Table A.1.
Siamese neural network encoder architecture
| LAYER (TYPE) | OUTPUT SHAPE | PARAMETERS | CONNECTED TO |
|---|---|---|---|
| INPUTLAYER | (192, 184, 30) | 0 | - |
| CONV_2D_1 | (192, 184, 64) | 48064 | INPUTLAYER |
| Nf = 64, ks = (5x5) | |||
| MAX_POOLING_2D_1 | (96, 92, 64) | 0 | CONV_2D_1 |
| BATCHNORM_1 | (96, 92, 64) | 256 | MAX_POOLING_2D_1 |
| CONV_2D_2 | (96, 92, 128) | 73856 | BATCHNORM_1 |
| Nf = 128, ks = (3x3) | |||
| MAX_POOLING_2D_2 | (48, 46, 128) | 0 | CONV_2D_2 |
| BATCHNORM_2 | (48, 46, 128) | 512 | MAX_POOLING_2D_1 |
| CONV_2D_3 | (48, 46, 256) | 295168 | BATCHNORM_2 |
| Nf = 256, ks = (3x3) | |||
| MAX_POOLING_2D_3 | (24, 23, 256) | 0 | CONV_2D_3 |
| BATCHNORM_3 | (96, 92, 256) | 1024 | MAX_POOLING_2D_3 |
| CONV_2D_4 | (24, 23, 512) | 1180160 | BATCHNORM_3 |
| Nf = 512, ks = (3x3) | |||
| MAX_POOLING_2D_4 | (12, 12, 512) | 0 | CONV_2D_4 |
| BATCHNORM_4 | (12, 12, 512) | 2048 | MAX_POOLING_2D_3 |
| GLOBAL_MAX_POOL_2D | (512) | 0 | BATCHNORM_4 |
| DENSE_1 | (512) | 262656 | GLOBAL_MAX_POOL_2D |
| BATCHNORM_4 | (512) | 2048 | DENSE_1 |
| DENSE_2 | (128) | 65664 | BATCHNORM_4 |
| BATCHNORM_5 | (128) | 512 | DENSE_2 |
| DENSE_3 | (64) | 8256 | BATCHNORM_5 |
| BATCHNORM_6 | (64) | 256 | DENSE_3 |
| Total params: 1,940,480 | |||
| Trainable params: 1,937,152 | |||
| Non-trainable params: 3,328 |
Notes. Encoder architecture of our Siamese neural network model. Nf, and ks stand for number of filters, and kernel size. DENSE_1 layer is the representation space of eCALIFA galaxies. This vector is then projected via fully connected layers to the constrastive space (BATCHNORM_6) where the loss function is computed.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.