Paper: Constructing fine-granularity functional brain network atlases via deep convolutional autoencoder

[gkiar]

URL: https://www.sciencedirect.com/science/article/pii/S1361841517301287

• finally a good graphic showing the network structure! (fig 3)
• they also use task fMRI from HCP, downsampled to 45x54x45
• I think we should use this structure as a starting point.
• for training they used adadelta, which, as far as I can tell, is similar in concept to adam but older. I'll use adam.