Skip to main content
. Author manuscript; available in PMC: 2020 Dec 25.
Published in final edited form as: Neuroimage. 2020 May 30;218:116989. doi: 10.1016/j.neuroimage.2020.116989

Table 1:

Network architecture of CNN and MLR

Layer name Output size time × channel Description
Input 46 × 1 Cropped time series (fragment)
Normalize 46 × 1 Temporal normalization
Conv1 32 × 16 Convolutional layer; [15,16] × 2
Conv2 30 × 16 Convolutional layer; [3,16] × 2
Conv3 28 × 16 Convolutional layer; [3,16] × 2
DownSample1 14 × 32 Down-sampling (split time-series into channels)
Conv4 12 × 32 Convolutional layer; [3, 32] × 2
Conv5 10 × 32 Convolutional layer; [3, 32] × 2
DownSample2 5 × 64 Down-sampling (split time-series into channels)
Conv6 3 × 64 Convolutional layer; [3, 64] × 2
Conv7 (feature) 1 × 5 Convolutional layer; [3, 5] × 2
MLR 1 × 1, 833 Fully connected layer; [5 × 1, 833]