【笔记】3.6卷积神经网络

密集层->卷积层

• 
• 
• 
  卷积层
• Locality: an output is compputed from k*k input windows
• Translation invariant: outputs use the same k*k weights(kernel)
• model params of a conv layer does not depend on input/output size
• A kernel may learn to identify a pattern

代码实现

h,w=K.shape
Y=torch.zeros((X.shape[0]-h+1,X.shape[1]-w+1))
for i in range(Y.shape[0]):
    for j in range(Y.shape[1]):
        Y[i,j]=(X[i:i+h,j:j+w]*K).sum()

池化层

T=torch.zeros((X.shape[0]-h+1,X.shape[1]-w+1))
for i in range(Y.shape[0]):
    for j in range(Y.shape[1]):
        if mode=='max':
            Y[i,j]=X[i:i+h,j:j+w].max()
        elif mode=='avg':
            Y[i,j]=X[i:i+h,j:j+w].mean()

卷积神经网络

• 堆叠卷积层来提取特征
  • 卷积层之后使用激活函数
  • 使用池化来降低位置敏感性
• 模型CNN是一种具有多种超参数和层连接的深度神经网络(AlexNet, VGG, Inceptions, ResNet, MobileNet)