Deep Learning Overview

Agenda
1. Introduction to Basic Terms and how they fit together:
○ AI, Big Data, Machine Learning, Deep Learning,
○ IOT, Neural Networks
2. Brief overview of use case of Deep Learning
3. How to Approach a Deep Learning Problem:
○ How to approach?
○ Which Tools?
○ Which Algorithms?
4. Questions & Answers

About CloudxLab
Videos Quizzes Hands-On Projects Case Studies
Real Life Use Cases
Making learning fun and for life

Automated Hands-on Assessments
Learn by doing

Problem Statement Hands On Assessment

Problem
Statement
Evaluation

Python Assessment Jupyter Notebook

Machine Learning
with
Python - Scikit Learn
Course Objective

Course Instructor
Sandeep Giri
Worked On Large Scale Computing
Graduated from IIT Roorkee
Software Engineer
Loves Explaining Technologies
Founder

Machine Learning
What Is Machine Learning?
Field of study that gives "computers the ability to
learn without being explicitly programmed"
-- Arthur Samuel, 1959

Machine Learning
What Is Machine Learning?
Let us understand it with real use case...

Machine Learning
Have You Played Mario?
How much time did it take you to learn & win the princess?

Machine Learning
Have You Played Mario?
Did Anyone teach you?

Machine Learning
How About Automating it?

Machine Learning
• Program Learns to Play Mario

Machine Learning
• Observes the game & pressed keys

Machine Learning
• Observes the game & pressed keys
• Maximises Score

Machine Learning
● So, the program learnt to play
○ Mario
○ And Other games
○ Without any programming

Machine Learning
Question
To make this program learn any other games such as PacMan we will have to
1. Write new rules as per the game
2. Just hook it to new game and let it play for a while

Machine Learning
Gather data and automatically solve problems
Imagine Doing The Same For Life

Machine Learning
The Machine Learning Tsunami - 1
● Self driving cars on the roads

Machine Learning
● Netflix movies recommendations

Machine Learning
● Amazon product recommendations

Machine Learning
● Accurate results in Google Search

Machine Learning
● Speech recognition in your smartphone

Machine Learning
Question
What do we need to
● Gather Data
● And automatically solving the problem?
IntelligenceData
+

Machine Learning
Collect Data or Take Actions - IOT
Phone & Devices
Cheaper, faster and smaller
Connectivity
Wifi, 4G, NFC, GPS

Machine Learning
Intelligence - Traditional vs ML.
How you would write a spam filter?

Machine Learning
Intelligence - Spam Filter - Traditional Approach
21
3

Machine Learning
Problems?

Machine Learning
● Problem is not trivial
○ Program will likely become a long list of complex rules
○ Pretty hard to maintain
● If spammers notice that
○ All their emails containing “4U” are blocked
○ They might start writing “For U” instead
○ If spammers keep working around spam filter, we will need to keep writing
new rules forever
Problems?

Machine Learning
Intelligence - Spam Filter - ML Approach

Machine Learning
● A spam filter based on Machine Learning techniques automatically learns
○ Which words and phrases are good predictors of spam
○ By detecting unusually frequent patterns of words
● The program will be
○ Much shorter
○ Easier to maintain
○ Most likely more accurate than traditional approach

Machine Learning
● Unlike traditional approach, ML techniques automatically notice that
○ “For U” has become unusually frequent in spam flagged by users and
○ It starts flagging them without our intervention

Machine Learning
Can help humans learn
● ML algorithms can be inspected to see what they have learned
● Spam filter after enough training
○ Reveals combinations of words that it believes are best predictors of spam
○ May reveal unsuspected correlations or new trend and
○ Lead to a better understanding of the problem for humans

Machine Learning
Can help humans learn

Machine Learning
What is AI?
Artificial intelligence (AI):
The intelligence exhibited by machines

Machine Learning
What is AI?
• The theory and development of
computer systems

Machine Learning
What is AI?
computer systems
• To perform tasks requiring human
intelligence such as

Machine Learning
What is AI?
computer systems
• Visual perception

Machine Learning
What is AI?
computer systems
• Speech Recognition

Machine Learning
What is AI?
computer systems
• Decision Making

Machine Learning
What is AI?
computer systems
• Decision Making
• Translation between languages

Machine Learning
History - Summer of 1956
• The term artificial intelligence was
coined by
• John McCarthy
• In a workshop at
• Dartmouth College in New
Hampshire
• Along with Marvin Minsky,
Claude Shannon, and Nathaniel
Rochester

Machine Learning
Sub-objectives of AI
Artificial
Intelligence
Natural
language
processing
Navigate
Represent
Knowledge
ReasoningPerception

Machine Learning
AI - Represent Knowledge
• Understanding and classifying terms or
things in world e.g.
• What is computer?
• What is a thought?
• What is a tool?
• Languages like lisp were created for the
same purpose

Machine Learning
AI - Reasoning
• Play puzzle game - Chess, Go, Mario
• Prove Geometry theorems
• Diagnose diseases

Machine Learning
AI - Navigate
• How to plan and navigate in the real world
• How to locate the destination?
• How to pick path?
• How to pick short path?
• How to avoid obstacles?
• How to move?

Machine Learning
AI - Natural Language Processing
• How to speak a language
• How to understand a language
• How to make sense out of a sentence

Machine Learning
AI - Perception
• How to we see things in the real world
• From sound, sight, touch, smell

Machine Learning
AI - Generalised Intelligence
• With these previous building blocks, the
following should emerge:
• Emotional Intelligence
• Creativity
• Reasoning
• Intuition

Machine Learning
AI - How to Achieve
Artificial Intelligence
Machine Learning
Rule Based Systems
Expert System
Domain Specific
Computing
Robotics
Deep
Learning

Machine Learning
AI - How to Achieve
Artificial Intelligence
Machine Learning
Rule Based Systems
Expert System
Domain Specific
Computing
Robotics
Deep
LearningWe will focus here.

Machine Learning
Machine Learning - Types
Human Supervision?
Machine Learning
How they generalize?
Learn Incrementally?

Machine Learning
Human Supervision?
Supervised
Machine Learning
Unsupervised
Reinforcement

Machine Learning
Machine Learning - Supervised Learning
Whether or not models are trained with human
supervision

Machine Learning
Human Supervision?
Supervised
Machine Learning
Unsupervised
Reinforcement
Classification
Regression

Machine Learning
Classification
● The training data we feed to the algorithm includes
○ The desired solutions, called labels
● Classification of spam filter is a supervised learning task

Machine Learning
Classification
● Spam filter
○ Is trained with many example emails called training data.
○ Each email in the training data contains the label if it is spam or ham(not spam)
○ Models then learns to classify new emails if they are spam or ham
Classify new email as
Ham or Spam

Machine Learning
Regression - Predict the price of the car (Value)

Machine Learning
Regression
● Predict price of the car
○ Given a set of features called predictors such as
○ Mileage, age, brand etc
● To train the model
○ We have to give many examples of cars
○ Including their predictors and labels(prices)

Machine Learning
Machine Learning - Gradient Descent
• Instead of trying all lines, go into
the direction yielding better
results

Machine Learning
Machine Learning - Gradient Descent
● Imagine yourself blindfolded on the
mountainous terrain
● And you have to find the best lowest
point
● If your last step went higher, you will
go in opposite direction
● Other, you will keep going just faster

Machine Learning
Machine Learning - Unsupervised Learning
● The training data is unlabeled
● The system tries to learn without a teacher

Machine Learning
Human Supervision?
Supervised
Machine Learning
Unsupervised
Reinforcement
Classification
Regression
Clustering

Machine Learning
Clustering - Detect group of similar visitors in your blog

Machine Learning
Clustering
● Detect group of similar visitors in blog
○ Notice the training set is unlabeled
● To train the model
○ We just feed the training set to clustering algorithm
○ At no point we tell the algorithm which group a visitor belongs to
○ It find groups without our help

Machine Learning
Clustering
● It may notice that
○ 40% visitors are comic lovers and read the blog in evening
○ 20% visitors are sci-fi lovers and read the blog during weekends
● This data helps us in targeting our blog posts for each group

Machine Learning
• In the form of a tree
• Nodes closer to each other are similar
Hierarchical Clustering - Bring similar elements together

Machine Learning
Anomaly Detection - Detecting unusual credit card transactions to prevent
fraud

Machine Learning
Machine Learning - Reinforcement Learning

Machine Learning
● The learning system an agent in this context
○ Observes the environment
○ Selects and performs actions and
○ Get rewards or penalties in return
○ Learns by itself what is the best strategy (policy) to get most reward over time

Machine Learning
Applications
● Used by robots to learn how to walk
● DeepMind’s AlphaGo
○ Which defeated world champion Lee Sedol at the game of Go

Machine Learning
Human Supervision?
Supervised
Machine Learning
Unsupervised
Reinforcement
Classification
Regression
Clustering
Batch Processing
Online

Machine Learning
Machine Learning - Batch Learning

Machine Learning
● Offline learning
● System is incapable of learning incrementally
○ It must be trained offline using all the available data
● Takes lot of time and computing resources
○ everytime training happens on the entire data

Machine Learning
● Once the system is trained, it gets
○ Pushed to production
○ Runs without learning anymore
○ Just applies what it has learned offline

Machine Learning
Machine Learning - Online Learning

Machine Learning
Human Supervision?
Supervised
Machine Learning
Unsupervised
Reinforcement
Classification
Regression
Clustering
Batch Processing
Model based
Online
Instance Based

Machine Learning
Machine Learning - Instance-based Learning

Machine Learning
Machine Learning - Instance-Based Learning
● Most trivial form of learning is
○ Learn by heart
● The system learns the examples by heart
● Then generalizes to new cases using a similarity measure

Machine Learning
Machine Learning - Instance-Based Learning
Example
● Spam filter flags emails
○ That are identical to known spam emails (emails marked spam by users)
○ Also the emails which are similar to known spam emails
○ This requires measure of similarity between two emails
○ A basic similarity measures between two emails can be
■ Count the number of words they have in common

Machine Learning
Machine Learning - Model-based Learning

Machine Learning
Machine Learning - Model-Based Learning
● Another way to generalize from a set of examples
○ Build a model of these examples
○ And then use model to make predictions
○ This is called inference
○ Hope that model will generalize well
○ We will learn more about it in next session

Machine Learning
Machine Learning - Artificial Neural Network(ANN)
Computing systems inspired by the biological neural networks that constitute animal
brains.

Machine Learning
Machine Learning - Artificial Neural Network(ANN)
• Learn (progressively improve
performance)
• To do tasks by considering examples
• Generally without task-specific
programming
• Example: Based on image - cat or no
cat?

Machine Learning
Machine Learning - Deep Learning
Each Neuron
Hot Water Cold Water
Each Neuron is like the knob.

Machine Learning
Machine Learning - Deep Learning
Each Neuron
Each Neuron is like the knob.
Soap

Machine Learning
Deep Learning
Each Neuron
What if there are many more parameters? So, physical input is conceptual input.
Soap
Person -
Male/Female
Climate?

Machine Learning
Deep Learning
Each Neuron
We can construct many architectures….
Soap
Person -
Male/Female
Climate?

Machine Learning
Deep Learning
Multiple layers of neutrons

Convolutional Neural Network
● Although in 1996
○ IBM’s Deep Blue supercomputer
○ Beat the chess world champion Garry Kasparov

● Yet computers were unable to do trivial tasks such as
○ Detecting a puppy in a picture or
○ Recognizing spoken words
○ Until quite recently

● Convolutional neural networks (CNNs) emerged
○ From the study of the brain’s visual cortex, and
○ They have been used in image recognition since the 1980s.

● In the last few years
○ CNNs have managed to achieve superhuman performance
○ On some complex visual tasks
● And all this was possible because of
○ Increase in computational power
○ The amount of available training data
○ And the tricks presented in last chapter on training deep neural nets

● Today CNNs power
○ Image search services
○ Self-driving cars
○ Automatic video classification systems
○ Voice recognition and
○ Natural language processing - NLP

● In this chapter we will present
○ Where CNNs came from
○ What their building blocks looks like and
○ How to implement them using TensorFlow
● Then we will present some of the best CNN architectures

The Architecture of the Visual Cortex

● In 1958 and 1959, David H. Hubel and Torsten Wiesel
○ Performed a series of experiments on cats and
○ Later on monkeys
● Their experiments gave crucial insights on the
○ Structure of the visual cortex
● They showed that many neurons in the visual cortex
○ Have a small local receptive field
○ Meaning they react only to
○ Visual stimuli located in a limited region of the visual field

Local receptive fields in the visual cortex

Question
Why not simply use a regular deep network with fully connected
layers for image recognition tasks?

Answer
● Deep neural network work fine for small images such as MNIST
● But they break for larger images because of
○ Huge number of parameters
● For example
○ A 100x100 image has 10,000 pixels
○ If the first layer has 1,000 neurons (which is a very small number)
○ This means a total of 10 million connections, that too in first layer
○ This will require a lot of computing power
● CNNs solve this problem by using partially connected layers

Convolutional Layer

● It is the most important
building block of a CNN
● Neurons in the first
convolutional layer are not
connected to every single
pixel in the input image , but
only to pixels in their
receptive fields
Convolutional Layer

● In turn, each neuron in the
second convolutional layer is
connected only to neurons
located within a small
rectangle in the first layer.
● This architecture allows the
network to concentrate on
low-level features in the first
hidden layer, then assemble
them into higher-level
features in the next hidden
layer, and so on.
Convolutional Layer

● This hierarchical structure is
common in real-world images,
which is one of the reasons
why CNNs work so well for
image recognition.
Convolutional Layer

Convolutional Layer
Stride
This figure shows how the layers of CNN are formed

CNN Architectures
Distinguishing dog Breeds based
on pics may require really
complex nerual network.

Recurrent Neural Network
● Predicting the future is what we do all the time
○ Finishing a friend’s sentence
○ Anticipating the smell of coffee at the breakfast or
○ Catching the ball in the field
● In this chapter, we will cover RNN
○ Networks which can predict future
● Unlike all the nets we have discussed so far
○ RNN can work on sequences of arbitrary lengths
○ Rather than on fixed-sized inputs

Recurrent Neural Network - Applications
● RNN can analyze time series data
○ Such as stock prices, and
○ Tell you when to buy or sell

● In autonomous driving systems, RNN can
○ Anticipate car trajectories and
○ Help avoid accidents

● RNN can take sentences, documents, or audio samples as input and
○ Make them extremely useful
○ For natural language processing (NLP) systems such as
■ Automatic translation
■ Speech-to-text or
■ Sentiment analysis

● RNNs’ ability to anticipate also makes them capable of surprising creativity.
○ You can ask them to predict which are the most likely next notes in a
melody
○ Then randomly pick one of these notes and play it.
○ Then ask the net for the next most likely notes, play it, and repeat the
process again and again.
Here is an example melody produced by Google’s Magenta project

● In this chapter we will learn about
○ Fundamental concepts in RNNs
○ The main problem RNNs face
○ And the solution to the problems
○ How to implement RNNs
● Finally, we will take a look at the
○ Architecture of a machine translation system

Recurrent Neurons

Recurrent Neurons
● Up to now we have mostly looked at feedforward neural networks
○ Where the activations flow only in one direction
○ From the input layer to the output layer
● RNN looks much like a feedforward neural network
○ Except it also has connections pointing backward

Recurrent Neurons
● Let’s look at the simplest possible RNN
○ Composed of just one neuron receiving inputs
○ Producing an output, and
○ Sending that output back to itself
Input
Output
Sending output back to itself

Recurrent Neurons
● At each time step t (also called a frame)
○ This recurrent neuron receives the inputs x(t)
○ As well as its own output from the previous time step y(t–1)
A recurrent neuron (left), unrolled through time (right)

● For example, at the first
step the word “Je” may
have a probability of
20%, “Tu” may have a
probability of 1%, and so
on
● The word with the
highest probability is
output
Machine Translation
An Encoder–Decoder Network for Machine Translation

Reinforcement Learning
Learning to Optimize Rewards

● In Reinforcement Learning
○ A software agent makes observations and
○ Takes actions within an environment and
○ In return it receives rewards

Goal?

Goal
Learn how to take actions in order to maximize
reward

● In short, the agent acts in the environment and
○ Learns by trial and error to
○ Maximize its reward

So how can we apply this in real-life applications?

Learning to Optimize Rewards - Walking Robot

● Agent - Program controlling a walking robot
● Environment - Real world
● The agent observes the environment through a set of sensors such as
○ Cameras and touch sensors
● Actions - Sending signals to activate motors

● It may be programmed to get
○ Positive rewards whenever it approaches the target destination and
○ Negative rewards whenever it
■ Wastes time
■ Goes in the wrong direction or
■ Falls down

Learning to Optimize Rewards - Ms. Pac-Man

● Agent - Program controlling Ms. Pac-Man
● Environment - Simulation of the Atari game
● Actions - Nine possible joystick positions
● Observations - Screenshots
● Rewards - Game points
Learning to Optimize Rewards - Ms. Pac-Man

Learning to Optimize Rewards - Thermostat

● Agent - Thermostat
○ Please note, the agent does not have to control a
○ Physically (or virtually) moving thing
● Rewards -
○ Positive rewards whenever agent is close to the target temperature
○ Negative rewards when humans need to tweak the temperature
● Important - Agent must learn to anticipate human needs
Learning to Optimize Rewards - Thermostat

Learning to Optimize Rewards - Auto Trader

● Agent -
○ Observes stock market prices and
○ Decide how much to buy or sell every second
● Rewards - The monetary gains and losses
Learning to Optimize Rewards - Auto Trader

● There are many other examples such as
○ Self-driving cars
○ Placing ads on a web page or
○ Controlling where an image classification system
■ Should focus its attention

● Note that there may not be any positive rewards at all
● For example
○ The agent may move around in a maze
○ Getting a negative reward at every time step
○ So it better find the exit as quickly as possible

Policy Search

● The algorithm used by the software agent to
○ Determine its actions is called its policy
● For example, the policy could be a neural network
○ Taking observations as inputs and
○ Outputting the action to take
Policy Search

Machine Learning
Machine Learning - Who is Using?
Almost Everyone

Machine Learning
Google Translate & Auto Draw
More use cases: https://aiexperiments.withgoogle.com/

Machine Learning
TensorFlow - Demo
http://playground.tensorflow.org/

Machine Learning
Deep Learning Frameworks

Machine Learning
Machine Learning Frameworks
MLlib - Distributed Simple

Machine Learning
Learn More?
Machine Learning Courses
Machine Learning
Specialization
Machine Learning

Questions?
https://discuss.cloudxlab.com
reachus@cloudxlab.com

Deep Learning Overview

Recommended

Recommended

More Related Content

What's hot

What's hot (17)

Similar to Deep Learning Overview

Similar to Deep Learning Overview (20)

More from CloudxLab

More from CloudxLab (20)

Recently uploaded

Recently uploaded (20)

Deep Learning Overview