SlideShare a Scribd company logo

convolutional neural network and its applications.pdf

•
S
SubhamKumar3239

The document describes how convolutional neural networks (CNNs) work. It explains that CNNs help reduce the number of parameters in a neural network by sharing weights and using small, local filters that are applied across the input. It provides examples of how CNNs can be used for image recognition by having filters detect local patterns, as well as applications in speech recognition and text classification.

Data & Analytics
1 of 28
Download to read offline
1
2
Most read
3
4
5
Most read
6
7
8
9
10
11
Most read
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Smaller Network: CNN ⚫ We know it is good to learn a small model. ⚫ From this fully connected model, do we really need all the edges? ⚫ Can some of these be shared?
Consider learning an image: ⚫Some patterns are much smaller than the whole image “beak” detector Can represent a small region with fewer parameters
Same pattern appears in different places: They can be compressed! What about training a lot of such “small” detectors and each detector must “move around”. “upper-left beak” detector “middle beak” detector They can be compressed to the same parameters.
A convolutional layer A filter A CNN is a neural network with some convolutional layers (and some other layers). A convolutional layer has a number of filters that does convolutional operation. Beak detector
Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 … … These are the network parameters to be learned. Each filter detects a small pattern (3 x 3).
Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -1 stride=1 Dot product
Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -3 If stride=2
Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 stride=1
Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 -1 -1 -1 -1 -1 -1 -2 1 -1 -1 -2 1 -1 0 -4 3 Repeat this for each filter stride=1 Two 4 x 4 images Forming 2 x 4 x 4 matrix Feature Map
Color image: RGB 3 channels 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 1 -1 -1 -1 1 -1 -1 -1 1 1 -1 -1 -1 1 -1 -1 -1 1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 Color image
1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 image convolution -1 1 -1 -1 1 -1 -1 1 -1 1 -1 -1 -1 1 -1 -1 -1 1 1 x 2 x … … 36 x … … 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 Convolution v.s. Fully Connected Fully- connected
1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 1 2 3 … 8 9 … 1 3 14 15 … Only connect to 9 inputs, not fully connected 4: 10: 16 1 0 0 0 0 1 0 0 0 0 1 1 3 fewer parameters!
1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 1: 2: 3: … 7: 8: 9: … 1 3: 14: 15: … 4: 10: 16: 1 0 0 0 0 1 0 0 0 0 1 1 3 -1 Shared weights 6 x 6 image Fewer parameters Even fewer parameters
The whole CNN Fully Connected Feedforward network cat dog …… Convolution Max Pooling Convolution Max Pooling Flattened Can repeat many times
Max Pooling 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 -1 -1 -1 -1 -1 -1 -2 1 -1 -1 -2 1 -1 0 -4 3 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1
Why Pooling ⚫ Subsampling pixels will not change the object Subsampling bird bird We can subsample the pixels to make image smaller fewer parameters to characterize the image
A CNN compresses a fully connected network in two ways: ⚫Reducing number of connections ⚫Shared weights on the edges ⚫Max pooling further reduces the complexity
Max Pooling 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 3 0 1 3 -1 1 3 0 2 x 2 image Each filter is a channel New image but smaller Conv Max Pooling
The whole CNN Convolution Max Pooling Convolution Max Pooling Can repeat many times A new image The number of channels is the number of filters Smaller than the original image 3 0 1 3 -1 1 3 0
The whole CNN Fully Connected Feedforward network cat dog …… Convolution Max Pooling Convolution Max Pooling Flattened A new image A new image
Flattening 3 0 1 3 -1 1 3 0 Flattened 3 0 1 3 -1 1 0 3 Fully Connected Feedforward network
Only modified the network structure and input format (vector -> 3-D tensor) CNN in Keras Convolution Max Pooling Convolution Max Pooling input 1 -1 -1 -1 1 -1 -1 -1 1 -1 1 -1 -1 1 -1 -1 1 -1 There are 25 3x3 filters. … … Input_shape = ( 28 , 28 , 1) 1: black/white, 3: RGB 28 x 28 pixels 3 -1 -3 1 3
Only modified the network structure and input format (vector -> 3-D array) CNN in Keras Convolution Max Pooling Convolution Max Pooling Input 1 x 28 x 28 25 x 26 x 26 25 x 13 x 13 50 x 11 x 11 50 x 5 x 5 How many parameters for each filter? How many parameters for each filter? 9 225= 25x9
Only modified the network structure and input format (vector -> 3-D array) CNN in Keras Convolution Max Pooling Convolution Max Pooling Input 1 x 28 x 28 25 x 26 x 26 25 x 13 x 13 50 x 11 x 11 50 x 5 x 5 Flattened 1250 Fully connected feedforward network Output
AlphaGo Neural Network (19 x 19 positions) Next move 19 x 19 matrix Black: 1 white: -1 none: 0 Fully-connected feedforward network can be used But CNN performs much better
AlphaGo’s policy network Note: AlphaGo does not use Max Pooling. The following is quotation from their Nature article:
CNN in speech recognition Time Frequency Spectrogram CNN Image The filters move in the frequency direction.
CNN in text classification Source of image: http://citeseerx.ist.psu.edu/viewdoc/downlo ad?doi=10.1.1.703.6858&rep=rep1&type=p df ?

More Related Content

PPT
Convolutional Neural Networks definicion y otros
ssuserf85a91
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
EngineeringTamilan
 
PPT
Deep-Learning
Amnaalia
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
archn4
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
AshishKumarSingh176
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
AminHa5
 
PPT
Deep-Learning presnetation by someone from the internet
clash12hero
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
kundurti
 
Convolutional Neural Networks definicion y otros
ssuserf85a91
 
Deep-Learning-2017-Lecture5CNN.ppt
EngineeringTamilan
 
Deep-Learning
Amnaalia
 
Deep-Learning-2017-Lecture5CNN.ppt
archn4
 
Deep-Learning-2017-Lecture5CNN.ppt
AshishKumarSingh176
 
Deep-Learning-2017-Lecture5CNN.ppt
AminHa5
 
Deep-Learning presnetation by someone from the internet
clash12hero
 
Deep-Learning-2017-Lecture5CNN.ppt
kundurti
 

Similar to convolutional neural network and its applications.pdf (20)

PPT
Deep learning-2017-lecture5 cnn
AnandShinde47
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
SaadMemon23
 
PPT
Deep Learning Techniques like CNN and RNN
SumaiyaSk
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
sghorai
 
PPT
Deep learning-smaller neural network
sonykhan3
 
PPT
Deep-Learning-Convolutional Neural Networks and Sequence Modeling.ppt
PraveenVundrajavarap
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
rohithprabhas1
 
PPT
Deep-Learning-2017-Lecture5CNN.ppt
sruthiksanalkumar
 
PPTX
Deep-LearningwithVisualExamplesExplaine.pptx
ansarinazish958
 
PPTX
Deep-Learning-2017-Lecture5CNN.pptx
Dr. Radhey Shyam
 
PPT
Deep Learning approach in Machine learning
vipulkondekar
 
PPT
Introduction to Deep-Learning-CNN Arch.ppt
khandarevaibhav
 
PPT
digital image processing - convolutional networks
Muhammad824617
 
PDF
AI_Theory: Covolutional_neuron_network.pdf
21146290
 
PPTX
Machine learning algorithms like CNN and LSTM
monihareni
 
PPT
Adv.TopicsAICNN.ppt
ZeeshanAhmad449947
 
PDF
convolutional neural networks for machine learning
omogire08
 
PDF
cnn.pdf
AKANKSHADIXIT52
 
PDF
Practical Deep Learning Using Tensor Flow - Sandeep Kath
Sandeep Kath
 
PPTX
Deep learning in E-Commerce Applications and Challenges (CNN)
Houda Bakir
 
Deep learning-2017-lecture5 cnn
AnandShinde47
 
Deep-Learning-2017-Lecture5CNN.ppt
SaadMemon23
 
Deep Learning Techniques like CNN and RNN
SumaiyaSk
 
Deep-Learning-2017-Lecture5CNN.ppt
sghorai
 
Deep learning-smaller neural network
sonykhan3
 
Deep-Learning-Convolutional Neural Networks and Sequence Modeling.ppt
PraveenVundrajavarap
 
Deep-Learning-2017-Lecture5CNN.ppt
rohithprabhas1
 
Deep-Learning-2017-Lecture5CNN.ppt
sruthiksanalkumar
 
Deep-LearningwithVisualExamplesExplaine.pptx
ansarinazish958
 
Deep-Learning-2017-Lecture5CNN.pptx
Dr. Radhey Shyam
 
Deep Learning approach in Machine learning
vipulkondekar
 
Introduction to Deep-Learning-CNN Arch.ppt
khandarevaibhav
 
digital image processing - convolutional networks
Muhammad824617
 
AI_Theory: Covolutional_neuron_network.pdf
21146290
 
Machine learning algorithms like CNN and LSTM
monihareni
 
Adv.TopicsAICNN.ppt
ZeeshanAhmad449947
 
convolutional neural networks for machine learning
omogire08
 
cnn.pdf
AKANKSHADIXIT52
 
Practical Deep Learning Using Tensor Flow - Sandeep Kath
Sandeep Kath
 
Deep learning in E-Commerce Applications and Challenges (CNN)
Houda Bakir
 
Ad

Recently uploaded (20)

PPTX
Data Security Breach: Immediate Action Plan
varmabhuvan266
 
PPTX
Introduction to computer chapter one 2017.pptx
mensunmarley
 
PDF
Chad Readey - An Independent Thinker
Chad Readey
 
PPT
Real Life Application of Set theory, Relations and Functions
manavparmar205
 
PPTX
Introduction-to-Python-Programming-Language (1).pptx
dhyeysapariya
 
PPTX
Databricks-DE-Associate Certification Questions-june-2024.pptx
pedelli41
 
PPTX
Data-Driven Machine Learning for Rail Infrastructure Health Monitoring
Sione Palu
 
PPTX
White Blue Simple Modern Enhancing Sales Strategy Presentation_20250724_21093...
RamNeymarjr
 
PPTX
short term internship project on Data visualization
JMJCollegeComputerde
 
PPTX
Power BI in Business Intelligence with AI
KPR Institute of Engineering and Technology
 
PDF
Company Presentation pada Perusahaan ADB.pdf
didikfahmi
 
PDF
Blue Futuristic Cyber Security Presentation.pdf
tanvikhunt1003
 
PPTX
INFO8116 - Week 10 - Slides.pptx data analutics
guddipatel10
 
PPTX
Introduction to Data Analytics and Data Science
KavithaCIT
 
PPTX
short term project on AI Driven Data Analytics
JMJCollegeComputerde
 
PPTX
World-population.pptx fire bunberbpeople
umutunsalnsl4402
 
PDF
blockchain123456789012345678901234567890
tanvikhunt1003
 
PDF
The_Future_of_Data_Analytics_by_CA_Suvidha_Chaplot_UPDATED.pdf
CA Suvidha Chaplot
 
PPTX
INFO8116 -Big data architecture and analytics
guddipatel10
 
PPTX
Pipeline Automatic Leak Detection for Water Distribution Systems
Sione Palu
 
Data Security Breach: Immediate Action Plan
varmabhuvan266
 
Introduction to computer chapter one 2017.pptx
mensunmarley
 
Chad Readey - An Independent Thinker
Chad Readey
 
Real Life Application of Set theory, Relations and Functions
manavparmar205
 
Introduction-to-Python-Programming-Language (1).pptx
dhyeysapariya
 
Databricks-DE-Associate Certification Questions-june-2024.pptx
pedelli41
 
Data-Driven Machine Learning for Rail Infrastructure Health Monitoring
Sione Palu
 
White Blue Simple Modern Enhancing Sales Strategy Presentation_20250724_21093...
RamNeymarjr
 
short term internship project on Data visualization
JMJCollegeComputerde
 
Power BI in Business Intelligence with AI
KPR Institute of Engineering and Technology
 
Company Presentation pada Perusahaan ADB.pdf
didikfahmi
 
Blue Futuristic Cyber Security Presentation.pdf
tanvikhunt1003
 
INFO8116 - Week 10 - Slides.pptx data analutics
guddipatel10
 
Introduction to Data Analytics and Data Science
KavithaCIT
 
short term project on AI Driven Data Analytics
JMJCollegeComputerde
 
World-population.pptx fire bunberbpeople
umutunsalnsl4402
 
blockchain123456789012345678901234567890
tanvikhunt1003
 
The_Future_of_Data_Analytics_by_CA_Suvidha_Chaplot_UPDATED.pdf
CA Suvidha Chaplot
 
INFO8116 -Big data architecture and analytics
guddipatel10
 
Pipeline Automatic Leak Detection for Water Distribution Systems
Sione Palu
 
Ad

convolutional neural network and its applications.pdf

  • 1. Smaller Network: CNN ⚫ We know it is good to learn a small model. ⚫ From this fully connected model, do we really need all the edges? ⚫ Can some of these be shared?
  • 2. Consider learning an image: ⚫Some patterns are much smaller than the whole image “beak” detector Can represent a small region with fewer parameters
  • 3. Same pattern appears in different places: They can be compressed! What about training a lot of such “small” detectors and each detector must “move around”. “upper-left beak” detector “middle beak” detector They can be compressed to the same parameters.
  • 4. A convolutional layer A filter A CNN is a neural network with some convolutional layers (and some other layers). A convolutional layer has a number of filters that does convolutional operation. Beak detector
  • 5. Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 … … These are the network parameters to be learned. Each filter detects a small pattern (3 x 3).
  • 6. Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -1 stride=1 Dot product
  • 7. Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -3 If stride=2
  • 8. Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 stride=1
  • 9. Convolution 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 -1 -1 -1 -1 -1 -1 -2 1 -1 -1 -2 1 -1 0 -4 3 Repeat this for each filter stride=1 Two 4 x 4 images Forming 2 x 4 x 4 matrix Feature Map
  • 10. Color image: RGB 3 channels 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 1 -1 -1 -1 1 -1 -1 -1 1 1 -1 -1 -1 1 -1 -1 -1 1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 -1 1 -1 Color image
  • 11. 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 image convolution -1 1 -1 -1 1 -1 -1 1 -1 1 -1 -1 -1 1 -1 -1 -1 1 1 x 2 x … … 36 x … … 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 Convolution v.s. Fully Connected Fully- connected
  • 12. 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 1 2 3 … 8 9 … 1 3 14 15 … Only connect to 9 inputs, not fully connected 4: 10: 16 1 0 0 0 0 1 0 0 0 0 1 1 3 fewer parameters!
  • 13. 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1 1: 2: 3: … 7: 8: 9: … 1 3: 14: 15: … 4: 10: 16: 1 0 0 0 0 1 0 0 0 0 1 1 3 -1 Shared weights 6 x 6 image Fewer parameters Even fewer parameters
  • 14. The whole CNN Fully Connected Feedforward network cat dog …… Convolution Max Pooling Convolution Max Pooling Flattened Can repeat many times
  • 15. Max Pooling 3 -1 -3 -1 -3 1 0 -3 -3 -3 0 1 3 -2 -2 -1 -1 1 -1 -1 1 -1 -1 1 -1 Filter 2 -1 -1 -1 -1 -1 -1 -2 1 -1 -1 -2 1 -1 0 -4 3 1 -1 -1 -1 1 -1 -1 -1 1 Filter 1
  • 16. Why Pooling ⚫ Subsampling pixels will not change the object Subsampling bird bird We can subsample the pixels to make image smaller fewer parameters to characterize the image
  • 17. A CNN compresses a fully connected network in two ways: ⚫Reducing number of connections ⚫Shared weights on the edges ⚫Max pooling further reduces the complexity
  • 18. Max Pooling 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 6 x 6 image 3 0 1 3 -1 1 3 0 2 x 2 image Each filter is a channel New image but smaller Conv Max Pooling
  • 19. The whole CNN Convolution Max Pooling Convolution Max Pooling Can repeat many times A new image The number of channels is the number of filters Smaller than the original image 3 0 1 3 -1 1 3 0
  • 20. The whole CNN Fully Connected Feedforward network cat dog …… Convolution Max Pooling Convolution Max Pooling Flattened A new image A new image
  • 21. Flattening 3 0 1 3 -1 1 3 0 Flattened 3 0 1 3 -1 1 0 3 Fully Connected Feedforward network
  • 22. Only modified the network structure and input format (vector -> 3-D tensor) CNN in Keras Convolution Max Pooling Convolution Max Pooling input 1 -1 -1 -1 1 -1 -1 -1 1 -1 1 -1 -1 1 -1 -1 1 -1 There are 25 3x3 filters. … … Input_shape = ( 28 , 28 , 1) 1: black/white, 3: RGB 28 x 28 pixels 3 -1 -3 1 3
  • 23. Only modified the network structure and input format (vector -> 3-D array) CNN in Keras Convolution Max Pooling Convolution Max Pooling Input 1 x 28 x 28 25 x 26 x 26 25 x 13 x 13 50 x 11 x 11 50 x 5 x 5 How many parameters for each filter? How many parameters for each filter? 9 225= 25x9
  • 24. Only modified the network structure and input format (vector -> 3-D array) CNN in Keras Convolution Max Pooling Convolution Max Pooling Input 1 x 28 x 28 25 x 26 x 26 25 x 13 x 13 50 x 11 x 11 50 x 5 x 5 Flattened 1250 Fully connected feedforward network Output
  • 25. AlphaGo Neural Network (19 x 19 positions) Next move 19 x 19 matrix Black: 1 white: -1 none: 0 Fully-connected feedforward network can be used But CNN performs much better
  • 26. AlphaGo’s policy network Note: AlphaGo does not use Max Pooling. The following is quotation from their Nature article:
  • 27. CNN in speech recognition Time Frequency Spectrogram CNN Image The filters move in the frequency direction.
  • 28. CNN in text classification Source of image: http://citeseerx.ist.psu.edu/viewdoc/downlo ad?doi=10.1.1.703.6858&rep=rep1&type=p df ?