Is it possible to see the activation function inside the layers of a neural network written entirely by hand?

2 次查看（过去 30 天）

Stefano 2019-4-10

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/455498-is-it-possible-to-see-the-activation-function-inside-the-layers-of-a-neural-network-written-entirely

编辑： Stefano 2019-4-10

I want to describe a very simple neural network, for example composed by an input image and a convolution layer, where all the parameters are defined by hand! Then providing an input image I want to see the outputs of that layer.

There are two problems:

1) If I describe the network simply through its layers. The "activation" function does not allow me to adopt as input the layers and I receive the following error.

"Undefined function 'activations' for input arguments of type 'nnet.cnn.layer.Layer'.

2) the only way to have a "net" suitable to be used with the activation function is or to train a network or to import a pretrained one, and in such a case the most parameters of the network is locked and cannot be changed by hand.