Skip to main content
. 2014 Jan 27;7:195. doi: 10.3389/fncom.2013.00195

Figure 2.

Figure 2

Structure of the convolutional autoencoder. Convolutional forward weights (red) compute the hidden layer activation levels and convolutional backward weights (blue) generate the reconstruction of the given input. The reconstruction layer is compared to the centered part (dashed block) of the input layer.