SlideShare a Scribd company logo
Min-Seo Kim
Network Science Lab
Dept. of Artificial Intelligence
The Catholic University of Korea
E-mail: kms39273@naver.com
1
 Ongoing studies
• GoogLeNet
2
GoogLeNet
• The GoogLeNet submission to ILSVRC 2014 used 12× fewer parameters than the winning architecture, VGG,
from two years prior, yet it was significantly more accurate.
• Notable factor is that with the ongoing traction of mobile and embedded computing, the efficiency of our
algorithms – especially their power and memory use – gains importance.
Introduction
3
GoogLeNet
• GoogLeNet have typically had a standard structure – stacked convolutional layers (optionally followed by
contrast normalization and maxpooling) are followed by one or more fully-connected layers.
• Use 1×1 convolutional layers, ReLU activation function in Network-in-Network.
Related Work
4
GoogLeNet
• The most straightforward way of improving the performance of deep neural networks is by increasing their
size.
• However this simple solution comes with two major drawbacks.
• Bigger size typically means a larger number of parameters, which makes the enlarged network more
prone to overfitting. To prevent overfitting, creation of high quality training sets can be tricky and
expensive.
• Uniformly increased network size is the dramatically increased use of computational resources.
• Since in practice the computational budget is always finite, an efficient distribution of computing resources is
preferred to an indiscriminate increase of size
Motivation and High Level Considerations
5
GoogLeNet - Architectural Details
• To effectively extract feature maps, 1x1, 3x3, and 5x5 convolution filters are performed in parallel.
• However, this inevitably increases the computational load.
Inception module
6
GoogLeNet - Architectural Details
• Therefore, to address this issue, the 1x1 convolution filter was used.
• By placing it before the 3x3 and 5x5 filters, it reduces the dimensions, which in turn reduces the
computational load and introduces non-linearity.
Inception module
7
GoogLeNet - Architectural Details
- input tensor = 28X28X192
- convolution filter = 5X5X192
- padding = 2
- strride = 1
- number of filter = 32
28X28X192X5X5X32=1.2 billion times
How does the 1x1 conv filter reduce the amount of computation?
- input tensor = 28X28X192
- convolution filter = 1X1X16
- number of filter = 16
192X1X1X28X28X16=2.4 million
operations
- input tensor = 28X28X16
- convolution filter = 5X5X192
- padding = 2
- strride = 1
- number of filter = 32
16x5x5x28x28x32 = 10 million operations
Total of 12.4 million operations.
The number of operations has decreased tenfold.
The non-linearity has increased.
8
GoogLeNet - Architectural Details
• This is the parameter calculation for the Inception 3a module inside the actual GoogLeNet.
Inception in GoogLeNet(inception 3a)
9
GoogLeNet - Architectural Details
Entire GoogLeNet
10
GoogLeNet - Architectural Details
• This is where the lower layers are located, close to the input image.
• For efficient memory usage, we applied a basic CNN-type model in the lower layer.
• The Inception module is used in the higher layers, so it is not used in this part.
Part 1
11
GoogLeNet - Architectural Details
• To extract various features, the Inception module described earlier is implemented.
Part 2
12
GoogLeNet - Architectural Details
• As the depth of the model becomes very deep, the vanishing gradient problem can occur even when using
the ReLU activation function.
• We added an auxiliary classifier to the middle layer, which outputs intermediate results so that the gradient
can be passed as an additional backprop.
• To prevent it from having too much influence, the loss of the auxiliary classifier is multiplied by 0.3 and added
to the total loss of the entire network.
• In the actual test, we removed the auxiliary classifier and used only the softmax of the far end.
Part 3
13
GoogLeNet - Architectural Details
• This is the end of the model with the prediction results.
• The average pooling layer with global average pooling is applied.
• This reduces the size of the feature map without any additional parameters.
Part 4
14
GoogLeNet
• We presented a new methodology that is different from the existing CNN methods that only build up depth.
• It won the first prize at ILSVRC 2014, beating VGGNet.
Conclusions

More Related Content

What's hot

UNetEliyaLaialy (2).pptx
UNetEliyaLaialy (2).pptxUNetEliyaLaialy (2).pptx
UNetEliyaLaialy (2).pptx
NoorUlHaq47
 
Lecture11
Lecture11Lecture11
Lecture11
zukun
 
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
Edge AI and Vision Alliance
 

What's hot (20)

UNetEliyaLaialy (2).pptx
UNetEliyaLaialy (2).pptxUNetEliyaLaialy (2).pptx
UNetEliyaLaialy (2).pptx
 
Machine learning for wireless networks @Bestcom2016
Machine learning for wireless networks @Bestcom2016Machine learning for wireless networks @Bestcom2016
Machine learning for wireless networks @Bestcom2016
 
ViT.pptx
ViT.pptxViT.pptx
ViT.pptx
 
Intelligent reflecting surface 2
Intelligent reflecting surface 2Intelligent reflecting surface 2
Intelligent reflecting surface 2
 
LeNet-5
LeNet-5LeNet-5
LeNet-5
 
Transfer Learning (D2L4 Insight@DCU Machine Learning Workshop 2017)
Transfer Learning (D2L4 Insight@DCU Machine Learning Workshop 2017)Transfer Learning (D2L4 Insight@DCU Machine Learning Workshop 2017)
Transfer Learning (D2L4 Insight@DCU Machine Learning Workshop 2017)
 
Domain adaptation
Domain adaptationDomain adaptation
Domain adaptation
 
CNN and its applications by ketaki
CNN and its applications by ketakiCNN and its applications by ketaki
CNN and its applications by ketaki
 
Ea 452 chap9
Ea 452 chap9Ea 452 chap9
Ea 452 chap9
 
Transfer Learning and Domain Adaptation (D2L3 2017 UPC Deep Learning for Comp...
Transfer Learning and Domain Adaptation (D2L3 2017 UPC Deep Learning for Comp...Transfer Learning and Domain Adaptation (D2L3 2017 UPC Deep Learning for Comp...
Transfer Learning and Domain Adaptation (D2L3 2017 UPC Deep Learning for Comp...
 
U-Net: Convolutional Networks for Biomedical Image Segmentation
U-Net: Convolutional Networks for Biomedical Image SegmentationU-Net: Convolutional Networks for Biomedical Image Segmentation
U-Net: Convolutional Networks for Biomedical Image Segmentation
 
La CRC: Un Regulador Inteligente
La CRC: Un Regulador Inteligente La CRC: Un Regulador Inteligente
La CRC: Un Regulador Inteligente
 
Lecture11
Lecture11Lecture11
Lecture11
 
Interference and system capacity
Interference and system capacityInterference and system capacity
Interference and system capacity
 
Lecture 15 DCT, Walsh and Hadamard Transform
Lecture 15 DCT, Walsh and Hadamard TransformLecture 15 DCT, Walsh and Hadamard Transform
Lecture 15 DCT, Walsh and Hadamard Transform
 
CRC implementation
CRC implementation CRC implementation
CRC implementation
 
PhD Defense
PhD DefensePhD Defense
PhD Defense
 
Deep learning for person re-identification
Deep learning for person re-identificationDeep learning for person re-identification
Deep learning for person re-identification
 
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
“Person Re-Identification and Tracking at the Edge: Challenges and Techniques...
 
Image Restoration (Frequency Domain Filters):Basics
Image Restoration (Frequency Domain Filters):BasicsImage Restoration (Frequency Domain Filters):Basics
Image Restoration (Frequency Domain Filters):Basics
 

Similar to GoogLeNet.pptx

(Im2col)accelerating deep neural networks on low power heterogeneous architec...
(Im2col)accelerating deep neural networks on low power heterogeneous architec...(Im2col)accelerating deep neural networks on low power heterogeneous architec...
(Im2col)accelerating deep neural networks on low power heterogeneous architec...
Bomm Kim
 

Similar to GoogLeNet.pptx (20)

Modern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentationModern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentation
 
PR-144: SqueezeNext: Hardware-Aware Neural Network Design
PR-144: SqueezeNext: Hardware-Aware Neural Network DesignPR-144: SqueezeNext: Hardware-Aware Neural Network Design
PR-144: SqueezeNext: Hardware-Aware Neural Network Design
 
PR-120: ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture De...
PR-120: ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture De...PR-120: ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture De...
PR-120: ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture De...
 
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.pptx
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.pptxEfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.pptx
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.pptx
 
240422_Thuy_Labseminar[Large Graph Property Prediction via Graph Segment Trai...
240422_Thuy_Labseminar[Large Graph Property Prediction via Graph Segment Trai...240422_Thuy_Labseminar[Large Graph Property Prediction via Graph Segment Trai...
240422_Thuy_Labseminar[Large Graph Property Prediction via Graph Segment Trai...
 
[2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review][2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review]
 
IRJET- Mango Classification using Convolutional Neural Networks
IRJET- Mango Classification using Convolutional Neural NetworksIRJET- Mango Classification using Convolutional Neural Networks
IRJET- Mango Classification using Convolutional Neural Networks
 
Sp19_P2.pptx
Sp19_P2.pptxSp19_P2.pptx
Sp19_P2.pptx
 
Traffic Sign Recognition System
Traffic Sign Recognition SystemTraffic Sign Recognition System
Traffic Sign Recognition System
 
08 neural networks
08 neural networks08 neural networks
08 neural networks
 
ConvNeXt: A ConvNet for the 2020s explained
ConvNeXt: A ConvNet for the 2020s explainedConvNeXt: A ConvNet for the 2020s explained
ConvNeXt: A ConvNet for the 2020s explained
 
lec6a.ppt
lec6a.pptlec6a.ppt
lec6a.ppt
 
IJET-V3I1P14
IJET-V3I1P14IJET-V3I1P14
IJET-V3I1P14
 
Graph Transformer with Graph Pooling for Node Classification, IJCAI 2023.pptx
Graph Transformer with Graph Pooling for Node Classification, IJCAI 2023.pptxGraph Transformer with Graph Pooling for Node Classification, IJCAI 2023.pptx
Graph Transformer with Graph Pooling for Node Classification, IJCAI 2023.pptx
 
IRJET- Image Classification – Cat and Dog Images
IRJET- Image Classification – Cat and Dog ImagesIRJET- Image Classification – Cat and Dog Images
IRJET- Image Classification – Cat and Dog Images
 
NS-CUK Seminar: S.T.Nguyen, Review on "Hierarchical Graph Convolutional Netwo...
NS-CUK Seminar: S.T.Nguyen, Review on "Hierarchical Graph Convolutional Netwo...NS-CUK Seminar: S.T.Nguyen, Review on "Hierarchical Graph Convolutional Netwo...
NS-CUK Seminar: S.T.Nguyen, Review on "Hierarchical Graph Convolutional Netwo...
 
(Im2col)accelerating deep neural networks on low power heterogeneous architec...
(Im2col)accelerating deep neural networks on low power heterogeneous architec...(Im2col)accelerating deep neural networks on low power heterogeneous architec...
(Im2col)accelerating deep neural networks on low power heterogeneous architec...
 
Recent Advances in CPLEX 12.6.1
Recent Advances in CPLEX 12.6.1Recent Advances in CPLEX 12.6.1
Recent Advances in CPLEX 12.6.1
 
Handwritten Digit Recognition(Convolutional Neural Network) PPT
Handwritten Digit Recognition(Convolutional Neural Network) PPTHandwritten Digit Recognition(Convolutional Neural Network) PPT
Handwritten Digit Recognition(Convolutional Neural Network) PPT
 
VGG.pptx
VGG.pptxVGG.pptx
VGG.pptx
 

More from ssuser2624f71

인공지능 로봇 윤리_1229_9차시.pptx
인공지능 로봇 윤리_1229_9차시.pptx인공지능 로봇 윤리_1229_9차시.pptx
인공지능 로봇 윤리_1229_9차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1228_8차시.pptx
인공지능 로봇 윤리_1228_8차시.pptx인공지능 로봇 윤리_1228_8차시.pptx
인공지능 로봇 윤리_1228_8차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1227_7차시.pptx
인공지능 로봇 윤리_1227_7차시.pptx인공지능 로봇 윤리_1227_7차시.pptx
인공지능 로봇 윤리_1227_7차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1226_6차시.pptx
인공지능 로봇 윤리_1226_6차시.pptx인공지능 로봇 윤리_1226_6차시.pptx
인공지능 로봇 윤리_1226_6차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1222_5차시.pptx
인공지능 로봇 윤리_1222_5차시.pptx인공지능 로봇 윤리_1222_5차시.pptx
인공지능 로봇 윤리_1222_5차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1221_4차시.pptx
인공지능 로봇 윤리_1221_4차시.pptx인공지능 로봇 윤리_1221_4차시.pptx
인공지능 로봇 윤리_1221_4차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1220_3차시.pptx
인공지능 로봇 윤리_1220_3차시.pptx인공지능 로봇 윤리_1220_3차시.pptx
인공지능 로봇 윤리_1220_3차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1219_2차시.pptx
인공지능 로봇 윤리_1219_2차시.pptx인공지능 로봇 윤리_1219_2차시.pptx
인공지능 로봇 윤리_1219_2차시.pptx
ssuser2624f71
 
인공지능 로봇 윤리_1218_1차시.pptx
인공지능 로봇 윤리_1218_1차시.pptx인공지능 로봇 윤리_1218_1차시.pptx
인공지능 로봇 윤리_1218_1차시.pptx
ssuser2624f71
 

More from ssuser2624f71 (20)

Vector and Matrix operationsVector and Matrix operations
Vector and Matrix operationsVector and Matrix operationsVector and Matrix operationsVector and Matrix operations
Vector and Matrix operationsVector and Matrix operations
 
240219_RNN, LSTM code.pptxdddddddddddddddd
240219_RNN, LSTM code.pptxdddddddddddddddd240219_RNN, LSTM code.pptxdddddddddddddddd
240219_RNN, LSTM code.pptxdddddddddddddddd
 
Sparse Graph Attention Networks 2021.pptx
Sparse Graph Attention Networks 2021.pptxSparse Graph Attention Networks 2021.pptx
Sparse Graph Attention Networks 2021.pptx
 
인공지능 로봇 윤리_1229_9차시.pptx
인공지능 로봇 윤리_1229_9차시.pptx인공지능 로봇 윤리_1229_9차시.pptx
인공지능 로봇 윤리_1229_9차시.pptx
 
인공지능 로봇 윤리_1228_8차시.pptx
인공지능 로봇 윤리_1228_8차시.pptx인공지능 로봇 윤리_1228_8차시.pptx
인공지능 로봇 윤리_1228_8차시.pptx
 
인공지능 로봇 윤리_1227_7차시.pptx
인공지능 로봇 윤리_1227_7차시.pptx인공지능 로봇 윤리_1227_7차시.pptx
인공지능 로봇 윤리_1227_7차시.pptx
 
인공지능 로봇 윤리_1226_6차시.pptx
인공지능 로봇 윤리_1226_6차시.pptx인공지능 로봇 윤리_1226_6차시.pptx
인공지능 로봇 윤리_1226_6차시.pptx
 
인공지능 로봇 윤리_1222_5차시.pptx
인공지능 로봇 윤리_1222_5차시.pptx인공지능 로봇 윤리_1222_5차시.pptx
인공지능 로봇 윤리_1222_5차시.pptx
 
인공지능 로봇 윤리_1221_4차시.pptx
인공지능 로봇 윤리_1221_4차시.pptx인공지능 로봇 윤리_1221_4차시.pptx
인공지능 로봇 윤리_1221_4차시.pptx
 
인공지능 로봇 윤리_1220_3차시.pptx
인공지능 로봇 윤리_1220_3차시.pptx인공지능 로봇 윤리_1220_3차시.pptx
인공지능 로봇 윤리_1220_3차시.pptx
 
인공지능 로봇 윤리_1219_2차시.pptx
인공지능 로봇 윤리_1219_2차시.pptx인공지능 로봇 윤리_1219_2차시.pptx
인공지능 로봇 윤리_1219_2차시.pptx
 
인공지능 로봇 윤리_1218_1차시.pptx
인공지능 로봇 윤리_1218_1차시.pptx인공지능 로봇 윤리_1218_1차시.pptx
인공지능 로봇 윤리_1218_1차시.pptx
 
디지털인문학9차시.pptx
디지털인문학9차시.pptx디지털인문학9차시.pptx
디지털인문학9차시.pptx
 
디지털인문학8차시.pptx
디지털인문학8차시.pptx디지털인문학8차시.pptx
디지털인문학8차시.pptx
 
디지털인문학7차시.pptx
디지털인문학7차시.pptx디지털인문학7차시.pptx
디지털인문학7차시.pptx
 
디지털인문학6차시.pptx
디지털인문학6차시.pptx디지털인문학6차시.pptx
디지털인문학6차시.pptx
 
디지털인문학 5차시.pptx
디지털인문학 5차시.pptx디지털인문학 5차시.pptx
디지털인문학 5차시.pptx
 
디지털인문학4차시.pptx
디지털인문학4차시.pptx디지털인문학4차시.pptx
디지털인문학4차시.pptx
 
디지털인문학3차시.pptx
디지털인문학3차시.pptx디지털인문학3차시.pptx
디지털인문학3차시.pptx
 
디지털인문학2차시.pptx
디지털인문학2차시.pptx디지털인문학2차시.pptx
디지털인문학2차시.pptx
 

Recently uploaded

Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdfAdversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
Po-Chuan Chen
 

Recently uploaded (20)

Introduction to Quality Improvement Essentials
Introduction to Quality Improvement EssentialsIntroduction to Quality Improvement Essentials
Introduction to Quality Improvement Essentials
 
How to Break the cycle of negative Thoughts
How to Break the cycle of negative ThoughtsHow to Break the cycle of negative Thoughts
How to Break the cycle of negative Thoughts
 
PART A. Introduction to Costumer Service
PART A. Introduction to Costumer ServicePART A. Introduction to Costumer Service
PART A. Introduction to Costumer Service
 
slides CapTechTalks Webinar May 2024 Alexander Perry.pptx
slides CapTechTalks Webinar May 2024 Alexander Perry.pptxslides CapTechTalks Webinar May 2024 Alexander Perry.pptx
slides CapTechTalks Webinar May 2024 Alexander Perry.pptx
 
Fish and Chips - have they had their chips
Fish and Chips - have they had their chipsFish and Chips - have they had their chips
Fish and Chips - have they had their chips
 
The Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve ThomasonThe Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve Thomason
 
Advances in production technology of Grapes.pdf
Advances in production technology of Grapes.pdfAdvances in production technology of Grapes.pdf
Advances in production technology of Grapes.pdf
 
Operations Management - Book1.p - Dr. Abdulfatah A. Salem
Operations Management - Book1.p  - Dr. Abdulfatah A. SalemOperations Management - Book1.p  - Dr. Abdulfatah A. Salem
Operations Management - Book1.p - Dr. Abdulfatah A. Salem
 
Mattingly "AI & Prompt Design: Limitations and Solutions with LLMs"
Mattingly "AI & Prompt Design: Limitations and Solutions with LLMs"Mattingly "AI & Prompt Design: Limitations and Solutions with LLMs"
Mattingly "AI & Prompt Design: Limitations and Solutions with LLMs"
 
Matatag-Curriculum and the 21st Century Skills Presentation.pptx
Matatag-Curriculum and the 21st Century Skills Presentation.pptxMatatag-Curriculum and the 21st Century Skills Presentation.pptx
Matatag-Curriculum and the 21st Century Skills Presentation.pptx
 
Basic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumersBasic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumers
 
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 
Solid waste management & Types of Basic civil Engineering notes by DJ Sir.pptx
Solid waste management & Types of Basic civil Engineering notes by DJ Sir.pptxSolid waste management & Types of Basic civil Engineering notes by DJ Sir.pptx
Solid waste management & Types of Basic civil Engineering notes by DJ Sir.pptx
 
Basic_QTL_Marker-assisted_Selection_Sourabh.ppt
Basic_QTL_Marker-assisted_Selection_Sourabh.pptBasic_QTL_Marker-assisted_Selection_Sourabh.ppt
Basic_QTL_Marker-assisted_Selection_Sourabh.ppt
 
Sectors of the Indian Economy - Class 10 Study Notes pdf
Sectors of the Indian Economy - Class 10 Study Notes pdfSectors of the Indian Economy - Class 10 Study Notes pdf
Sectors of the Indian Economy - Class 10 Study Notes pdf
 
INU_CAPSTONEDESIGN_비밀번호486_업로드용 발표자료.pdf
INU_CAPSTONEDESIGN_비밀번호486_업로드용 발표자료.pdfINU_CAPSTONEDESIGN_비밀번호486_업로드용 발표자료.pdf
INU_CAPSTONEDESIGN_비밀번호486_업로드용 발표자료.pdf
 
Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdfAdversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
Adversarial Attention Modeling for Multi-dimensional Emotion Regression.pdf
 
Basic Civil Engineering Notes of Chapter-6, Topic- Ecosystem, Biodiversity G...
Basic Civil Engineering Notes of Chapter-6,  Topic- Ecosystem, Biodiversity G...Basic Civil Engineering Notes of Chapter-6,  Topic- Ecosystem, Biodiversity G...
Basic Civil Engineering Notes of Chapter-6, Topic- Ecosystem, Biodiversity G...
 
Gyanartha SciBizTech Quiz slideshare.pptx
Gyanartha SciBizTech Quiz slideshare.pptxGyanartha SciBizTech Quiz slideshare.pptx
Gyanartha SciBizTech Quiz slideshare.pptx
 
Forest and Wildlife Resources Class 10 Free Study Material PDF
Forest and Wildlife Resources Class 10 Free Study Material PDFForest and Wildlife Resources Class 10 Free Study Material PDF
Forest and Wildlife Resources Class 10 Free Study Material PDF
 

GoogLeNet.pptx

  • 1. Min-Seo Kim Network Science Lab Dept. of Artificial Intelligence The Catholic University of Korea E-mail: kms39273@naver.com
  • 3. 2 GoogLeNet • The GoogLeNet submission to ILSVRC 2014 used 12× fewer parameters than the winning architecture, VGG, from two years prior, yet it was significantly more accurate. • Notable factor is that with the ongoing traction of mobile and embedded computing, the efficiency of our algorithms – especially their power and memory use – gains importance. Introduction
  • 4. 3 GoogLeNet • GoogLeNet have typically had a standard structure – stacked convolutional layers (optionally followed by contrast normalization and maxpooling) are followed by one or more fully-connected layers. • Use 1×1 convolutional layers, ReLU activation function in Network-in-Network. Related Work
  • 5. 4 GoogLeNet • The most straightforward way of improving the performance of deep neural networks is by increasing their size. • However this simple solution comes with two major drawbacks. • Bigger size typically means a larger number of parameters, which makes the enlarged network more prone to overfitting. To prevent overfitting, creation of high quality training sets can be tricky and expensive. • Uniformly increased network size is the dramatically increased use of computational resources. • Since in practice the computational budget is always finite, an efficient distribution of computing resources is preferred to an indiscriminate increase of size Motivation and High Level Considerations
  • 6. 5 GoogLeNet - Architectural Details • To effectively extract feature maps, 1x1, 3x3, and 5x5 convolution filters are performed in parallel. • However, this inevitably increases the computational load. Inception module
  • 7. 6 GoogLeNet - Architectural Details • Therefore, to address this issue, the 1x1 convolution filter was used. • By placing it before the 3x3 and 5x5 filters, it reduces the dimensions, which in turn reduces the computational load and introduces non-linearity. Inception module
  • 8. 7 GoogLeNet - Architectural Details - input tensor = 28X28X192 - convolution filter = 5X5X192 - padding = 2 - strride = 1 - number of filter = 32 28X28X192X5X5X32=1.2 billion times How does the 1x1 conv filter reduce the amount of computation? - input tensor = 28X28X192 - convolution filter = 1X1X16 - number of filter = 16 192X1X1X28X28X16=2.4 million operations - input tensor = 28X28X16 - convolution filter = 5X5X192 - padding = 2 - strride = 1 - number of filter = 32 16x5x5x28x28x32 = 10 million operations Total of 12.4 million operations. The number of operations has decreased tenfold. The non-linearity has increased.
  • 9. 8 GoogLeNet - Architectural Details • This is the parameter calculation for the Inception 3a module inside the actual GoogLeNet. Inception in GoogLeNet(inception 3a)
  • 10. 9 GoogLeNet - Architectural Details Entire GoogLeNet
  • 11. 10 GoogLeNet - Architectural Details • This is where the lower layers are located, close to the input image. • For efficient memory usage, we applied a basic CNN-type model in the lower layer. • The Inception module is used in the higher layers, so it is not used in this part. Part 1
  • 12. 11 GoogLeNet - Architectural Details • To extract various features, the Inception module described earlier is implemented. Part 2
  • 13. 12 GoogLeNet - Architectural Details • As the depth of the model becomes very deep, the vanishing gradient problem can occur even when using the ReLU activation function. • We added an auxiliary classifier to the middle layer, which outputs intermediate results so that the gradient can be passed as an additional backprop. • To prevent it from having too much influence, the loss of the auxiliary classifier is multiplied by 0.3 and added to the total loss of the entire network. • In the actual test, we removed the auxiliary classifier and used only the softmax of the far end. Part 3
  • 14. 13 GoogLeNet - Architectural Details • This is the end of the model with the prediction results. • The average pooling layer with global average pooling is applied. • This reduces the size of the feature map without any additional parameters. Part 4
  • 15. 14 GoogLeNet • We presented a new methodology that is different from the existing CNN methods that only build up depth. • It won the first prize at ILSVRC 2014, beating VGGNet. Conclusions