Computational Photography Si Lu Spring 2018 - - PowerPoint PPT Presentation

▶
computational photography
SMART_READER_LITE
LIVE PREVIEW

Computational Photography Si Lu Spring 2018 - - PowerPoint PPT Presentation

Jun 14, 2023 477 likes •1.35k views

Computational Photography Si Lu Spring 2018 http://web.cecs.pdx.edu/~lusi/CS510/CS510_Computati onal_Photography.htm 05/29/2018 Last Time o 3D Video Stabilization 2 Introduction of Neural Networks 3 Content Introduction

slide-1
SLIDE 1

Computational Photography

Si Lu

Spring 2018

http://web.cecs.pdx.edu/~lusi/CS510/CS510_Computati

  • nal_Photography.htm

05/29/2018

slide-2
SLIDE 2

Last Time

  • 3D Video Stabilization

2

slide-3
SLIDE 3

Introduction of Neural Networks

3

slide-4
SLIDE 4

Content

4

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks?
  • Neural Network-Basic
  • What is in the black box
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-5
SLIDE 5

Content

5

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks?
  • Neural Network-Basic
  • What is in the black box
  • Gradient Descent Method
  • Different Optimizors
  • Activation Function
  • Morden Neural Networks
slide-6
SLIDE 6

Artificial vs. Biological Neural Nets

http://7xiur2.com1.z0.glb.clouddn.com/0137.png

slide-7
SLIDE 7

7

https://cdn-images-1.medium.com/max/1600/1*xR4m0oOKz_jRgQU4Oge53g.jpeg

slide-8
SLIDE 8

8

https://cdn-images-1.medium.com/max/1600/1*xR4m0oOKz_jRgQU4Oge53g.jpeg

slide-9
SLIDE 9

9

https://cdn-images-1.medium.com/max/1600/1*xR4m0oOKz_jRgQU4Oge53g.jpeg https://www.robotics.org/

slide-10
SLIDE 10

10

https://cdn-images-1.medium.com/max/1600/1*xR4m0oOKz_jRgQU4Oge53g.jpeg https://www.geekwire.com

slide-11
SLIDE 11

11

Artificial = Biological ? 90 Billion Neuron

slide-12
SLIDE 12

Artificial = Biological ?

slide-13
SLIDE 13

Artificial = Biological ?

slide-14
SLIDE 14

Artificial = Biological ?

slide-15
SLIDE 15

Artificial = Biological ?

slide-16
SLIDE 16

Artificial = Biological ?

slide-17
SLIDE 17

Artificial = Biological ?

slide-18
SLIDE 18

Artificial = Biological ?

slide-19
SLIDE 19

Artificial = Biological ?

slide-20
SLIDE 20

Artificial = Biological ?

slide-21
SLIDE 21

Artificial = Biological ?

slide-22
SLIDE 22

Artificial = Biological ?

slide-23
SLIDE 23

Artificial = Biological ?

=

slide-24
SLIDE 24

Content

24

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-25
SLIDE 25

What are Neural Networks?

https://i.imgur.com/Vbsk7t5.jpg
slide-26
SLIDE 26

What are Neural Networks?

Neural network, or artificial neural network, is a computing system inspired by the biological neural networks that constitute animal brains

wikipedia

slide-27
SLIDE 27

What are Neural Networks?

wikipedia

slide-28
SLIDE 28

What are Neural Networks?

wikipedia

Input Hidden layer 1 Hidden layer 2 Output

slide-29
SLIDE 29

What are Neural Networks?

slide-30
SLIDE 30

What are Neural Networks?

slide-31
SLIDE 31

What are Neural Networks?

slide-32
SLIDE 32

What are Neural Networks?

Trainging dataset Ground truth labels

slide-33
SLIDE 33

What are Neural Networks?

slide-34
SLIDE 34

What are Neural Networks?

slide-35
SLIDE 35

What are Neural Networks?

slide-36
SLIDE 36

What are Neural Networks?

Error

slide-37
SLIDE 37

What are Neural Networks?

Error

slide-38
SLIDE 38

What are Neural Networks?

Repeat millions of times

slide-39
SLIDE 39

What are Neural Networks?

slide-40
SLIDE 40

What are Neural Networks?

slide-41
SLIDE 41

Content

41

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
  • Overfitting-Batch Normalization, Dropout
  • From LeNet to ResNet
slide-42
SLIDE 42

Neural Network Basics

https://futureoflife.org
slide-43
SLIDE 43

Content

43

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-44
SLIDE 44

What is in the black box ?

slide-45
SLIDE 45

What is in the black box ?

slide-46
SLIDE 46

What is in the black box ?

slide-47
SLIDE 47

What is in the black box ?

slide-48
SLIDE 48

What is in the black box ?

slide-49
SLIDE 49

What is in the black box ?

Neural networks are trained to extract higher and higher levels of abstract features to better represent the dataset via back-propogation

slide-50
SLIDE 50

What is in the black box ?

Σ AF

w1 w2 w3 x1 x2 x3 y

Basic Unit: Neuron

input parameters

  • utput
slide-51
SLIDE 51

What is in the black box ?

Σ AF

w1 w2 w3 x1 x2 x3 y

y = w1x1+w2x2+w3x3

input parameters

  • utput
slide-52
SLIDE 52

What is in the black box ?

Σ AF

w1 w2 w3 x1 x2 x3 y

y = WxT

input parameters

  • utput
slide-53
SLIDE 53

Content

53

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-54
SLIDE 54

Gradient Descent Method

Σ AF

w1 w2 w3 x1 x2 x3 y

y = WxT

input parameters

  • utput
slide-55
SLIDE 55

Gradient Descent Method

y = WxT

  • Q1. What is the optimal W?
  • Q2. How to obtain the optimal W?
slide-56
SLIDE 56

Gradient Descent Method

Wopt = argminW|y-WxT|2

  • Q1. What is the optimal W?
  • Q2. How to obtain the optimal W?
slide-57
SLIDE 57

Gradient Descent Method

Wopt = argminW|y-WxT|2

  • Q1. What is the optimal W?
  • Q2. How to obtain the optimal W?

loss

slide-58
SLIDE 58

Gradient Descent Method

Optimization: Gradient Descent

  • Q1. What is the optimal W?
  • Q2. How to obtain the optimal W?
slide-59
SLIDE 59

Gradient Descent Method

Simplification

Original loss function: f=|y-WxT|2

slide-60
SLIDE 60

Gradient Descent Method

Simplification

Original loss function: f=|y-WxT|2 Simplify 1: single w/x/y: f=(y-wx)2

slide-61
SLIDE 61

Gradient Descent Method

Simplification

Original loss function: f=|y-WxT|2 Simplify 1: single w/x/y: f=(y-wx)2 Simplify 2: y=0, x=1: f=w2

slide-62
SLIDE 62

Gradient Descent Method

f w

slide-63
SLIDE 63

Gradient Descent Method

f w

slide-64
SLIDE 64

Gradient Descent Method

f w

slide-65
SLIDE 65

Gradient Descent Method

slide-66
SLIDE 66

Gradient Descent Method

slide-67
SLIDE 67

Gradient Descent Method

f w

slide-68
SLIDE 68

Content

68

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-69
SLIDE 69

Speed up Training: dataset

Large numbe of x, y, w

y = WxT

slide-70
SLIDE 70

Speed up Training: dataset

slide-71
SLIDE 71

Speed up Training: dataset

Batch

slide-72
SLIDE 72

Speed up Training: dataset

Batch

slide-73
SLIDE 73

Speed up Training: dataset

Batch

Stochastic Gradient Descent (SGD)

slide-74
SLIDE 74

Speed up Training: optimizer

Original: W += - LR * dx

slide-75
SLIDE 75

Speed up Training: optimizer

Original: W += - LR * dx m += b1*m- LR * dx W += m Momentum: Adding “dowhill”- inertia

slide-76
SLIDE 76

Speed up Training: optimizer

Original: W += - LR * dx v += dx^2 W += -LR * dx/sqrt(v) AdaGrad: Adding “breaking shoes”- resistance

slide-77
SLIDE 77

Speed up Training: optimizer

Momentum + AdaGrad

RMSProp Adam (Popular)

slide-78
SLIDE 78

Content

78

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks
slide-79
SLIDE 79

Activation Function

Σ AF

w1 w2 w3 x1 x2 x3 y

y = WxT

input parameters

  • utput
slide-80
SLIDE 80

https://medium.com/@shrutijadon10104776

slide-81
SLIDE 81

Activation Function

slide-82
SLIDE 82

Activation Function

Activate different neurons for different input

slide-83
SLIDE 83

Activation Function

Activate different neurons for different input

slide-84
SLIDE 84

Activation Function

Essentially: adding non-linearty

Team Salary Championship

slide-85
SLIDE 85

Activation Function

Essentially: adding non-linearty

Team Salary Team Salary Championship Championship

slide-86
SLIDE 86

Next Time

86

  • Introduction
  • Artificial vs. Biological Neural Nets
  • What are Neural Networks ?
  • Neural Network-Basic
  • What is in the black box ?
  • Gradient Descent Method
  • Speed up training
  • Activation Function
  • Morden Neural Networks