You are on page 1of 15

Machine Learning

Introduction to Machine Learning

is a field of study that gives computers the ability to learn


without being explicitly

1
By: Pedro Domingos
3

What is Machine Learning ?

Simple Definition -- Using Data to answer questions


Study of computer algorithms
that improve automatically
through experience.
Tom M. Mitchell provided a widely quoted, formal
definition of the algorithms studied in the machine
learning field (1998)
A well-defined learning task is given by <P, T, E>
Performance measure P
Class of tasks T
Experience E
A computer program is said to if its performance at
tasks in T, as measured by P, improves with experience E
2
Why Machine Learning is Used ?
Where human presence may not be possible/difficult
Navigating on Mars, Deep Oceans etc.
Humans are unable to use their skills effectively
Speech recognition, Face recognition, Handwritten Digit Recognition
etc.
Solution changes quickly from time to time and on the fly
Routing on a Computer Network, Electric Grid
Manual calculation not possible due to huge amounts of data
Recommending a product to buyers by Amazon, Flipkart, Netflix etc.
Predicting an outcome from sensor data, e.g., Weather prediction

When Machine Learning should NOT be Used ?


Machine Learning is not always useful!
Examples:
Calculate payroll of an Organisation
Dispense money from ATM's and link back to your account balance
Calculate utility bills (Telephone, Electricity etc.)

3
Artificial Intelligence

Artificial Intelligence
The designing and building of intelligent agents that receive percepts
from the environment and take actions that affect that environment.
Russel & Norvig, Artificial Intelligence: A Modern Approach
The theory and development of computer systems able to perform
tasks normally requiring human intelligence, such as visual
perception, speech recognition, decision-making, and translation
between languages.
Oxford Dictionary, https://www.lexico.com/definition/artificial_intelligence
The ability of a digital computer or computer-controlled robot to
perform tasks commonly associated with intelligent beings.
B.J. Copeland, https://www.britannica.com/technology/artificial-intelligence

4
Emerging AI Technologies

Credit: Accenture

AI Applications

Credit: https://www.accenture.com/nz-en/topic-accenture-government-artificial-intelligence

10

5
Deep Learning

11

Deep Learning
Machine learning using Artificial Neural Networks
ANNs are inspired by the structure and function of the brain
Deep => multiple layers in the network

12

6
AI
vs

ML

vs

DL Credit: https://en.wikipedia.org/wiki/Deep_learning

13

History and Evolution of AI/ML

12

7
Evolution
ENIAC In 1946, first computer system ENIAC was developed.
First game playing program for checkers in 1952 by Arthur Samuel
Perception In 1957 Frank Rosenblatt invented the Perception at the
Cornell Aeronautical Laboratory. It one of the earliest algorithm for ML.
A very simple linear classifier
By combining a large number of them in a network, a powerful model could be
created.
ELIZA Developed in early by Jospeph Weizenbaum.
Simulated a psychotherapist by using tricks like string substitution and canned
responses based on keywords.
When the original ELIZA appeared, some people mistook her for human.

15

Evolution (Contd.)
Statistical Al
Started in early 90s. The field shifted to more data driven approach as
compared to the knowledge-driven expert systems developed earlier.
Big Data Analysis
As a result of the exponential growth in the amount of data that
is available for scientific research, the science is on the brink of
huge changes
Statistical Al is a centre piece of Big Data analysis
The aim is to develop new methods to store these substantial amounts,
and to quickly find, analyse, and validate complex patterns in Big Data

16

8
Evolution (Contd.)
1994 Self driving car road test
1997 Deep blue beats Gary Kasparov
2009 Google builds self driving car
2011 Watson wins Jeopardy
2014 Human Vision Surpassed by ML System
2015 AlphaGo beats a human professional Go player
2019 Deepfake - videos from just one picture as input
The saga continues

17

Reasons for the growing popularity of AI/ML


New Software / Algorithms
Neural Network
Deep Learning
New Hardware GPUs
Cloud as an Enabler
Advances in Big Data processing

20

9
20
19
AI, ML, DL today: DeepFace

AI, ML, DL today: Self Driving Cars

Credit: https://thebfd.co.nz/2020/10/27/five-
things-to-know-about-making-self-driving-cars-
safe/

10
AI, ML, DL today: Personalized Recommendations

Credit: https://becominghuman.ai/how-netflix-uses-
ai-and-machine-learning-a087614630fe
21

AI, ML, DL today: Personal Assistants (Digital)

Credit: https://www.vox.com/2017/5/17/15655458/digital-assistants-voice-activated-operating-systems-alexa-siri-cortana-ios

20

11
Summary of Common Uses
Speech recognition
Drive automobiles
Healthcare: Discover Medicines, Diagnose Diseases
Routing in communication network
Understand handwritten text
Data mining
Personalised feedbacks, recommendations

23

And a few Failures CES 2018: Robot refuses to co-operate with LG chief
https://youtu.be/tQMtbWwbduA

24

12
And a few Failures Ai Camera Follows Bald Head Mistaking It For Football
https://youtu.be/9zoJP2FkpgU

25

And a few Failures ? Chihuahua vs. the Blueberry Muffin


https://blog.cloudsight.ai/chihuahua-or-muffin-1bdf02ec1680

24

13
27

Building ML Applications

28

14
Steps Involved in Building a ML Application
1. Collect data/Gathering Data
Deciding
2. Prepare the input data
Ensuring that there is no bias, cleanse , format etc.
3. Analyse the input data
Plotting, finding features etc.
4. Choosing a Model/Algorithm
Examples Random Forest, ANNs, Hidden Markov Models,etc
5. Training
Run the designed algorithm on training data (Not for unsupervised learning)
6. Testing/ Evaluation
How well did we do?
See to what extent it works on test data
7. Productization
After successful testing, use the application with the real (field) data

15

You might also like