Download 1 Week Lec 1 Introduction to Deep Learning

COSC 4121: Deep Learning and Applications
Lecture 1
Introduction to Deep Learning
COSC-4121 DL and Applications
Shahzad Hussain
Lecturer
Today’s Lecture outline
1. Course Evaluation Detail
2. Artificial
Intelligence,
Learning, and Deep Learning
1.
2.
3.
4.
5.
6.
7.
2
Machine
Artificial Intelligence
Machine Learning
Learning Representation from Data
The “Deep” in deep Learning
Understanding how deep learning works, in three
figures
Don’t believe the short-term hype
The promise of AI
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
1. Course Evaluation Procedure
COSC-4121 DL and Applications
Course Evaluation Procedure
• Mid Term 30%
• Final Term 50%
• Sessional
20%
– Quizzes
5%
– Assignments
5%
– Term Project
10%
– Presentations
– Class behavior
• Attendance must be maintained as per rule (>=75%)
4
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Recommended Book
•
Deep Learning with python
Francis Chollet
Practical Implementation & Reference Book
• Deep Learning
Ian Goodfellow
Yoshua Bengio
Aaron Courville
• Grokking Deep Learning
Andrew W. Trask
5
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
2. Artificial Intelligence, Machine Learning, and Deep Learning
COSC-4121 DL and Applications
6
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• Promised future of intelligent chatbots, selfdriving cars, and virtual assistants where
deep learning will play crucial roles.
COSC-4121 DL and Applications
7
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• The field of artificial intelligence: the
effort to automate intellectual tasks
normally performed by humans
COSC-4121 DL and Applications
8
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• The dominant paradigm in AI from 1950s to
the late 1980s:
Expert System (known as symbolic AI).
– Intractable to figure out explicit rules for
solving complicated rules (e.g., image
classification, speech recognition, language translation)
COSC-4121 DL and Applications
An example of a rule-based system
9
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• Machine learning system is trained
rather than explicitly programmed
– Many examples (data) are needed for
finding out the rules automatically
Rules
Data
Data
Answers
Classical
Programming
Machine
Learning
Answers
Rules
Machine Learning a new Programming Paradigm
10
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Artificial Intelligence, Machine Learning and Deep Learning
• To do machine learning we need three things:
– Input data points.
– Examples of the expected output (=labels)
– A measure for goodness (objective function)
• Machine learning models are all about finding
appropriate representations for input data
– e.g., coordinate change
COSC-4121 DL and Applications
11
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• The “deep” in deep learning
– Specifically, deep learning learns representations
automatically by involving successive layers of
increasingly meaningful representations.
COSC-4121 DL and Applications
Better representation
12
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
• The layers are parameterized by weights
o Weights are called ‘parameters’ of layers
o Learning means finding a set of values for the
weights of all layers such that the network will
correctly map example inputs to their associated
targets
Input X
Goal: finding
the right
values for
weights
Weights
Layer
(data transformation)
Weights
Layer
(data transformation)
Figure 1.7 A neural network is
parameterized by its weights
13
Prediction
Y’
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Artificial Intelligence, Machine Learning and Deep Learning
• Loss function (= objective function)
o Measuring how far the predictions are from
the true targets (quality of prediction)
Input X
Weights
Layer
(data transformation)
Weights
Layer
(data transformation)
Figure 1.8 A loss function
measures the quality of the
network’s output
Loss Score
14
Prediction
Y’
COSC-4121 DL and Applications
True targets
Y
Loss Function
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
•
Optimizer adjusts the weights with respect to minimizing the loss
Input X
Weights
Layer
(data transformation)
Weights
Layer
(data transformation)
COSC-4121 DL and Applications
Weights update
Prediction
Y’
Optimizer
Loss Score
True targets
Y
Loss Function
Figure 1.8 A loss score is used as a feedback signal to adjust the weights
15
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
•
What deep learning has achieved so far:
o Near-human-level image classification
o Near-human-level speech recognition
o Near-human-level handwriting transcription
o Improved machine translation
o Improved text-to-speech conversion
o Digital assistants such as Google Now and
Amazon Alexa
o Near-human-level autonomous driving
o Improved ad targeting, as used by Google,
Baidu, Bing
o Improved search results on the web
o Ability
to
answer
natural-language
questions
o Superhuman Go playing (defeating tophuman players)
16
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Artificial Intelligence, Machine Learning and Deep Learning
• Don’t believe the short-term hype
o Talk of human-level general intelligence should not
be taken too seriously.
• Too much data are required for training
• Too expensive in computation
• Too much expectations for AI
COSC-4121 DL and Applications
17
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Artificial Intelligence, Machine Learning and Deep Learning
•
The promise of AI
– Although we may have unrealistic short-term expectations for AI,
the long-term picture is looking bright
•
Similarly to Internet (most people didn’t see how the internet was relevant to
them and how it was going to change their lives)
– AI will answer your questions, help educate your kids, and watch
over your health
COSC-4121 DL and Applications
18
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Conversational Agents
COSC-4121 DL and Applications
Idea: Conversational Agent
• HAL 9000 is a fictional
artificial intelligence character
and the main antagonist in
Arthur C. Clarke's Space
Odyssey series.
• First appearing in the 1968
film
2001:
A
Space
Odyssey, HAL is a sentient
computer that controls the
systems of the Discovery One
spacecraft and interacts
with the ship's astronaut crew
20
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
Digital Personal Assistants and Chatbots
COSC-4121 DL and Applications
Digital Personal Assistants and
Chatbots
Google Assistant:
Google
Assistant
was
launched in 2016 and is
one of the most advanced
chatbots available.
It can be found in a variety
of appliances such as
telephones,
headphones, speakers,
washers,
TVs,
and
refrigerators.
22
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
Digital Personal Assistants and
Chatbots
Amazon Alexa:
Alexa is a virtual assistant
developed and marketed
by Amazon.
It can interact with users
by
voice
and
by
executing
commands
such as playing music,
creating to-do lists, setting
up
alarms,
playing
audiobooks, and answering
basic questions.
23
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
Digital Personal Assistants and
Chatbots
Apple Siri:
Siri can accept user voice
commands and a natural
language user interface to
answer
questions,
make
suggestions, and perform actions
by parsing these voice commands
and delegating these requests to a
set of internet services. The
software can adapt to users'
individual language usage,
their searches, and preferences.
The more it is used the more it
learns and the better it gets.
24
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
Digital Personal Assistants and
Chatbots
Microsoft Cortana:
Cortana
is
another
digital virtual assistant,
designed and created
by Microsoft. Cortana
can set reminders
and
alarms,
recognize
natural
voice commands, and
it answers questions
using information.
25
Shahzad Hussain, Lecturer, Khawaja Fareed University
of Engineering and Information Technology
COSC-4121 DL and Applications
Personal Chauffeur
COSC-4121 DL and Applications
Google’s Waymo
COSC-4121 DL and Applications
27
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
Google’s Waymo
• As of 2018, Waymo's autonomous cars have
driven eight million miles on public roads as well
as five billion miles in simulated environments.
• In the next few years, it is all but a certainty that
we will be able to purchase a car capable of full
driving autonomy.
• Tesla, among others, already offers driver
assistance with their Autopilot feature and
possibly will be the first company to offer full selfdriving capabilities.
• Waymo launched the first commercial driverless
service in 2018 in Arizona, USA with plans to
expand nationally and worldwide.
28
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Shipping and Warehouse Management
COSC-4121 DL and Applications
Shipping and Warehouse Management
• An Amazon sorting facility is one of the best
examples of the symbiotic relationship that is
forming between humans, computers, and
robots. Computers take customer orders
and decide where to route merchandise,
the robots act as mules carrying the pallets
and inventory around the warehouse.
• Even Scott Anderson, Amazon's director
of robotics fulfillment acknowledged in May
2019 that a fully automated warehouse is at
least 10 years away.
• So, we will continue to see this configuration
in warehouses across the world for a little
longer.
30
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Human Health
COSC-4121 DL and Applications
Human Health
1.
2.
3.
4.
5.
6.
7.
32
Drug Discovery
Insurance pricing
Patient Diagnosis
Medical Image Interpretation
Psychiatric Analysis
Smart Health Records
Disease Detection and Prediction
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Knowledge Search
COSC-4121 DL and Applications
Knowledge Search
• Google Search, is most often used
for exploring information.
• Early in 2015, as was reported by
Bloomberg, Google began using a
deep
learning
system
called
RankBrain to assist in generating
search query responses.
34
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Recommendation Systems
COSC-4121 DL and Applications
Recommendation Systems
• Recommendation systems are another
example of AI technology that has
been weaved into our everyday lives.
• Amazon,
YouTube,
Netflix,
LinkedIn, and Facebook all rely on
recommendation technology and we
don't even realize we are using it.
• Netflix Prize
• Pandora
• Robo-Advisors – betterment
36
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
The Smart Home
COSC-4121 DL and Applications
The Smart Home
• Home Monitoring:
– Ring Video Doorbell – Amazon
– Google Nest Thermostat
• Vacucmming and mopping
– Roomba
– Brava
– Packbot – Bomb disposal robot
• Picking up your mess
• Personal Chef
– Moley Robotic Kitchen
38
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Gaming
COSC-4121 DL and Applications
StartCraft 2
• StarCraft II is considered one of
the hardest, though it is an ancient
game by video game standards.
• The team at DeepMind introduced
a program dubbed AlphaStar,
defeat a top professional player, in
December 2018, AlphaStar whooped
a team put together by Grzegorz
"MaNa" Komincz, one of the
world's
strongest
professional
StarCraft players with a score of 50.
40
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
IBM Watson –
Question Answering Machine
• IBM and the Watson team
made history in 2011 when
they devised a system that
was able to beat two of the
most successful Jeopardy
champions.
• Watson
is
a
questionanswering system that can
answer questions posed in
natural language.
• It was initially created by
IBM's DeepQA research
team,
led
by
principal
investigator David Ferrucci.
41
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Chess
• The Deep Blue team went back to
the drawing board, made a lot of
enhancements to the software, and
played Kasparov in 1997.
• Deep Blue won the second round
against Kasparov winning the sixgame rematch by a score of 3½–
2½.
• It then became the first computer
system to beat a current world
champion in a match under standard
chess tournament rules and time
controls.
42
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
AlphaGo
• There is great example of a
powerful program that can play the
game of Go also developed by
DeepMind called AlphaGo.
• March 2016, it beat Lee Sedol in
a five-game match. This became
the first time a Go program beat a
9-dan
professional
without
handicaps. Although AlphaGo lost
to Lee Sedol in the fourth game,
Lee resigned in the final game,
giving a final score
• of 4 games to 1.
43
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Movie Making
COSC-4121 DL and Applications
Movie Making
• Deepfakes
– A deepfake is a portmanteau, or blend
of “deep learning” and “fake”.
– It an AI technique to merge video
images.
– Common application is to overlap
someone’s face onto another.
– Used to create fake news or hoaxes.
Example:
– Zao – developed by Chinese Company
Momo, allows you to overlap someone’s
face over short movie clips like Titanic
and the results are impressive.
45
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Movie Making
• Movie Script Generation
– One of the most famous
examples is Sunspring, award
winning.
– Sunspring is an experimental
fiction short film released in
2016.
– Written
by
using
deep
learning techniques.
– Script was created by LSTM
(Long short-term memory)
model dubbed Benjamin.
46
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Data Cleaning and Transformation
COSC-4121 DL and Applications
Data Cleaning and Transformation
• Clean and Accurate data is paramount to
producing consistent, reproducible, and
accurate AI models.
• Data scientist spends about 80% of their time
cleaning, preparing and transforming their
input data and 20% for optimizing their
models.
• Example: image data sets Image Net and
MS COCO.
• Amazon Lake formation, August 2019,
automates some steps of creation of a data
lake
including
collection,
cleansing,
deduplication,
cataloging,
and
publication of data.
48
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications
Today’s Lecture Summary
• Course Evaluation Detail
• Artificial Intelligence, Machine Learning,
and Deep Learning
–
–
–
–
–
–
–
49
Artificial Intelligence
Machine Learning
Learning Representation from Data
The “Deep” in deep Learning
Understanding how deep learning works, in three
figures
Don’t believe the short-term hype
The promise of AI
Shahzad Hussain, Lecturer, Khawaja Fareed University of Engineering and Information Technology
COSC-4121 DL and Applications