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

Figure 3.

Figure 3

Receptive fields of the feed-forward connections generating the activation levels for the phase simulations. (A) The regular Gabor filters are generated with 8 different orientations and 6 different color channels. (B) The convolutional autoencoder weights are learned by optimizing the reconstruction cost, sparseness and weight decay.