Where can I find the detailed structure of the autoencoder network variable "net" obtained by the trainautoencoder function? The network structure diagram provided by the "vie

Where can I find the detailed structure of the autoencoder network variable "net" obtained by the trainautoencoder function? The network structure diagram provided by the "view" function is too broad, which layers are included in encoder and decoder?

Answer 1

[Ben]

You can view the network by calling the network function:

% Set up toy data and autoencoder 
t = linspace(0,2*pi,10).';
phi = randn(1,20);
x = sin(t+phi);
enc = trainAutoencoder(x);
net = network(enc);
layers = net.layers;

This gave me a cell array where layers{1} is the encoder layer and layers{2} is the decoder layer.

[jiajun]

I see. But I'm more interested in the internal structure of the encoder and decoder layer. Can you provide relevant information?

[Ben]

I'm not sure what internal structure you require - the encoder and decoder networks appear to be described by one layer each, each layer is a fully connected layer where are a weight matrix and bias vector, and σ an activation function.

The precise configuration and algorithmic details can be found on the doc page.