5.1-5.2 卷积神经网络——卷积和池化

卷积神经网络——卷积和池化

Convolution Layer

• 可以保全输入的空间结构
• 卷积神经网络越深所学习到的特征越高阶
• 卷积层输出大小公式：
  $$size=(N+2P-F)/stride+1$$
• $1\times 1$ convolution layers make perfect sense

视觉之外的卷积神经网络

• $5\times 5$ filters -> $5\times 5$ receptive field for each neuron

Pooling layer

• make the representations smaller and more manageable
• operates over each activation map independently
• Note that it is not common to use zero-padding for Pooling layers