0. Revision: Fully connected Neural Network 全连接

1. Convolution Neural Network

保留空间信息

1.1 Convolution

Convolution-Single Input Channel 单通道 

数乘

3 input Channels 3通道

N input Channels

N input Channels and M output channel

M 个卷积核

1.2 convolution layer

3×3卷积核 图像大小-2

convolution layer-padding 想要保持维数不变

3×3 3/2=1 添1圈

5×5 5/2=2 添2圈

padding=1

convolution layer-stride=2

1.3 Max Pooling Layer

通道数量不变

2. A simple convolution neural network

3. How to use GPU

1. Move Model to GPU

2. Move Tensor to GPU

Results

错误率来说1/3的提升