by a
satisfying result were most likely to become
established patterns and to occur again in response to
the same stimulus.

• The law of exercise: frequent connections of stimulus

|and response establishes stronger behavior
36
Experiment by Edward Thorndike
 Deep Learning Breakthroughs Since 2012

-Deep Neural Networks for Speech -Invention of GANs (Ian Goodfellow) – -TensorFlow released
Recognition (Hinton et al. in 2012) power of imagination and creativity -FaceNet paper by Google
-Google’s Brain Recognizes Cat Videos -VGGNet -Microsoft Asia ResNet (Deep Residual
-AlexNet: 2012 ImageNet Large Scale Learning for Image Recognition)
Visual Recognition Challenge (ILSVRC) -OpenAI
-YOLO Paper

2012 2014 2015

37 |
 Deep Learning Breakthroughs Since 2016

-DeepMind’s AlphaGo beat world champion Lee

Sedol at Go four out of five times.
-TPU was released
-Sophia by Hanson Robotics
-PyTorch was released
-Google Assistant
2017

2016

-AlphaZero
-Transformer Architecture - Attention Is All You Need

38 |
 Deep Learning Breakthroughs Since 2018

2018

-Deepfakes
-BERT - Bidirectional Encoder
Representations from Transformers

-Solving Rubik’s Cube with Robot Hand – OpenAI

Dactyl
-Deepfakes Detection Challenge
-TensorFlow 2.0 released
-OpenAI Five world champions in an esports
game Dota2

2019

39 |
 Deep Learning
Breakthroughs in
2020

40 |
 Deep Deep learning is part of a broader family of machine learning
methods based on artificial neural networks with representation

Learning learning. Learning can be supervised, semi-supervised or

unsupervised.

A Graph with depth more than 2

41 |
Deep Learning – Common Algorithms

42 |
Convolutional Neural Network for Computer Vision Tasks like Image Recognition
 Deep Learning – Common Algorithms

Recurrent Neural Network for Sequence Tasks like NLP

43 |
 Deep Learning – Common Algorithms

Generative Pre-trained
Transformer 3 is an
autoregressive language
model that uses deep
learning to produce
human-like text. It is the
third-generation language
prediction model in the
GPT-n series created by
OpenAI.

44 | GPT3 Demo
 Reinforcement Learning

Explanation:
https://www.youtube.com/watch?v=ECmG0nNJE98

Experience:
https://www.youtube.com/watch?v=tlThdr3O5Qo

Video Link- Tesla Auto Pilot

45 |
 Leading Cloud AI Platforms

Oracle AI Wipro HOLMES

Salesforce API.AI
Baidu Premonition
Pega Rainbird
Infosys Nia
46 |
 TensorFlow Playground

https://playground.tensorflow.org/

47 |
Real Machine
Learning
Examples

48 |
https://github.com/bhagi8289/RMachineLearning
 Artificial Intelligence

AI’s Projected Impact on Industries

49 |
 Industries Impacted by AI

Manufacturing Healthcare

Retail Legal Services

Logistics Advertising

Aviation
BFSI

50 |
 Industries Impacted by AI

Chemicals Online Shopping

Education Sports

Agriculture Media & Entertainment

Software Development &

Cybersecurity Call Centers

51 |
 Industries Impacted by AI

Energy, Utilities & Mining Customer Experience

Hospitality Fashion

Intellectual Property Automotive

IT Services Management Fortune Telling

52 |
 Industries Impacted by AI

Gaming Materials Discovery

Casino Defense

Pharma & Drug Discovery Music Industry

Law Enforcement, Security &

Job Search & Recruitment
Surveillance

53 |
 Artificial Intelligence

AI Adoption & Use-Cases

54 |
 AI Application Areas

Continuo
Classificat us Optimizat Machine
Clustering
ion Estimatio ion Vision
n

Natural Anomaly Recommen Data

Ranking
Language Detection dation Generation

55 |
 Sophia The Robot Video

https://www.youtube.com/watch?v=Bg_tJvCA8zw

https://www.youtube.com/watch?v=G-zyTlZQYpE

56 |
 Notable AI Use-
Cases

57 |
 Some Real AI Use-Cases

S.N. Company Published Use-Case

1 Google Maximizing The Potential Of Artificial Intelligence
• Artificial Intelligence Personal Assistants
• Language Translation
• Self-Driving Cars
• Captioning Millions Of Videos
• Diagnosing Disease
• Google Brain
• Deep Mind

58 | Credit: Artificial Intelligence in Practice by Bernard Marr & Matt Ward, 2019, Wiley
 Select Real AI Use-Cases
S.N. Company Published Use-Case
2 The Walt Disney Using Artificial Intelligence To Make Magical Memories
Company In 2013, Disney introduced its MagicBand wristbands, which are
issued to every visitor and let them book rides and attractions,
access their hotel rooms, order meals at the park's restaurants and
pay for purchases in gift shops.
They also give Disney detailed information about what each visitor is
doing at every point in the day. This lets them offer personalized
experiences . It also gives park planners detailed aggregated
datasets about overall visitor movements.
This means that planners can repurpose spots that fail to attract
footfall to ease congestion around the top attractions and hotspots
that cause bottlenecks throughout the day.
Because data is analyzed in real time, response can be real time too
– for example, staging an impromptu character parade to draw
crowds from a heavily congested area to a quieter one.

59 | Credit: Artificial Intelligence in Practice by Bernard Marr & Matt Ward, 2019, Wiley
 Select Real AI Use-Cases
S.N. Company Published Use-Case
3 BMW Using Artificial Intelligence To Build And Drive The Cars Of
Tomorrow

Back in 2016, a partnership with IBM saw four BMW i8 vehicles

connected to the IBM Watson cognitive computing platform through
their IBM cloud service. The idea was that the car could learn how
to improve its understanding of driver behavior and then adapt its
system to suit personal preferences. By uploading all the data it
gathers to the cloud, the system is able to build a vast database of
user behaviors, and then use machine learning to anticipate the
needs and preferences of other drivers.

60 | Credit: Artificial Intelligence in Practice by Bernard Marr & Matt Ward, 2019, Wiley
 Responsible Artificial
Intelligence

An Overview
61 |
 Responsible AI Practices

The development of AI is creating new opportunities to solve challenging, real-world

problems. It is also raising new questions about the best way to build AI systems that benefit
everyone.

62 |
 Fairness Reliability &
Safety

Microsoft
Privacy &
Security Inclusiveness Responsible AI
Framework
Transparency Accountability

63 |
Google Responsible AI Framework

64 |
 Responsible AI
Document by
NITI AAYOG

65 |
 Explainable Artificial
Intelligence

(XAI)

An Overview
66 |
 Explainable AI

• Explainable AI is a set of tools and

frameworks to help you understand
and interpret predictions made by
your machine learning models.

• With it, you can debug and improve

model performance, and help others
understand your models' behavior.

• XAI helps build interpretable and

inclusive AI systems from the ground
up with tools designed to help detect
and resolve bias, drift, and other gaps
in data and models.

67 |
 Why Explainable AI

The Explainable AI (XAI) program aims to create a

suite of machine learning techniques that:

• Produce more explainable models, while

maintaining a high level of learning performance
(prediction accuracy); and

• Enable human users to understand, appropriately

trust, and effectively manage the emerging
generation of artificially intelligent systems.

68 |
Courtesy: Dr. Matt Turek, DARPA
 Billions to
Trillions of
Parameters

69 |
 Need for Explainable AI

Applications
• We are entering a new age of AI • Medicine
systems and AI applications. Healthcare • Why did you do it?
• Machine Learning is the core • Security • Why not something else?
technology Finance • When do you succeed?
• Machine Learning models are • Legal • When do you fail?
• Opaque • Human Resources • When can I trust you?
• Non-intuitive • Military • How do I correct the error?
• Difficult for people to • Supply Chain
understand
70 |
 XAI Concept

71 |
https://www.darpa.mil/program/explainable-artificial-intelligence
 XAI Goals

• New machine-learning systems will have the ability to explain

their rationale, characterize their strengths and weaknesses, and
convey an understanding of how they will behave in the future.
• The strategy for achieving that goal is to develop new or
modified machine-learning techniques that will produce more
explainable models.
• These models will be combined with state-of-the-art human-
computer interface techniques capable of translating models
into understandable and useful explanation dialogues for the
end user

72 |
https://www.darpa.mil/program/explainable-artificial-intelligence
 More Technical Details for Explainable AI

73 |
 More Details on Taxonomy for Explainable AI

74 |
https://www.sciencedirect.com/science/article/pii/S1566253519308103
 Artificial Intelligence
Research Trends

Brief Overview

75 |
 Hot Trends in AI Research
• Machine Learning • Image processing
• NLP – Large Language Models • Unsupervised Learning
• Computer Vision – Deep Learning • Neural networks
Powered Models • Data Mining
• Transformers • Data Analytics
• GANs and Deep-Fake • Neuro-Symbolic AI
• Adversarial Machine Learning • Explainable and Responsible AI
• Robotics • Human-Centred AI
• Evolutionary computation • Human Bio-metrics and Pattern
• Speech Recognition Recognition
• AI Powered Music Generation • Autonomous Vehicles
• AI Artworks

76 |
 AI Research Trends

77 |
https://ai100.stanford.edu/2016-report/section-i-what-artificial-intelligence/ai-research-trends
 Some Notable Research Papers on Future of AI

https://arxiv.org/abs/2002.06177
78 |
 Some Notable Research Papers on Future of AI

https://arxiv.org/abs/1911.01547
79 |
 Some Notable Research Papers on Future of AI

https://arxiv.org/abs/2012.05876
80 |
 Six researchers who are shaping the future of artificial intelligence

https://www.nature.com/articles/d41586-020-03411-0
81 |
 The future of deep learning, according to its pioneers

https://vimeo.com/554817366

https://cacm.acm.org/magazines/2021/7/253464-deep-learning-for-ai/fulltext
82 | https://venturebeat.com/2021/07/05/the-future-of-deep-learning-according-to-its-pioneers/
 The future of artificial intelligence for Humanities & Social Science
Research

https://www.nature.com/articles/s41599-021-00750-9
83 |
 On the Opportunities and Risks of Foundation Models (Deep
Learning) by Stanford University

84 | https://arxiv.org/abs/2108.07258?sf149288348=1
 Follow on bhagirathl

85 |