1 Image Classification (Neuron Version)

–

1.1 Observation 1

1.2 Observation 2

如果不同的receptive field需要相同功能的neuron，可以使这些neuron共享参数

1.3 Benefit of Convolutional Layer

2 Image Classification (Filter Version)

不用担心filter大小3x3会很小，只要network够深，filter能侦探到的size会越来越大

3 Comparison of Two Stories

–

4 Pooling

pooling 可以减少运算量

–

5 Application: Playing Go