SlideShare a Scribd company logo

Cnn method

Download as PPT, PDF
A
AmirSajedi1

This document provides an overview of convolutional neural networks (CNNs). It describes that CNNs are a type of deep learning model used in computer vision tasks. The key components of a CNN include convolutional layers that extract features, pooling layers that reduce spatial size, and fully-connected layers at the end for classification. Convolutional layers apply learnable filters in a local receptive field, while pooling layers perform downsampling. The document outlines common CNN architectures, such as types of layers, hyperparameters like stride and padding, and provides examples to illustrate how CNNs work.

Science
1 of 29
Downloaded 88 times
1
2
3
4
5
6
Most read
7
8
Most read
9
Most read
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Convolutional Neural Network… Convolutional Neural Network AmirHossein Sajedi Master of Computer Engineering – Artificial Inteligence
Deep learning Deep learning is an artificial intelligence function that imitates the workings of the human brain in processing data and creating patterns for use in decision making. Deep learning is a subset of machine learning in artificial intelligence that has networks capable of learning unsupervised from data that is unstructured or unlabeled. Also known as deep neural learning or deep neural network.
Convolutional Neural Network CNN is a supervised deep learning method.The origin of the applications of deep learning to object recognition tasks can be traced to the convolutional neural networks (CNNs) in the early 90s. The CNNbased architectures have captured intense interest in computer vision since October 2012 shortly after the ImageNet competition results were released.
CNN ARCHITECTURE They are very similar to ordinary NNs. They are made up of neurons that have learnable weights and biases. Each neuron receives some inputs, performs a dot product and optionally follows it with a nonlinearity. Their connectivity is now restricted to be local spatially.
CNN ARCHITECTURE
CNN ARCHITECTURE CNNs are comprised of three types of layers: —Convolutional layers Pooling layers Fully-connected layers Unlike standard ANNs, the neurons within Convolutional and Poolinglayers will only connect to a small region of the layer preceding it
CNN ARCHITECTURE The neurons in a layer will only be connected to a small region of the layer before it, instead of all of the neurons in a fully-connected manner.
CONVOLUTIONAL LAYER The convolution operation extracts different features of the input. The first convolution layer extracts lowlevel features like edges, lines, and corners. Higherlevel layers extract higher- level features.
CONVOLUTIONAL LAYER
LOCAL CONNECTIVITY Every neuron in a convolutional layer is only connected to small region of the input volume. The dimensionality of this region is commonly referred to as the receptive field size of the neuron.The magnitude of the connectivity through the depth is nearly always equal to the depth of the input.
CONVOLUTIONAL LAYER-DEPTH The depth of the output volume produced by the convolutional layer is the number of neurons within the layer looking at the same region in the input. Reducing this hyperparameter can significantly minimisethe total number of neurons of the network, but it can also significantly reduce the pattern recognition capabilities of the model.
STRIDE we must specify the stride with which we slide the filter. When the stride is 1 then we move thefilters one pixel at a time. When the stride is 2 (or uncommonly 3 or more, though this is rare in practice) then the filters jump 2 pixels at a time as we slide them around. This will produce smaller output volumes spatially.
CONSTRAINTS ON STRIDES When the input has size W=10, no zero padding is used P=0, and the filter size is F=3, then it would be impossible to use stride S=2 (W−F+2P)/S+1=(10−3+0)/2+1=4.5
ZERO-PADDING Sometimes it will be convenient to pad the input volume with zeros around the border. The nice feature of zero padding is that it will allow us to control the spatial size of the output volumes. We will use it to exactly preserve the spatial size of the input volume so the input and output width and height are the same.
THE SPATIAL SIZE OF THE OUTPUT W: the input volume size F: the receptive field size of the Conv Layer neurons S: the stride with which they are applied P: the amount of zero padding
POOLING LAYER/SUBSAMPLING LAYERS  It is common to periodically insert a Pooling layer in-between successive Conv layers.  It progressively reduce the spatial size of the representation to reduce the amount of parameters and computation in the network, and hence to also control overfitting.  It makes the features robust against noise and distortion.  The Pooling Layer operates independently on every depth slice of the input and resizes it spatially using the pooling operation.
TYPES OF POOLING
NON-LINEAR LAYERS Neural networks in general and CNNs in particular rely on a nonlinear trigger function to signal distinct identification of likely features on each hidden layer. CNNs may use a variety of specific functions such as rectified linear units (ReLUs) and continuous trigger (non- linear) functions to efficiently implement this nonlinear triggering.
RELU In comparison to the other non-linear functions used in CNNs (e.g.,hyperbolic tangent, absolute of hyperbolic tangent, and sigmoid), the advantage of a ReLU is that the network trains many times faster.
NORMALIZATION LAYER Many types of normalization layers have been proposed for use in ConvNet architectures, sometimes with the intentions of implementing inhibition schemes observed in the biological brain. However, these layers have since fallen out of favor because in practice their contribution has been shown to be minimal, if any.For various types of normalizations.
FULLY-CONNECTED LAYER They are often used as the final layers of a CNN. These layers mathematically sum a weighting of the previous layer of features, indicating the precise mix of “ingredients” to determine a specific target output result. In case of a fully connected layer, all the elements of all the features of the previous layer get used in the calculation of each element of each output feature.
EXAMPLE ARCHITECTURE: OVERVIEW  A simple ConvNet for CIFAR-10 classification could have the architecture [INPUT - CONV - RELU - POOL - FC]: v INPUT [32x32x3] will hold the raw pixel values of the image, (width,height,color).v CONV layer will compute the output of neurons that are connected to local regions in the input, each computing a dot product between their weights and a small region they are connected to in the input volume. This may result in volume such as [32x32x12] if we decided to use 12 filters. v RELU layer will apply an element wise activation function, such as the max(0,x). This leaves the size of the volume unchanged. v POOL layer will perform a down sampling operation along the spatial dimensions, resulting in volume such as [16x16x12]. v FC (i.e. fully-connected) layer will compute the class scores, resulting in volume of size [1x1x10], where each of the 10 numbers correspond to a class score, such as among the 10 categories of CIFAR-10.
CONV DEMO  Since 3D volumes are hard to visualize, all the volumes (the input volume (in blue), the weight volumes (in red), the output volume (in green)) are visualized with each depth slice stacked in rows. The input volume is of size W1=5,H1=5,D1=3, and the CONV layer parameters are K=2,F=3,S=2,P=1. That is, we have two filters of size 3×33×3, and they are applied with a stride of 2. Therefore, the output volume size has spatial size (5 - 3 + 2)/2 +1 = 3. Moreover, notice that a padding of P=1 is applied to the input volume, making the outer border of the input volume zero. The visualization below iterates over the output activations (green), and shows that each element is computed by element wise multiplying the highlighted input (blue) with the filter (red), summing it up, and then offsetting the result by the bias.
CONV DEMO
CONV DEMO
CONV DEMO
GOOGLE-NET -Example  GoogLeNet. The ILSVRC 2014 winner was a Convolutional Network from Szegedy et al. from Google. Its main contribution was the development of an Inception Module that dramatically reduced the number of parameters in the network (4M, compared to AlexNet with 60M). Additionally, this paper uses Average Pooling instead of Fully Connected layers at the top of the ConvNet, eliminating a large amount of parameters that do not seem to matter much. There are also several follow up versions to the GoogLeNet, most recently Inception-v4.
GOOGLE-NET
Thank you

More Related Content

PPTX
CNN and its applications by ketaki
Ketaki Patwari
 
PPTX
CONVOLUTIONAL NEURAL NETWORK
Md Rajib Bhuiyan
 
PPTX
Convolutional neural network
Ferdous ahmed
 
PPTX
Convolutional Neural Network (CNN) - image recognition
YUNG-KUEI CHEN
 
PPTX
AlexNet
Bertil Hatt
 
PPTX
Convolutional Neural Networks
milad abbasi
 
PDF
CNN Algorithm
georgejustymirobi1
 
PPTX
Cnn
Mehrnaz Faraz
 
CNN and its applications by ketaki
Ketaki Patwari
 
CONVOLUTIONAL NEURAL NETWORK
Md Rajib Bhuiyan
 
Convolutional neural network
Ferdous ahmed
 
Convolutional Neural Network (CNN) - image recognition
YUNG-KUEI CHEN
 
AlexNet
Bertil Hatt
 
Convolutional Neural Networks
milad abbasi
 
CNN Algorithm
georgejustymirobi1
 
Cnn
Mehrnaz Faraz
 

What's hot (20)

PPTX
Convolution Neural Network (CNN)
Basit Rafiq
 
PPTX
Convolutional Neural Networks
Ashray Bhandare
 
PPTX
Image classification using cnn
SumeraHangi
 
PPTX
Convolutional Neural Network (CNN)
Muhammad Haroon
 
PPTX
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
PPTX
CNN Tutorial
Sungjoon Choi
 
PPTX
Image Classification using deep learning
Asma-AH
 
PDF
Convolutional Neural Network Models - Deep Learning
Mohamed Loey
 
PPTX
Convolutional Neural Network and Its Applications
Kasun Chinthaka Piyarathna
 
PPTX
Introduction to Convolutional Neural Networks
ParrotAI
 
PDF
Deep Learning - Convolutional Neural Networks
Christian Perone
 
PPTX
Introduction to CNN
Shuai Zhang
 
PPTX
Convolutional Neural Network
Vignesh Suresh
 
PDF
MobileNet - PR044
Jinwon Lee
 
PDF
Convolutional neural network
Yan Xu
 
PDF
Resnet
ashwinjoseph95
 
PDF
Convolutional Neural Networks (CNN)
Gaurav Mittal
 
PDF
Mobilenetv1 v2 slide
威智 黃
 
PPTX
Deep Learning - CNN and RNN
Ashray Bhandare
 
PDF
Object Detection Using R-CNN Deep Learning Framework
Nader Karimi
 
Convolution Neural Network (CNN)
Basit Rafiq
 
Convolutional Neural Networks
Ashray Bhandare
 
Image classification using cnn
SumeraHangi
 
Convolutional Neural Network (CNN)
Muhammad Haroon
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
CNN Tutorial
Sungjoon Choi
 
Image Classification using deep learning
Asma-AH
 
Convolutional Neural Network Models - Deep Learning
Mohamed Loey
 
Convolutional Neural Network and Its Applications
Kasun Chinthaka Piyarathna
 
Introduction to Convolutional Neural Networks
ParrotAI
 
Deep Learning - Convolutional Neural Networks
Christian Perone
 
Introduction to CNN
Shuai Zhang
 
Convolutional Neural Network
Vignesh Suresh
 
MobileNet - PR044
Jinwon Lee
 
Convolutional neural network
Yan Xu
 
Resnet
ashwinjoseph95
 
Convolutional Neural Networks (CNN)
Gaurav Mittal
 
Mobilenetv1 v2 slide
威智 黃
 
Deep Learning - CNN and RNN
Ashray Bhandare
 
Object Detection Using R-CNN Deep Learning Framework
Nader Karimi
 
Ad

Similar to Cnn method (20)

PDF
cnn-170917175001 (1).pdf
EngineeringTamilan
 
PDF
convolutional_neural_networks in deep learning
ssusere5ddd6
 
PDF
Overview of Convolutional Neural Networks
ananth
 
PPTX
build a Convolutional Neural Network (CNN) using TensorFlow in Python
Kv Sagar
 
PDF
_AI_Stanford_Super_#DeepLearning_Cheat_Sheet!_😊🙃😀🙃😊.pdf
SongsDrizzle
 
PDF
super-cheatsheet-deep-learning.pdf
DeanSchoolofElectron
 
ODP
Convolutional Neural Networks
Tianxiang Xiong
 
PDF
DL.pdf
ssuserd23711
 
PDF
cnn.pdf
Amnaalia
 
PPTX
Cnn
Nirthika Rajendran
 
PDF
Convolutional Neural Networks - Veronica Vilaplana - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
PPTX
Mnist report ppt
RaghunandanJairam
 
PDF
Mnist report
RaghunandanJairam
 
PDF
Convolutional Neural Networks (DLAI D5L1 2017 UPC Deep Learning for Artificia...
Universitat Politècnica de Catalunya
 
PPTX
Machine Learning - Convolutional Neural Network
Richard Kuo
 
PDF
Faire de la reconnaissance d'images avec le Deep Learning - Cristina & Pierre...
Jedha Bootcamp
 
PPTX
intro-to-cnn-April_2020.pptx
ssuser3aa461
 
PPTX
Convolutional Neural Network (CNN)of Deep Learning
alihassaah1994
 
PPTX
Convolutional neural networks
Learning Courses Online
 
PPTX
CNN_Presentation to learn the basics of CNN Model.pptx
bani30122004
 
cnn-170917175001 (1).pdf
EngineeringTamilan
 
convolutional_neural_networks in deep learning
ssusere5ddd6
 
Overview of Convolutional Neural Networks
ananth
 
build a Convolutional Neural Network (CNN) using TensorFlow in Python
Kv Sagar
 
_AI_Stanford_Super_#DeepLearning_Cheat_Sheet!_😊🙃😀🙃😊.pdf
SongsDrizzle
 
super-cheatsheet-deep-learning.pdf
DeanSchoolofElectron
 
Convolutional Neural Networks
Tianxiang Xiong
 
DL.pdf
ssuserd23711
 
cnn.pdf
Amnaalia
 
Cnn
Nirthika Rajendran
 
Convolutional Neural Networks - Veronica Vilaplana - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
Mnist report ppt
RaghunandanJairam
 
Mnist report
RaghunandanJairam
 
Convolutional Neural Networks (DLAI D5L1 2017 UPC Deep Learning for Artificia...
Universitat Politècnica de Catalunya
 
Machine Learning - Convolutional Neural Network
Richard Kuo
 
Faire de la reconnaissance d'images avec le Deep Learning - Cristina & Pierre...
Jedha Bootcamp
 
intro-to-cnn-April_2020.pptx
ssuser3aa461
 
Convolutional Neural Network (CNN)of Deep Learning
alihassaah1994
 
Convolutional neural networks
Learning Courses Online
 
CNN_Presentation to learn the basics of CNN Model.pptx
bani30122004
 
Ad

Recently uploaded (20)

PDF
Biophysics 2.pdffffffffffffffffffffffffff
MiraMusleh
 
PDF
AlphaEarth Foundations and the Satellite Embedding dataset
gdgforscience
 
PDF
IFIT3 RNA-binding activity primores influenza A viruz infection and translati...
saracelre
 
PPT
POSITIONING IN OPERATION THEATRE ROOM.ppt
PrabakaranMP1
 
PPTX
BIOMOLECULES PPT........................
vachieagrawal1221
 
PPTX
Derivatives of integument scales, beaks, horns,.pptx
Dr Showkat Ahmad Wani
 
PPTX
Protein & Amino Acid Structures Levels of protein structure (primary, seconda...
Muhammad Sajid Afridi
 
PDF
An interstellar mission to test astrophysical black holes
SĂŠrgio Sacani
 
PDF
Sciences of Europe No 170 (2025)
Sciences of Europe
 
PDF
VARICELLA VACCINATION: A POTENTIAL STRATEGY FOR PREVENTING MULTIPLE SCLEROSIS
ijab2
 
PDF
Phytochemical Investigation of Miliusa longipes.pdf
IrfanShahirSharafi
 
PPTX
G5Q1W8 PPT SCIENCE.pptx 2025-2026 GRADE 5
Rodel Gordzkie
 
PPTX
INTRODUCTION TO EVS | Concept of sustainability
manumaddy15198
 
PPT
The World of Physical Science, • Labs: Safety Simulation, Measurement Practice
joybejerano
 
PPTX
microscope-Lecturecjchchchchcuvuvhc.pptx
jaylaurencenobleza
 
PPTX
2. Earth - The Living Planet earth and life
markjustinebarolobau
 
PPTX
ANEMIA WITH LEUKOPENIA MDS 07_25.pptx htggtftgt fredrctvg
vedvyas5668
 
PPTX
famous lake in india and its disturibution and importance
judesebina
 
PDF
SEHH2274 Organic Chemistry Notes 1 Structure and Bonding.pdf
ivan1108lau
 
PPTX
The KM-GBF monitoring framework – status & key messages.pptx
pensoftservices
 
Biophysics 2.pdffffffffffffffffffffffffff
MiraMusleh
 
AlphaEarth Foundations and the Satellite Embedding dataset
gdgforscience
 
IFIT3 RNA-binding activity primores influenza A viruz infection and translati...
saracelre
 
POSITIONING IN OPERATION THEATRE ROOM.ppt
PrabakaranMP1
 
BIOMOLECULES PPT........................
vachieagrawal1221
 
Derivatives of integument scales, beaks, horns,.pptx
Dr Showkat Ahmad Wani
 
Protein & Amino Acid Structures Levels of protein structure (primary, seconda...
Muhammad Sajid Afridi
 
An interstellar mission to test astrophysical black holes
SĂŠrgio Sacani
 
Sciences of Europe No 170 (2025)
Sciences of Europe
 
VARICELLA VACCINATION: A POTENTIAL STRATEGY FOR PREVENTING MULTIPLE SCLEROSIS
ijab2
 
Phytochemical Investigation of Miliusa longipes.pdf
IrfanShahirSharafi
 
G5Q1W8 PPT SCIENCE.pptx 2025-2026 GRADE 5
Rodel Gordzkie
 
INTRODUCTION TO EVS | Concept of sustainability
manumaddy15198
 
The World of Physical Science, • Labs: Safety Simulation, Measurement Practice
joybejerano
 
microscope-Lecturecjchchchchcuvuvhc.pptx
jaylaurencenobleza
 
2. Earth - The Living Planet earth and life
markjustinebarolobau
 
ANEMIA WITH LEUKOPENIA MDS 07_25.pptx htggtftgt fredrctvg
vedvyas5668
 
famous lake in india and its disturibution and importance
judesebina
 
SEHH2274 Organic Chemistry Notes 1 Structure and Bonding.pdf
ivan1108lau
 
The KM-GBF monitoring framework – status & key messages.pptx
pensoftservices
 

Cnn method

  • 1. Convolutional Neural Network… Convolutional Neural Network AmirHossein Sajedi Master of Computer Engineering – Artificial Inteligence
  • 2. Deep learning Deep learning is an artificial intelligence function that imitates the workings of the human brain in processing data and creating patterns for use in decision making. Deep learning is a subset of machine learning in artificial intelligence that has networks capable of learning unsupervised from data that is unstructured or unlabeled. Also known as deep neural learning or deep neural network.
  • 3. Convolutional Neural Network CNN is a supervised deep learning method.The origin of the applications of deep learning to object recognition tasks can be traced to the convolutional neural networks (CNNs) in the early 90s. The CNNbased architectures have captured intense interest in computer vision since October 2012 shortly after the ImageNet competition results were released.
  • 4. CNN ARCHITECTURE They are very similar to ordinary NNs. They are made up of neurons that have learnable weights and biases. Each neuron receives some inputs, performs a dot product and optionally follows it with a nonlinearity. Their connectivity is now restricted to be local spatially.
  • 6. CNN ARCHITECTURE CNNs are comprised of three types of layers: —Convolutional layers Pooling layers Fully-connected layers Unlike standard ANNs, the neurons within Convolutional and Poolinglayers will only connect to a small region of the layer preceding it
  • 7. CNN ARCHITECTURE The neurons in a layer will only be connected to a small region of the layer before it, instead of all of the neurons in a fully-connected manner.
  • 8. CONVOLUTIONAL LAYER The convolution operation extracts different features of the input. The first convolution layer extracts lowlevel features like edges, lines, and corners. Higherlevel layers extract higher- level features.
  • 10. LOCAL CONNECTIVITY Every neuron in a convolutional layer is only connected to small region of the input volume. The dimensionality of this region is commonly referred to as the receptive field size of the neuron.The magnitude of the connectivity through the depth is nearly always equal to the depth of the input.
  • 11. CONVOLUTIONAL LAYER-DEPTH The depth of the output volume produced by the convolutional layer is the number of neurons within the layer looking at the same region in the input. Reducing this hyperparameter can significantly minimisethe total number of neurons of the network, but it can also significantly reduce the pattern recognition capabilities of the model.
  • 12. STRIDE we must specify the stride with which we slide the filter. When the stride is 1 then we move thefilters one pixel at a time. When the stride is 2 (or uncommonly 3 or more, though this is rare in practice) then the filters jump 2 pixels at a time as we slide them around. This will produce smaller output volumes spatially.
  • 13. CONSTRAINTS ON STRIDES When the input has size W=10, no zero padding is used P=0, and the filter size is F=3, then it would be impossible to use stride S=2 (W−F+2P)/S+1=(10−3+0)/2+1=4.5
  • 14. ZERO-PADDING Sometimes it will be convenient to pad the input volume with zeros around the border. The nice feature of zero padding is that it will allow us to control the spatial size of the output volumes. We will use it to exactly preserve the spatial size of the input volume so the input and output width and height are the same.
  • 15. THE SPATIAL SIZE OF THE OUTPUT W: the input volume size F: the receptive field size of the Conv Layer neurons S: the stride with which they are applied P: the amount of zero padding
  • 16. POOLING LAYER/SUBSAMPLING LAYERS  It is common to periodically insert a Pooling layer in-between successive Conv layers.  It progressively reduce the spatial size of the representation to reduce the amount of parameters and computation in the network, and hence to also control overfitting.  It makes the features robust against noise and distortion.  The Pooling Layer operates independently on every depth slice of the input and resizes it spatially using the pooling operation.
  • 18. NON-LINEAR LAYERS Neural networks in general and CNNs in particular rely on a nonlinear trigger function to signal distinct identification of likely features on each hidden layer. CNNs may use a variety of specific functions such as rectified linear units (ReLUs) and continuous trigger (non- linear) functions to efficiently implement this nonlinear triggering.
  • 19. RELU In comparison to the other non-linear functions used in CNNs (e.g.,hyperbolic tangent, absolute of hyperbolic tangent, and sigmoid), the advantage of a ReLU is that the network trains many times faster.
  • 20. NORMALIZATION LAYER Many types of normalization layers have been proposed for use in ConvNet architectures, sometimes with the intentions of implementing inhibition schemes observed in the biological brain. However, these layers have since fallen out of favor because in practice their contribution has been shown to be minimal, if any.For various types of normalizations.
  • 21. FULLY-CONNECTED LAYER They are often used as the final layers of a CNN. These layers mathematically sum a weighting of the previous layer of features, indicating the precise mix of “ingredients” to determine a specific target output result. In case of a fully connected layer, all the elements of all the features of the previous layer get used in the calculation of each element of each output feature.
  • 22. EXAMPLE ARCHITECTURE: OVERVIEW  A simple ConvNet for CIFAR-10 classification could have the architecture [INPUT - CONV - RELU - POOL - FC]: v INPUT [32x32x3] will hold the raw pixel values of the image, (width,height,color).v CONV layer will compute the output of neurons that are connected to local regions in the input, each computing a dot product between their weights and a small region they are connected to in the input volume. This may result in volume such as [32x32x12] if we decided to use 12 filters. v RELU layer will apply an element wise activation function, such as the max(0,x). This leaves the size of the volume unchanged. v POOL layer will perform a down sampling operation along the spatial dimensions, resulting in volume such as [16x16x12]. v FC (i.e. fully-connected) layer will compute the class scores, resulting in volume of size [1x1x10], where each of the 10 numbers correspond to a class score, such as among the 10 categories of CIFAR-10.
  • 23. CONV DEMO  Since 3D volumes are hard to visualize, all the volumes (the input volume (in blue), the weight volumes (in red), the output volume (in green)) are visualized with each depth slice stacked in rows. The input volume is of size W1=5,H1=5,D1=3, and the CONV layer parameters are K=2,F=3,S=2,P=1. That is, we have two filters of size 3×33×3, and they are applied with a stride of 2. Therefore, the output volume size has spatial size (5 - 3 + 2)/2 +1 = 3. Moreover, notice that a padding of P=1 is applied to the input volume, making the outer border of the input volume zero. The visualization below iterates over the output activations (green), and shows that each element is computed by element wise multiplying the highlighted input (blue) with the filter (red), summing it up, and then offsetting the result by the bias.
  • 27. GOOGLE-NET -Example  GoogLeNet. The ILSVRC 2014 winner was a Convolutional Network from Szegedy et al. from Google. Its main contribution was the development of an Inception Module that dramatically reduced the number of parameters in the network (4M, compared to AlexNet with 60M). Additionally, this paper uses Average Pooling instead of Fully Connected layers at the top of the ConvNet, eliminating a large amount of parameters that do not seem to matter much. There are also several follow up versions to the GoogLeNet, most recently Inception-v4.