SlideShare a Scribd company logo

Day 3 take up convolutional neural network

H
HuyPhmNht2

The document appears to be a presentation outlining a workshop on convolutional neural networks (CNNs) including reasons for using CNNs over traditional neural networks, as well as technical details such as matrix representation, layer structure, and image classification challenges. It also includes a Kaggle challenge related to image classification with a registration deadline and rewards for participants. Overall, it aims to educate attendees on CNNs and their application in image recognition tasks.

Engineering
1 of 47
Download to read offline
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
Day 3 take up convolutional neural network
Day 3 take up convolutional neural network
Day 3 take up convolutional neural network
Day 3 take up convolutional neural network
Day 3 take up convolutional neural network
Day 3 take up convolutional neural network
ICEBREAKER Phạm Nguyễn Anh Thư @thu.phamhcmut Let’s get to know each other!
Record shop _____ ___ ____ Thật bất ngờ ___ ____ ____ Em dạo này Tình đắng như ly cà phê ____ _____ ____ _ __ ____ ___ ___ ___ ___ ___ ______ Để Mị nói cho mà nghe ____ ____ ____ ___ __ ____ ____ Đâu cần một bài ca tình yêu
how-to-AI Series: Unlock Potential Day 3: Take UP Convolutional Neural Network Nguyễn Luật Gia Khôi @giakhoi.nguyenluat Nguyễn Thế Bình @binh.nguyen288
Outline 1. Break the ice 2. Convolutional Neural Network 3. Why using CNN instead of NN? 4. Demo code 5. Kaggle Challenge
Convolutional Neural Network
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 -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: black ● 1: white Matrix representation
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 -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 -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 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 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 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 ANN can solve this variety in digits
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 -1 -1 -1 1 1 -1 ... 1 -1 1 1 -1 -1 1 1 x2 x33 x32 x3 x4 x34 x1 x35 ... h2 x33 x32 h3 h4 x34 h1 x35 ... h2 h8 h3 h9 h1 h10 ... 1 9 8 2 3 0 0.01 0.92 0.003 0.008 0.015 0.02
WAIT A SECOND, how about this?
Image size = 1920 x 1080 x 3 Input layer #neurons = 1920 x 1080 x 3 = 6 million Hidden layer #neurons = Say you keep it = 4 million Weights between input and first hidden layer = 6 * 4 = 24 trillion parameters
Day 3 take up convolutional neural network
Disadvantages of ANN in image classification - Computationally expensive. - Treat local pixels same as pixels far apart => no locality concern - Sensitive to location of the object in the image.
How does human recognize images?
Eyes Nose Ears Hand Leg Head Body Koala
Nine Curves Long curve Circular pattern
How can we make computers recognize these patterns?
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 -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 -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 -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 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 Circular pattern filter Diagonal line filter Diagonal line filter
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 -1 -1 -1 1 -1 -1 -1 -1 * = -1 9 -1 -5 1 -3 -3 3 -3 -5 -1 -5 -3 -5 -3 Circular pattern filter -1 x -1 + -1 x -1 + … + -1 x -1 + 1 x 1 = 9
x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter = x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter = x x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter =
x x * Eye filter = * Eye filter = * Eye filter = x x x x
x x * Eye filter = x * Nose filter = x x * Ear filter = * Head filter x = x x
x x X ... X ... h2 x8 h3 x9 h1 x10 ... 0.01 (Giraffe) 0.80 (Koala) 0.003 (Horse) 0.008 (Buffalo) 0.15 (Bear) 0.02 (Squirrel)
Flatten Feature extraction Classification
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 -1 -1 -1 1 -1 -1 -1 -1 * = -1 9 -1 -5 1 -3 -3 3 -3 -5 -1 -5 -3 -5 -3 Circular pattern filter ReLU 0 9 0 0 1 0 0 3 0 0 0 0 0 0 0
Stride & Padding stride = 2, padding = “valid” stride = 1, padding = “same” input output input output
Day 3 take up convolutional neural network
0 9 0 0 0 1 0 0 0 3 0 0 0 0 0 0 0 0 0 0 Max pooling pool_size = (2,2) stride = 2 9 0 3 0 output_shape = math.floor((input_shape - pool_size) / strides) + 1
Benefits of max pooling - Reduce feature map size => Reduce computational cost. - Fewer parameters needed => Reduce overfitting. - Make model tolerant towards variations and distortions.
Complete pipeline Conv + ReLU Pooling Conv + ReLU Pooling w1 x h1x c1 w2 x h2 x c1 w2 x h2 x c2 w3 x h3 x c2 Flatten
Why using CNN instead of NN?
- Can capture spatial features => Connection sparsity => Reduce overfitting x x * Eye filter = x * Nose filter = x x * Ear filter =
0 0 9 0 0 1 0 0 0 3 0 0 0 0 0 0 0 0 0 0 - Translation invariant. 1 9 3 0 0 0 0 0 0 0 9 0 0 3 0 0 0 0 0 0 0 0 0 0 0 9 3 0 Conv Conv Pooling Pooling
- Location invariant feature detection.
- Parameter sharing.
KAHOOT TIME
DEMO CODE
GDSC AI CHALLENGE
● Challenge: Build a model that can automatically classifies images to their correct labels. ● Registration form is now open! ● Timeline: 12/12/2021 - 30/12/2021 ● Rewards: ○ First prize: 1.000.000 VND ○ Second prize: 700.000 VND ○ Third prize: 500.000 VND ○ 2 AI potential prizes: 400.000 VND GDSC AI CHALLENGE
Q&A
Feedback form https://tinyurl.com/37kk6dkw
Thank You!

More Related Content

PDF
Oracle中Sql解析过程
guest4ef2ea7
 
PDF
Free test practice ccat
RiteshChandokh
 
PPT
Trade First Subtraction
Brenda Vaitkevicius/Wayne State Student
 
DOCX
วรรณคดีอันทรงคุณค่า
wichaikraisorn
 
PDF
Yesterday
Yateamo
 
PDF
Deep Learning - STM 6
Tricode (part of Dept)
 
PPTX
Optimization techniques in formulation Development- Plackett Burmann Design a...
D.R. Chandravanshi
 
PPTX
Deep Learning con CNTK by Pablo Doval
Plain Concepts
 
Oracle中Sql解析过程
guest4ef2ea7
 
Free test practice ccat
RiteshChandokh
 
Trade First Subtraction
Brenda Vaitkevicius/Wayne State Student
 
วรรณคดีอันทรงคุณค่า
wichaikraisorn
 
Yesterday
Yateamo
 
Deep Learning - STM 6
Tricode (part of Dept)
 
Optimization techniques in formulation Development- Plackett Burmann Design a...
D.R. Chandravanshi
 
Deep Learning con CNTK by Pablo Doval
Plain Concepts
 

Similar to Day 3 take up convolutional neural network (20)

PPTX
DL_Lecture_29_06_2020.pptx
AmlanChakrabarti3
 
PDF
convolutional neural networks for machine learning
omogire08
 
PPTX
How the convolutional neural network works behind the hood.
USMANKHAN73908
 
PPTX
Machine learning algorithms like CNN and LSTM
monihareni
 
PDF
Lecture 4: How it Works: Convolutional Neural Networks
Mohamed Loey
 
PDF
Lecture_15_General _Schmidt _factor_.pdf
AmanpreetKaur617053
 
PDF
Digital Image Processing
Sofian Hadiwijaya
 
PDF
how CNN works for tech Every parts introductions.pdf
JunZhao68
 
PDF
cnn.pdf
AKANKSHADIXIT52
 
PDF
CNN Basics.pdf
NiharikaThakur32
 
PPTX
Convolutional Neural Networks
Ashray Bhandare
 
PPTX
Desing of Factorial Experiments
Ramon Saviñon
 
PPTX
Estado del Arte de la IA
Plain Concepts
 
PDF
DeepLearning with Neon
Sofian Hadiwijaya
 
PDF
How to caculus
Kike García Manyari
 
PDF
Your Town
Amie Whitney
 
PDF
Dia 2 elaboramos el plan de alimentación para los cuyes.
ChioGarcia4
 
PDF
1.pdf
Dhiraj Bhaskar
 
PPTX
Deep-Learning-2017-Lecture5CNN.pptx
Dr. Radhey Shyam
 
PPTX
Deep-LearningwithVisualExamplesExplaine.pptx
ansarinazish958
 
DL_Lecture_29_06_2020.pptx
AmlanChakrabarti3
 
convolutional neural networks for machine learning
omogire08
 
How the convolutional neural network works behind the hood.
USMANKHAN73908
 
Machine learning algorithms like CNN and LSTM
monihareni
 
Lecture 4: How it Works: Convolutional Neural Networks
Mohamed Loey
 
Lecture_15_General _Schmidt _factor_.pdf
AmanpreetKaur617053
 
Digital Image Processing
Sofian Hadiwijaya
 
how CNN works for tech Every parts introductions.pdf
JunZhao68
 
cnn.pdf
AKANKSHADIXIT52
 
CNN Basics.pdf
NiharikaThakur32
 
Convolutional Neural Networks
Ashray Bhandare
 
Desing of Factorial Experiments
Ramon Saviñon
 
Estado del Arte de la IA
Plain Concepts
 
DeepLearning with Neon
Sofian Hadiwijaya
 
How to caculus
Kike García Manyari
 
Your Town
Amie Whitney
 
Dia 2 elaboramos el plan de alimentación para los cuyes.
ChioGarcia4
 
1.pdf
Dhiraj Bhaskar
 
Deep-Learning-2017-Lecture5CNN.pptx
Dr. Radhey Shyam
 
Deep-LearningwithVisualExamplesExplaine.pptx
ansarinazish958
 
Ad

More from HuyPhmNht2 (7)

PPTX
Slide.pptx
HuyPhmNht2
 
PDF
Introduction to figma
HuyPhmNht2
 
PDF
Info seminar 2021
HuyPhmNht2
 
PDF
Day 5
HuyPhmNht2
 
PDF
Day 4
HuyPhmNht2
 
PDF
Day 2 build up your own neural network
HuyPhmNht2
 
PDF
Day 1 wazz up ai
HuyPhmNht2
 
Slide.pptx
HuyPhmNht2
 
Introduction to figma
HuyPhmNht2
 
Info seminar 2021
HuyPhmNht2
 
Day 5
HuyPhmNht2
 
Day 4
HuyPhmNht2
 
Day 2 build up your own neural network
HuyPhmNht2
 
Day 1 wazz up ai
HuyPhmNht2
 
Ad

Recently uploaded (20)

PDF
Automation-in-Manufacturing-Chapter-Introduction.pdf
pcmth867
 
PDF
Operating System & Kernel Study Guide-1 - converted.pdf
mranoninvisib16
 
DOCX
ASol_English-Language-Literature-Set-1-27-02-2023-converted.docx
AYUSHMANAV
 
PDF
Evaluating the Democratization of the Turkish Armed Forces from a Normative P...
jpsjournal1
 
PPTX
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
ShriNRPrasad
 
PPTX
Geodesy 1.pptx...............................................
abhi1361yadav
 
PDF
Well-logging-methods_new................
ZafriFarid
 
PPT
Project quality management in manufacturing
BarMusaTetik
 
PDF
PPT on Performance Review to get promotions
prashant593122
 
PPTX
CH1 Production IntroductoryConcepts.pptx
RacemMellouli1
 
PDF
July 2025 - Top 10 Read Articles in International Journal of Software Enginee...
sebastianku31
 
PDF
Mitigating Risks through Effective Management for Enhancing Organizational Pe...
IJAEMSJORNAL
 
PPTX
KTU 2019 -S7-MCN 401 MODULE 2-VINAY.pptx
VinayB68
 
PPTX
IOT PPTs Week 10 Lecture Material.pptx of NPTEL Smart Cities contd
ssusera400e8
 
PPTX
Construction Project Organization Group 2.pptx
vj5agdales
 
PPTX
UNIT 4 Total Quality Management .pptx
gokuld13012005
 
DOCX
573137875-Attendance-Management-System-original
AYUSHMANAV
 
PDF
R24 SURVEYING LAB MANUAL for civil enggi
MNANDITHACIVILSTAFF
 
PPT
CRASH COURSE IN ALTERNATIVE PLUMBING CLASS
rene militante
 
PPTX
Sustainable Sites - Green Building Construction
mhdumerali
 
Automation-in-Manufacturing-Chapter-Introduction.pdf
pcmth867
 
Operating System & Kernel Study Guide-1 - converted.pdf
mranoninvisib16
 
ASol_English-Language-Literature-Set-1-27-02-2023-converted.docx
AYUSHMANAV
 
Evaluating the Democratization of the Turkish Armed Forces from a Normative P...
jpsjournal1
 
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
ShriNRPrasad
 
Geodesy 1.pptx...............................................
abhi1361yadav
 
Well-logging-methods_new................
ZafriFarid
 
Project quality management in manufacturing
BarMusaTetik
 
PPT on Performance Review to get promotions
prashant593122
 
CH1 Production IntroductoryConcepts.pptx
RacemMellouli1
 
July 2025 - Top 10 Read Articles in International Journal of Software Enginee...
sebastianku31
 
Mitigating Risks through Effective Management for Enhancing Organizational Pe...
IJAEMSJORNAL
 
KTU 2019 -S7-MCN 401 MODULE 2-VINAY.pptx
VinayB68
 
IOT PPTs Week 10 Lecture Material.pptx of NPTEL Smart Cities contd
ssusera400e8
 
Construction Project Organization Group 2.pptx
vj5agdales
 
UNIT 4 Total Quality Management .pptx
gokuld13012005
 
573137875-Attendance-Management-System-original
AYUSHMANAV
 
R24 SURVEYING LAB MANUAL for civil enggi
MNANDITHACIVILSTAFF
 
CRASH COURSE IN ALTERNATIVE PLUMBING CLASS
rene militante
 
Sustainable Sites - Green Building Construction
mhdumerali
 

Day 3 take up convolutional neural network

  • 7. ICEBREAKER Phạm Nguyễn Anh Thư @thu.phamhcmut Let’s get to know each other!
  • 8. Record shop _____ ___ ____ Thật bất ngờ ___ ____ ____ Em dạo này Tình đắng như ly cà phê ____ _____ ____ _ __ ____ ___ ___ ___ ___ ___ ______ Để Mị nói cho mà nghe ____ ____ ____ ___ __ ____ ____ Đâu cần một bài ca tình yêu
  • 9. how-to-AI Series: Unlock Potential Day 3: Take UP Convolutional Neural Network Nguyễn Luật Gia Khôi @giakhoi.nguyenluat Nguyễn Thế Bình @binh.nguyen288
  • 10. Outline 1. Break the ice 2. Convolutional Neural Network 3. Why using CNN instead of NN? 4. Demo code 5. Kaggle Challenge
  • 12. 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 -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: black ● 1: white Matrix representation
  • 13. 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 -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 -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 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 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 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 ANN can solve this variety in digits
  • 14. 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 -1 -1 -1 1 1 -1 ... 1 -1 1 1 -1 -1 1 1 x2 x33 x32 x3 x4 x34 x1 x35 ... h2 x33 x32 h3 h4 x34 h1 x35 ... h2 h8 h3 h9 h1 h10 ... 1 9 8 2 3 0 0.01 0.92 0.003 0.008 0.015 0.02
  • 15. WAIT A SECOND, how about this?
  • 16. Image size = 1920 x 1080 x 3 Input layer #neurons = 1920 x 1080 x 3 = 6 million Hidden layer #neurons = Say you keep it = 4 million Weights between input and first hidden layer = 6 * 4 = 24 trillion parameters
  • 18. Disadvantages of ANN in image classification - Computationally expensive. - Treat local pixels same as pixels far apart => no locality concern - Sensitive to location of the object in the image.
  • 19. How does human recognize images?
  • 22. How can we make computers recognize these patterns?
  • 23. 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 -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 -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 -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 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 Circular pattern filter Diagonal line filter Diagonal line filter
  • 24. 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 -1 -1 -1 1 -1 -1 -1 -1 * = -1 9 -1 -5 1 -3 -3 3 -3 -5 -1 -5 -3 -5 -3 Circular pattern filter -1 x -1 + -1 x -1 + … + -1 x -1 + 1 x 1 = 9
  • 25. x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter = x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter = x x -1 -1 -1 -1 1 -1 -1 -1 -1 * Circular pattern filter =
  • 27. x x * Eye filter = x * Nose filter = x x * Ear filter = * Head filter x = x x
  • 28. x x X ... X ... h2 x8 h3 x9 h1 x10 ... 0.01 (Giraffe) 0.80 (Koala) 0.003 (Horse) 0.008 (Buffalo) 0.15 (Bear) 0.02 (Squirrel)
  • 30. 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 -1 -1 -1 1 -1 -1 -1 -1 * = -1 9 -1 -5 1 -3 -3 3 -3 -5 -1 -5 -3 -5 -3 Circular pattern filter ReLU 0 9 0 0 1 0 0 3 0 0 0 0 0 0 0
  • 31. Stride & Padding stride = 2, padding = “valid” stride = 1, padding = “same” input output input output
  • 33. 0 9 0 0 0 1 0 0 0 3 0 0 0 0 0 0 0 0 0 0 Max pooling pool_size = (2,2) stride = 2 9 0 3 0 output_shape = math.floor((input_shape - pool_size) / strides) + 1
  • 34. Benefits of max pooling - Reduce feature map size => Reduce computational cost. - Fewer parameters needed => Reduce overfitting. - Make model tolerant towards variations and distortions.
  • 35. Complete pipeline Conv + ReLU Pooling Conv + ReLU Pooling w1 x h1x c1 w2 x h2 x c1 w2 x h2 x c2 w3 x h3 x c2 Flatten
  • 36. Why using CNN instead of NN?
  • 37. - Can capture spatial features => Connection sparsity => Reduce overfitting x x * Eye filter = x * Nose filter = x x * Ear filter =
  • 38. 0 0 9 0 0 1 0 0 0 3 0 0 0 0 0 0 0 0 0 0 - Translation invariant. 1 9 3 0 0 0 0 0 0 0 9 0 0 3 0 0 0 0 0 0 0 0 0 0 0 9 3 0 Conv Conv Pooling Pooling
  • 39. - Location invariant feature detection.
  • 44. ● Challenge: Build a model that can automatically classifies images to their correct labels. ● Registration form is now open! ● Timeline: 12/12/2021 - 30/12/2021 ● Rewards: ○ First prize: 1.000.000 VND ○ Second prize: 700.000 VND ○ Third prize: 500.000 VND ○ 2 AI potential prizes: 400.000 VND GDSC AI CHALLENGE
  • 45. Q&A