Deep Learning

Dr. Farrukh Bhatti

Text/Reference Books
• Fundamentals of Deep Learning: Designing Next-generation machine intelligence
algorithms by Nikhil Buduma, 2017

Reference
• Deep Learning, An MIT Press book, by Ian Goodfellow and Yoshua Bengio and
Aaron Courville

• Neural Networks and Deep Learning by Michael Nielsen

Assessments
• Quizzes – 10%
• Course Project – 20%
• Presentation on project – 10%
• Paper writing on project – 10%
• Mid-term exam – 20%
• Final exam – 30%
 Highlights
• Introduction – Applications, History
• Perceptron – Linear perceptron as Neuron, Logistic Regression
• Neural Network
• Training Feed-Forward NN
• Practical Aspects – Test/training set, overfitting, hyper parameter tunning, data augmentation,
gradient descent
• ConvNet – Convolution-1D and 2D, Conv Filters, Forward and backward propagation, Feature
Maps, Pooling, FC, Batch Normalization, Transfer learning, Modern CNNs
• Object Detection – Classification + Regression, Regional proposal network, RCNN, Faster RCNN,
YOLO, Single Shot Detector
• Autoencoders - Learning lower-dimensional representations, PCA, AE architecture, Sparcity in
AEs, Stacked AEs
• Sequence Models – Recurrent NN, Vanishing gradients, LSTM
• Deep Reinforcement Learning
AI in Robots
• In the next 20 years, the
report said, the global stock
of robots would reach as
many as 20 million by 2030,
with 14 million in China only

• Millions of manufacturing
jobs to be lost to robots.

https://www.trtworld.com/life/how-robots-are-stealing-human-jobs-and-threatening-our-future-28285
Self-Driving Cars
• AI simulates human perceptual and decision-making processes using deep
learning and controls actions in driver control systems, such as steering and
brakes.
• Every piece of data (concerning
self-driving cars, we talk about
data received by the vehicle’s
sensors) goes through the multi-
layered neural network, enabling
analyzing images in a much more
comprehensive way.
Self-Driving Cars
• Waymo - a US-based company that’s working on the world’s first
autonomous ride-hailing service and autonomous trucking and local
delivery solutions.
• Network of
radars, lidars, and
cameras
• Waymo’s cars
have already
driven over 20
billion miles both
in the real world
and in simulations
 Self-Driving Cars
• Autonomous buses produced by the Aurrigo company will be operating
on the streets of British Cambridge.

https://youtu.be/6ttW78t4XKg
Image segmentation and recognition
AI for Urban Surveillance –
An Example of China

• https://www.youtube.com/watch
?v=7gSU_Xes3GQ
• 3.48 – 7.20
 Image Recognition

• Vehicle Make
and Model
Classification
• SafeCity
Project
Captions
Generated
using Neural
Networks
ThisPersonDoesNotExist.com uses AI to
generate fake faces
https://www.theverge.c
om/tldr/2019/2/15/182
26005/ai-generated-
fake-people-portraits-
thispersondoesnotexist-
stylegan
Image Segmentation using Deep Learning
• Urban mapping and
planning
• Precision agriculture
• Flood detection
• Target detection
• Mineral exploration
• Environmental monitoring
• Disaster monitoring
• Estimation of soil erosion
and salinity
Image Segmentation: Example
Deep Learning for Natural Language
Processing (NLP)
• NLP allows a computer to understand human language using deep
learning
• Computers require structured data – human speech is unstructured
and often ambiguous
• “The boy saw the man with the telescope”. (Syntactic Ambiguity)
• “The car hit the pole while it was moving”. (Semantic Ambiguity)

• Chatbots, Google voice search, Siri

Deep Learning for Natural Language
Processing (NLP)
Image recognition use cases
• AI image recognition software in the medical field is
used for detecting anomalies in tissues, including
tumours, bone cracks or various types of cancer.
Image recognition use cases
• In agricultural farms – used to identify which plants need watering and even spot plant diseases,
insects, and worms.
• AI image recognition can be used to scan large amounts of online content. Pre-trained models can
be used for content moderation (e.g. identify guns or nudity)
• Targeted advertising, brand awareness in shopping malls
Image recognition use cases
• Retail: Image recognition machine learning technologies can improve and automate store
management. The software could scan the image of a shelf surface and deduce various data such
as the shortage of certain products, misplaced items or even spot the use-by date of certain
products.
• Theft detection
• Self-checkout stations to scan and identify
products
Image recognition use cases
• Defect Detection on production lines
Image recognition use cases
• Crowd Monitoring
• UK-based startup Cawamo enables social distancing video monitoring. The
startup uses AI and a deep machine learning engine that detects people
gathering in public areas.
• It sends voice alerts to the gathering crowd without face masks and to the
relevant safety authorities. Moreover, the system records the events pre- and
post-alerts to analyze the impact of warnings.
• Government organizations and public authorities leverage this solution to
ensure the safety of people in public spaces.
Image recognition use cases
• Monitoreal is a Cypriot startup that
makes home security systems. The
startup’s product, Monitoreal Security
Assistant Pro, uses AI-based smart
object detection to identify small
moving objects from a considerable
distance even under poor lighting
conditions.
• Moreover, it integrates video
surveillance with other smart action
accessories such as security lights. It
also detects objects from low-
resolution camera footage, enabling
its use for surveillance analytics.
• This allows homeowners to monitor
their surroundings in real-time and
ignore false detections.
Early Models of Human Cognition

• Associationism
– Humans and animals learn through association
• 400BC-1900AD: Plato, David Hume, Ivan Pavlov..
What are “Associations”
• Lightning is generally followed by thunder
– Ergo – “hey here’s a bolt of lightning, we’re
going to hear
thunder”
– Ergo – “We just heard thunder; did someone
get hit by
lightning”?

• Experiments with dogs by Ivan Pavlov

Dogs learn to associate whistle with food

Examples?
Associationism
• Collection of ideas stating a basic philosophy:
– “Pairs of thoughts become associated based on the organism’s past experience”
– Learning is a mental process that forms associations between temporally related phenomena

360 BC: Aristotle

– "Hence, too, it is that we hunt through the mental train,
excogitating from the present or some other, and from similar or
contrary or coadjacent. Through this process reminiscence takes
place. For the movements are, in these cases, sometimes at the
same time, sometimes parts of the same whole, so that the
subsequent movement is already more than half accomplished.“

• In English: we memorize and rationalize through association

Aristotle and Associationism
Aristotle’s four laws of association:
– The law of contiguity. Things or events that occur close together in
space or time get linked together
– The law of frequency. The more often two things or events are
linked, the more powerful that association.
– The law of similarity. If two things are similar, the thought of one will
trigger the thought of the other
– The law of contrast. Seeing or recalling something may trigger the
recollection of something opposite
Where are Associations stored and how?

David Hartley’s Observations on man (1749)

• We receive input through vibrations and those are transferred to the

brain
• Memories could also be small vibrations (called vibratiuncles) in the
same regions
• Our brain represents compound or connected ideas by connecting
our memories with our current senses
• Current science did not know about neurons
Brain –
Interconnected
Neurons
Many neurons connect into each
neuron
Each neuron connects out to many
neurons
Connectionism

• Alexander Bain, philosopher, psychologist,

mathematician, logician, linguist, professor
• 1873: The information is in the connections
• Mind and body (1873)
Bain’s Idea 1: Neural Grouping
• Neurons excite and stimulate each other
• Different combinations of inputs can result in different outputs
Bain’s Idea 1: Neural Grouping
• Different intensities of
activation of a lead to
the differences in
when X and Y are activated

• Even proposed a
learning mechanism..
Bain’s Idea 2: Making Memories
• “When two impressions concur, or closely succeed one another, the
nerve-currents find some bridge or place of continuity, better or worse,
according to the abundance of nerve matter available for the
transition.”
Bain’s Doubts
• In 1873, Bain postulated that there must be one million neurons and
5 billion connections relating to 200,000 “acquisitions”
• In 1883, Bain was concerned that he hadn’t taken into account the
number of “partially formed associations” and the number of neurons
responsible for recall/learning
• By the end of his life (1903), recanted all his ideas!
– Too complex; the brain would need too many neurons and
connections
Connectionism lives on..
• Connectionism presents a cognitive theory based on
simultaneously occurring, distributed signal activity via
connections that can be represented numerically, where
learning occurs by modifying connection strengths based on
experience

• The human brain is a connectionist machine

• Bain, A. (1873). Mind and body. The theories of their relation. London:
Henry King.
• Ferrier, D. (1876). The Functions of the Brain. London: Smith, Elder
and Co
• Neurons connect to other neurons. The processing/capacity of
the brain is a function of these connections
• Connectionist machines emulate this structure
Connectionist Machines
Network of processing elements
• All world knowledge is stored in the
connections between the elements
• In most connectionist models, networks
change over time.
• A common aspect of connectionist models is
activation – an action potential spike
• Activation typically spreads to all the other
units connected to it
 Synaptic terminals

Brain Neuron
• Neuron – foundational unit of brain
• Approx. 86 billion neurons in brain
• Up to 15,000 connections with other
neurons via synapses
• Optimized to receive information from
other neurons, process this information in a
unique way, and send its result to other cells
 Synaptic terminals

Brain Neuron
• Each of these incoming connections is
dynamically strengthened or weakened
based on how often it is used
• Strength of each connection that
determines the contribution of the input to
the neuron’s output
• After being weighted by the strength of
their respective connections, the inputs are
summed together in the cell body
• The sum is then transformed into a new
signal that’s propagated along the cell’s
axon and sent off to other neurons
Modeling a Neuron
Connectionist Machines

Neural networks are connectionist machines

As opposed to Von Neumann Machines

VN machines have to learn by rules, while artificial neural networks learn by example
Dedicated chips for NN - have many non-linear processing units
- The program is the connections between these units
- Connections may also define memory
 Artificial
Intelligence Machine Learning
Any technique that Ability to learn without Deep Learning
enables computers to explicitly being
mimic human programmed, use human Extract patterns from data using neural
behavior created features, training networks, learn features through
training

In machine learning, instead of teaching a computer Deep learning is essentially a neural network with three or more
a massive list of rules to solve the problem, we give layers. These neural networks attempt to simulate the behavior
it a model with which it can evaluate examples, and of the human brain—albeit far from matching its ability—
a small set of instructions to modify the model allowing it to “learn” from large amounts of data. While a neural
when it makes a mistake. network with a single layer can still make approximate
predictions, additional hidden layers can help to optimize and
refine for accuracy. (IBM)
Example: Use of ANN
Three Types of AI
• Artificial narrow intelligence (ANI), which has a narrow range of
abilities – only this has been achieved
• Artificial general intelligence (AGI), which is on par with human
capabilities
• Artificial superintelligence (ASI), which is more capable than a human.

• Reading Assignment – read this link

• https://www.spiceworks.com/tech/artificial-
intelligence/articles/types-of-
ai/#:~:text=While%20narrow%20AI%20refers%20to,function%20just%
20as%20humans%20do.