Deep learning face questions 18: network of networks structures (Network in Network)

table of Contents

　　For example

　　Reference material

---

Computing network in the network structure by a plurality of branches (or convolution pooling), the calculation result on the branch connection in the depth

For example

Tensor a 3 * 3 * 2,

1 to 3 * 1 * 2 convolution kernels same convolution, step = 1,

And 2 2 * 2 * 2 convolution kernels same convolution, step = 1,

With a 2 * 3 * 3 mask in the same pool of maximum, step size = 1,

The results obtained in these three splice in the depth direction

GoogLeNet network architecture is based on a similar design of the network in the network module, the module is called in the GoogLeNet Inception Module, the plurality of modules may be combined into Inception Module GoogLeNet.

 Back to Contents

 

Reference material

"Illustrates the depth of learning and neural networks: from tensor to achieve TensorFlow" _ Zhang Ping

 

 Back to Contents