Why there is no bias and coefficient trainable parameters in the Max- and Average-Pooling Layers?

Question

LINGJUN KONG 2017-7-25

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/350254-why-there-is-no-bias-and-coefficient-trainable-parameters-in-the-max-and-average-pooling-layers

评论： LINGJUN KONG 2017-7-29

在 MATLAB Online 中打开

For many Network, there are 2 trainable parameters in Polling Layers: Bias and Coefficient

Y=Coefficient*Max(X)+Bias;

Why the Matlab Layer doesn't support it? Are those 2 parameters not very useful for the final performance?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Sailesh Sidhwani 2017-7-28

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/350254-why-there-is-no-bias-and-coefficient-trainable-parameters-in-the-max-and-average-pooling-layers#answer_275919

Hi Lingjun,

Based on the MATLAB documentation, max-pooling and average-pooling layers does not perform any learning and are only used for down-sampling the data. Refer to the the section "More About" in the following documentation pages:

Average-pooling layer: https://www.mathworks.com/help/nnet/ref/averagepooling2dlayer.html

Max-pooling layer: https://www.mathworks.com/help/nnet/ref/maxpooling2dlayer.html

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

LINGJUN KONG 2017-7-29

Is that because of that, the parameters are not very useful for the downsampling layer? I've seen that there are 2 trainable parameters in the LeNet-5

请先登录，再进行评论。