Design
How to solve a problem using
Deep Learning
AI for Good Workshop
July 2019
Session 3
https://sites.google.com/view/AIforEveryone
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key Learning objective
for Day 1
How to solve a problem with Deep Learning ?
How to design an neural network architecture?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz #1
Is it possible to map hand signs for deaf  speech
 Hand signs   Speech
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
A B
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz #2
Is it possible to “see the depth” for visual impaired
 Yes / no ?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz #2
Is it possible to feel the depth for visual impaired
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Topics in this presentation
•New way to think from today !
• How to use Deep Learning to solve any problem
• How to model & architect Neural Networks
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
New way to think from today !
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
New way to think from today !
Do you have
 Large dataset?
 Compute resource (can pay easy to cloud)
 Ability to use frameworks such as Tensorflow
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key Learning objective
for Day 1
How to design an neural network architecture?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key Learning objective
for Day 1How to design an neural network architecture?
How to map x to y?
What
architecture
can connect
x to y?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key Learning objective
for Day 1How to design an neural network architecture?
Predicted
class
How to map x to y?
What architecture can map x to y?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Deep Learning idea-a-thon:
How to solve a problem
by designing novel neural network architectures?
CNN
RNN
Fully
Connected
Merge
CNN is suitable architecture to extract
features out of any data that has
spatial relationships. For example, An
photo image spatial info..or info
spread across space
In general, a recurrent neural network could
be considered as the best neural network
model for extracting features out of
temporal data. A temporal data is basically
a data that varies over time such as ECG,
music, speech, sentences, words. LSTM is
a advanced RNN
To classify / predict
To combine the learnings of
two neural networks.
(to combine the power of 2
people’s brains)
Attention
To learn to focus on the
most important aspects to
achieve a particular task
Generative
Deep Learning
To model creative tasks as
to compose a music or
to come up a painting
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
How to use
Deep learning
to learn
to connect A and B
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learning objective:
How to learn to map A  B
Deep
Learning
A
B
Rules or math function
to find B given A.
B = f(A)
What should be f()?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
How to translate A to B
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learn to map Xy
• Examples of x, y
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learn to map Xy
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learn to map A  B
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Fun activity
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learning objective:
How do solve a problem
using Deep Learning?
Case study: self driving car
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key Learning objective
for Day 1
How to solve a problem with Deep Learning ?
How to design an neural network architecture?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #1
classification problems
CNN
Fully
Connected
Prediction of what
of objects is in this
photo
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #1
CNN
Fully
Connected
Turn car left
Turn car right
Don’t Turn the car
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Learn to map x->y
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #1
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #1
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Camera view ? steering angle
•
Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car
Credits: Nvidia
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
what the neural network learns depends
upon its purpose
• How to set the purpose of a “learning machine”?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
What is the function to map x->y
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #2
Regression problems
CNN
Fully
Connected
Continuous number
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #2
Regression problems
CNN
Fully
Connected
Turn the car by 10 degrees
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #2
Regression problems
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #2
Regression problems
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #3
multimodal classification
CNN
Fully
Connected
Predicted
class
RNN
Merge
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #3
multimodal classification
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #3
multimodal classification
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
Neural Network Design Pattern #3
multimodal classification
•
Quiz: What was x and y in the problem?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
Neural Network Design Pattern #3
• Quiz: What was x and y in the problem?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
multimodal classification
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Answering Visual Questions from Blind People
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #3
multimodal classification
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #3
multimodal classification
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #3
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz/Thinking activity
how to classify a genomic sequence
healthcare/genomic sequence classification
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
How to Solve this problem with deep learning:
how to classify a genomic sequence
?
?
?
Credits: DeepMotif (ICLR’16) http://www.cs.virginia.edu/yanjun/paperA14/2017_demo_slides.pdf
This is a classification problem.
A: Sequence of characters
B: Category
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
How to Solve this problem with deep learning:
how to classify a genomic sequence
?
?
?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
how to classify a genomic sequence
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #4
video question answering
CNN
Fully
Connected
Predicted
class
RNN
Merge
Video
Question
RNN
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
•
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #4
video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #5
image question answering
CNN
Answer
(text)
RNN
Merge
Image
(photo)
Question
(text)
RNN
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #5
image question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #6
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #6
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Architecture patterns
Neural Network Design Pattern #7
image to image translation
CNN
Image
(text)
Image
(photo)
CNN
Features of
the image
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #7
image to image translation
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #7
image to image translation
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #7
image to image translation
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Neural Network Design Pattern #7
image to image translation
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
Can a neural network output 2 types
of output?
• How to architect Multi output network?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
Multi output network
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
how to extract features from multiple
sources ?
• I have a input of audio and video and text. How to combine the features
from all 3?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Quiz
how to extract features from multiple sources ?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Our inspiration / acknowledgements
Friendly approaches :
1) KERAS.io
François Chollet’s
Book on “Deep Learning with Python”
2) Deeplearning.ai (Coursera.org)
Andrew Ng
3) Udacity Deep Learning Free course
4) Google Machine Learning Course
https://developers.google.com/machine-learning/crash-course/
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
More inspiration/ acknowledgements
Excellent Resources
• Stanford cs231 n
http://cs231n.stanford.edu
• MIT Deep Learning
http://introtodeeplearning.com/
https://deeplearning.mit.edu
• IIT Madras
my classes notes with Prof. Anurag (Deep Learning)
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
What did you learn?
(Day 1)
How to solve a problem with Deep Learning ?
How to design an neural network architecture?
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key ideas to remember
You can map any thing to anything with Deep Learning AB
• Anything  Machine Learns to map  Anything
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key ideas to remember
How to design a architectures to handle various types of A  B
How to design this architecture?
CNN / RNN ?
Fully Connected or RNN ?
Predicted
class
RNN /
Fully connected ?
Merge / Attention ?
Video
Question
A  --------------------------------------------------------------------------------------------------------------------------  B
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key ideas to remember
From architecture to code ? (just learn early glimpse of how code looks like.)
Just 1 page of code is enough for a complex problems such as video question answering
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Key ideas to remember
If you didn’t understand the code*, just don’t worry.
• Why not to worry?
• Solving the video question answering problem was not possible until 2017.
• To solve this problem, it required dozens of experts in Google lab to dedicate 6 months of effort!
*code = code in the previous slide
Acknowledgments & Credits are mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
Team activity
idea-a-ton
Goal of this thinking activity
How to solve a problem with Deep Learning ?
How to design an neural network architecture?

How to architect Deep Learning

  • 1.
    Design How to solvea problem using Deep Learning AI for Good Workshop July 2019 Session 3 https://sites.google.com/view/AIforEveryone
  • 2.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key Learning objective for Day 1 How to solve a problem with Deep Learning ? How to design an neural network architecture?
  • 3.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz #1 Is it possible to map hand signs for deaf  speech  Hand signs   Speech
  • 4.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone A B •
  • 5.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz #2 Is it possible to “see the depth” for visual impaired  Yes / no ?
  • 6.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz #2 Is it possible to feel the depth for visual impaired
  • 7.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
  • 8.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Topics in this presentation •New way to think from today ! • How to use Deep Learning to solve any problem • How to model & architect Neural Networks
  • 9.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone New way to think from today !
  • 10.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone New way to think from today ! Do you have  Large dataset?  Compute resource (can pay easy to cloud)  Ability to use frameworks such as Tensorflow
  • 11.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key Learning objective for Day 1 How to design an neural network architecture?
  • 12.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key Learning objective for Day 1How to design an neural network architecture? How to map x to y? What architecture can connect x to y?
  • 13.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key Learning objective for Day 1How to design an neural network architecture? Predicted class How to map x to y? What architecture can map x to y?
  • 14.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Deep Learning idea-a-thon: How to solve a problem by designing novel neural network architectures? CNN RNN Fully Connected Merge CNN is suitable architecture to extract features out of any data that has spatial relationships. For example, An photo image spatial info..or info spread across space In general, a recurrent neural network could be considered as the best neural network model for extracting features out of temporal data. A temporal data is basically a data that varies over time such as ECG, music, speech, sentences, words. LSTM is a advanced RNN To classify / predict To combine the learnings of two neural networks. (to combine the power of 2 people’s brains) Attention To learn to focus on the most important aspects to achieve a particular task Generative Deep Learning To model creative tasks as to compose a music or to come up a painting
  • 15.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone How to use Deep learning to learn to connect A and B
  • 16.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learning objective: How to learn to map A  B Deep Learning A B Rules or math function to find B given A. B = f(A) What should be f()?
  • 17.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone How to translate A to B
  • 18.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
  • 19.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learn to map Xy • Examples of x, y
  • 20.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learn to map Xy
  • 21.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learn to map A  B
  • 22.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone
  • 23.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Fun activity
  • 24.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learning objective: How do solve a problem using Deep Learning? Case study: self driving car
  • 25.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key Learning objective for Day 1 How to solve a problem with Deep Learning ? How to design an neural network architecture?
  • 26.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #1 classification problems CNN Fully Connected Prediction of what of objects is in this photo
  • 27.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #1 CNN Fully Connected Turn car left Turn car right Don’t Turn the car
  • 28.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Learn to map x->y
  • 29.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #1
  • 30.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #1
  • 31.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Camera view ? steering angle • Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car Credits: Nvidia
  • 32.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz what the neural network learns depends upon its purpose • How to set the purpose of a “learning machine”?
  • 33.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone •
  • 34.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone What is the function to map x->y •
  • 35.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #2 Regression problems CNN Fully Connected Continuous number
  • 36.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #2 Regression problems CNN Fully Connected Turn the car by 10 degrees
  • 37.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #2 Regression problems
  • 38.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #2 Regression problems •
  • 39.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #3 multimodal classification CNN Fully Connected Predicted class RNN Merge
  • 40.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #3 multimodal classification
  • 41.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #3 multimodal classification
  • 42.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz Neural Network Design Pattern #3 multimodal classification • Quiz: What was x and y in the problem?
  • 43.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz Neural Network Design Pattern #3 • Quiz: What was x and y in the problem?
  • 44.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz multimodal classification
  • 45.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Answering Visual Questions from Blind People
  • 46.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #3 multimodal classification
  • 47.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #3 multimodal classification •
  • 48.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #3
  • 49.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz/Thinking activity how to classify a genomic sequence healthcare/genomic sequence classification
  • 50.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone How to Solve this problem with deep learning: how to classify a genomic sequence ? ? ? Credits: DeepMotif (ICLR’16) http://www.cs.virginia.edu/yanjun/paperA14/2017_demo_slides.pdf This is a classification problem. A: Sequence of characters B: Category
  • 51.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone How to Solve this problem with deep learning: how to classify a genomic sequence ? ? ?
  • 52.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone how to classify a genomic sequence
  • 53.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #4 video question answering CNN Fully Connected Predicted class RNN Merge Video Question RNN
  • 54.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #4 video question answering
  • 55.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering •
  • 56.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering
  • 57.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering
  • 58.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering
  • 59.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering
  • 60.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #4 video question answering
  • 61.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #5 image question answering CNN Answer (text) RNN Merge Image (photo) Question (text) RNN
  • 62.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #5 image question answering
  • 63.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #6
  • 64.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #6
  • 65.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Architecture patterns Neural Network Design Pattern #7 image to image translation CNN Image (text) Image (photo) CNN Features of the image
  • 66.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #7 image to image translation
  • 67.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #7 image to image translation
  • 68.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #7 image to image translation
  • 69.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Neural Network Design Pattern #7 image to image translation
  • 70.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz Can a neural network output 2 types of output? • How to architect Multi output network?
  • 71.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz Multi output network
  • 72.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz how to extract features from multiple sources ? • I have a input of audio and video and text. How to combine the features from all 3?
  • 73.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Quiz how to extract features from multiple sources ?
  • 74.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Our inspiration / acknowledgements Friendly approaches : 1) KERAS.io François Chollet’s Book on “Deep Learning with Python” 2) Deeplearning.ai (Coursera.org) Andrew Ng 3) Udacity Deep Learning Free course 4) Google Machine Learning Course https://developers.google.com/machine-learning/crash-course/
  • 75.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone More inspiration/ acknowledgements Excellent Resources • Stanford cs231 n http://cs231n.stanford.edu • MIT Deep Learning http://introtodeeplearning.com/ https://deeplearning.mit.edu • IIT Madras my classes notes with Prof. Anurag (Deep Learning)
  • 76.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone What did you learn? (Day 1) How to solve a problem with Deep Learning ? How to design an neural network architecture?
  • 77.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key ideas to remember You can map any thing to anything with Deep Learning AB • Anything  Machine Learns to map  Anything
  • 78.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key ideas to remember How to design a architectures to handle various types of A  B How to design this architecture? CNN / RNN ? Fully Connected or RNN ? Predicted class RNN / Fully connected ? Merge / Attention ? Video Question A  --------------------------------------------------------------------------------------------------------------------------  B
  • 79.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key ideas to remember From architecture to code ? (just learn early glimpse of how code looks like.) Just 1 page of code is enough for a complex problems such as video question answering
  • 80.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Key ideas to remember If you didn’t understand the code*, just don’t worry. • Why not to worry? • Solving the video question answering problem was not possible until 2017. • To solve this problem, it required dozens of experts in Google lab to dedicate 6 months of effort! *code = code in the previous slide
  • 81.
    Acknowledgments & Creditsare mentioned to inspirational resources presented in the end of presentation. For more like this, https://sites.google.com/view/AIforEveryone Team activity idea-a-ton Goal of this thinking activity How to solve a problem with Deep Learning ? How to design an neural network architecture?