卷积神经网络-CNN

The Basic Conception：

 

Case Study：

LeNet-5

AlexNet

To be mensionend：

1.archtecture picture above is crop from krizhevsky's paper, its firster layer which is 224*224 that actually can't output 55*55 volume, so we make the input images as 227*227*3

 2.now we no longer use the norm layer because it actully do nothing to make provement

 ZFNet

VGGNet

 

 

Note:

1. different from AlexNet using so many conv layers with large shape, VGGNet only use 3*3 conv layer

2. as the spatial size is decresing, the number of parameters is increasing.

3. most memory is in early CONV, and most params is in late FC

GoogLeNet

 

ResNet

 

AlphaGo

 Summary

 Reference:

CNN course

[1]CS231n Winter 2016: Lecture 7: Convolutional Neural Networks

https://www.youtube.com/watch?v=LxfUGhug-iQ

CNN case

[2]ImageNet Classification with Deep Convolutional Neural Networks

http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf

[3]MSRA - Deep residual learning

https://www.youtube.com/watch?v=1PGLj-uKT1w

Understand CNN using excel

[4]Architecture of Convolutional Neural Networks (CNNs) demystified

https://www.analyticsvidhya.com/blog/2017/06/architecture-of-convolutional-neural-networks-simplified-demystified/