What is Artificial Intelligence ?

Artificial Intelligence : (Merriam-Webster ) The capability of

a machine to imitate intelligent human behavior.

What is Machine Learning ?

Machine learning is a branch of artificial intelligence based on
the idea that systems can learn from data, identify patterns and
make decisions with minimal human intervention.

38
Data Type
• Discrete data:
• Discrete non-ordered numbers
• Random collection of words
• Unrelated audio sounds
• Random music notes
• Sequential (temporal) data:
Sequential
• Stochastic process
• Sequence of words in a sentence
Spatio-temporal
• Audio speech data data
• Music
• Spatial data:
• Image data
• Geo-spatial data
Major Components of Data analytics
1. Sensor Technology
• Newer modalities
• Higher resolution
• Higher speeds Data
2. Communication and storage
• Challenge of volume, variety and velocity
• Security challenges
3. Computational hardware
• Cloud computing, GPUs, TPUs, Quantum computing
4. Algorithms
• Artificial intelligence (AI) Analytics
• Classical Machine Learning
• Deep learning
• Reinforcement Learning
Who’s Generating Big Data…
Old Model: Few companies are generating data, all others are consuming data

New Model: all of us are generating data, and all of us are consuming data

41
Major Companies are tracking every move we make
Our cell phones are continuously collecting data:
• Time stamped GPS coordinates and movements (e.g. walking, onBicycle,
inRailVehicle, in car, etc)
• The barometric pressure
• The MAC address, signal strength, and frequency of every nearby wifi access point
and Bluetooth beacon
• Battery life and charging state of the phone
• Open databases
• Social Media (FB, Twitter, true-caller,…)
• Voter ID
• Auto registration
• Utility bills
• Amazon Echo and other devices
42
How much data?
• One Crystal plasticity FEM calculation
produced 65 GB data (2008)
• 1 sq cm sample imaged at 1 nm resolution
(16 bit) would need 2 TB
• CERN’s Large Hydron Collider (LHC)
generates 15 PB a year
640K ought to
be enough for
anybody.
Information creation is outpacing available storage

Ref:IDC
Machine Learning
Supervise learning
(classification and
regression)

Unsupervisor
learning
(clustering,
Federated learning
dimensionality
reduction, an
normal detection)

Machine
learning
Semi supervised
Transfer learning
learning

Reinforcement
learning
(multi-step
problem)
Evolution of Artificial Intelligence

• Artificial Intelligence : (Merriam-Webster ) The capability of a machine to

imitate intelligent human behavior.
Artificial Artificial
Artificial Artificial Super
Narrow General
Intelligence Intelligence
Intelligence Intelligence
AI ASI
ANI AGI

First Wave Second Wave Third Wave

Activity Approach Driver Capability and performance

AI Perform a task Rule based Definite cost Domain specific; lower that human performance
function
Machine Learning Techniques in Data analytics
• Information-based Learning • Error-based Learning
– Decision Trees – Multivariable Regression
– Shannon’s Entropy – Linear discriminate analysis
– Information Gain – Multinomial Logistic Regression
• Similarity-based Learning – Support Vector Machines
– Feature Space • Expert-system based learning
– Distance Metrics
• Probability-based Learning
– Naïve Bayes Model

Performance
– Markovian model Conventional ML

Data 47
Machine Learning Techniques in Data analytics

• Information-based Learning • Error-based Learning

– Decision Trees – Multivariable Regression
– Shannon’s Entropy – Linear discriminate analysis
– Information Gain – Multinomial Logistic Regression
• Similarity-based Learning – Support Vector Machines
– Feature Space • Expert-system based learning
– Distance Metrics
• Probability-based Learning Deep Learning
– Naïve Bayes Model

Performance
– Markovian model Conventional ML

Deep Learning
Convolutional neural network
Recurrent neural network Data 48
Self-driving Cars

Dominos self-driving
pizza delivery vehicle

Waymo’s self-driving cars

has driven 13 Million Km
DL surpasses human ability in narrow areas

https://devopedia.org/imagenet 50
Google DL Retinopathy

Age Biological Sex Smoking

Predicted: 59.1 years Predicted: Female Predicted: Non-smoking
Actual: 57.6 years Actual: Female Actual: Non-smoking
Image of retina

A1C BMI Systolic blood Pressure

Predicted: Non-diabetic Predicted: 24.1 kg/m Predicted: 148.0 mmHg
Actual: Non-diabetic Actual: 26.3 kg/m Actual: 148.5 mmHg

51
Lip-reading AI

Google’s DeepMind
• AI trained on 5000 hours of TV
• 118,000 sentences

Other resources: LipNet AI, WAS

52
Deep image reconstruction:
Reading the brain

Ref: Prof. Yukiyasu Kamitani, University of Kyoto Japan

Google’s AlphaGo AI beats the world’s best
human Go player
● Go is one of the most complex board game
● AlphaGo secured the victory after winning the second
game in a three-part match against Ke Jie

https://www.bbc.com/news/technology-40042581
AlphaGo Zero

55
DEEP LEARNING FOR SYMBOLIC MATHEMATICS

Comparison of our model with

Mathematica, Maple and Matlab
on a test set of 500 equations

Examples of problems that DL model is able to solve, on which Mathematica and Matlab
were not able to find a solution. For each equation, DL model finds a valid solution with
greedy decoding.
Lample and Charton arXiv: 1912.01412
56
What can Artificially intelligent not do?

57
AI Is Now Writing Poetry, Music And song
popularity recommendations
A team of researchers from Microsoft and
Kyoto University developed a poet AI

Poem Written by AI

OptimiseLab uses AI to predict ‘music popularity

scores’

Song composed by AI
https://www.youtube.com/watch?time_continue=22&v=LSHZ_b05W7o
AI can beat us at games—but sometimes, that’s by cheating

Watch Google's AI master the infamously difficult Atari game

Montezuma's Revenge
Generative Adversarial Network (GAN)

Fine Tune the Generator

Generative Adversarial Network (GAN)

2014 2018

https://thispersondoesnotexist.com/
What is common in all these persons?
 Various robots for human companionship

Paro robot (Source National Institute

of Advanced Industrial Science and
The Sophia robot (Source Hanson Robotics) Technology)

A real doll (Source real doll)

AI in legal justice system

• Many courts in the US are adopting an AI based Online Dispute

Resolution (ODR) system

• At Superior Court, Los Angeles, AI handle their traffic citations

• Wisconsin Department of Corrections uses AI to determine jail

sentence

64
Ethical Aspects of AI in society

• Can social AI influence elections around the world

• AI healer: Replika is an AI personal companion for mental wellness.

• Is it an alternate to chat therapy?
• Do we not talk to god?

Is this ethical?

65
Immortal twin, the human digital clones.
The Holy Grail of artificial
•Immortal twin is a cyber clone/replica of a person.
•It knows all the past memories of the person, has the
exact personality, intelligence, mannerism as the
person.
•It also looks (on the digital screen) and speaks exactly
as the person. It actively Learns with the person, and
keeps learning from the happenings of current affairs
even after the person is no more.
Future of conflicts
• Autonomous weapons
• Swarm drones
• Space and Deep sea wars
• Directed-energy (laser, microwaves & particle beams)
• Electromagnetic pulse disruptor
• Active Cloaking (future of stealth)
• Cyber attack on strategic installations

Deep Learning enabled Artificial Intelligence

Autonomous Real-Time Ground Ubiquitous
Surveillance Imaging System (ARGUS-IS)
• 100 sq km area
• 20,000 feet above ground
• 1.8 Gigapixels video system
• 1 exabyte of HD video per day
• Resolution of 6”
• Automatic object-tracking
• 18.5 M USD
X-47B Pegasus
• Unmanned combat air vehicle (UCAV)
• Aircraft carrier-based operation
• Tailless jet-powered blended-wing-body
aircraft
• Autonomous operation & aerial
refueling
• Cruise speed: Mach 0.9
• Range: 3,889+ km
• Service ceiling: 42,000 ft
• 2 weapon bays, 2,000 kg of ordnance
• Stealth: Invisible to radar & eye
• Project cost: 0.8 B USD

70
Moral Machine experiment

Example question from the Moral Machine experiment that

confronted people with trolley problems (Source MIT)
Knowledge Based Learning
• Deep learning
• Bottom-up statistical approach analyzing and deriving insights from data.
• Knowledge Graphs (Simple Taxonomies to Expressive Ontologies)
• Top-down symbolic conceptual reasoning based inferences from data

87
Knowledge Graphs are currently in use
schema.org

Size of Knowledge graph

Primary entities Assertions/facts

Microsoft 2 billion 55 billion
Google 1 billion 70 billion
Facebook 50 million 0.5 billion
eBay 100 million 1 billion
IBM 100 million 5 billion

88
Evolution of Artificial Intelligence
• Artificial Intelligence : (Merriam-Webster ) The capability of a machine to
imitate intelligent human behavior.
Artificial Artificial
Artificial Artificial Super
Narrow General
Intelligence Intelligence
Intelligence Intelligence
AI ASI
ANI AGI

First Wave Second Wave Third Wave

Activity Approach Driver Capability and performance

AI Perform a task Rule based Definite cost Domain specific; lower that human performance
function
ANI Perform a task Self learned (ML) Non-explicitly (RL) Domain specific; surpasses human performance
Automatic
AGI Overarching Self learned (ML) Goal Universal domain; equivalent to human
goal performance
Evolution of Artificial Intelligence
• Artificial Intelligence : (Merriam-Webster ) The capability of a machine to
imitate intelligent human behavior.
Artificial Artificial
Artificial Artificial Super
Narrow General
Intelligence Intelligence
Intelligence Intelligence
AI ASI
ANI AGI

First Wave Second Wave Third Wave

Activity Approach Driver Capability and performance

AI Perform a task Rule based Definite cost Domain specific; lower that human performance
function
ANI Perform a task Self learned (ML) Non-explicitly (RL) Domain specific; surpasses human performance
Automatic
AGI Overarching ASI ->
Self learned Self awareness
(ML) Goal -> Consciousness
Universal domain; equivalent to human
goal performance
ASI Overarching Self learned Goal Universal domain; surpasses human performance
goal (ML)
From Data to Wisdom
1. Data (numbers, text, images, sounds, etc.)
2. Information (structured data about a particular subject)
3. knowledge (abstraction of information to ideas and principles)
4. Knowledge graph (aggregation of knowledge from various domains to
form and Holistic understanding of an entire ecosystem)
5. Intellect is defined as the faculty of reasoning and understanding
objectively. It is usually accepted that intellect has reasoning but has no
emotions.
6. Collective intelligence
7. Wisdom

94