Table 1.
Overview of trained networks.
| Double | Quad | θE, min [″] | wθE | Loss | Epoch | rlearn | N | Seed | Section | Figures |
|---|---|---|---|---|---|---|---|---|---|---|
| Natural distribution of Einstein radii of lenses | ||||||||||
| ✓ | ✓ | 0.5 | 1 | 0.0201 | 115 | 0.005 | 64 | 3 | 4.1 | |
| ✓ | ✓ | 0.5 | 5 | 0.0496 | 123 | 0.001 | 64 | 3 | 4.1 | 6, 7, 8, 13, 14 |
| ✓ | ✓ | 2.0 | 1 | 0.0120 | 85 | 0.01 | 32 | 3 | 4.2 | |
| ✓ | ✓ | 2.0 | 5 | 0.0209 | 85 | 0.008 | 32 | 2 | 4.2 | 7, 8, 9, 13, 14 |
| ✓ | 0.5 | 1 | 0.0193 | 242 | 0.008 | 64 | 1 | 5.1 | ||
| ✓ | 0.5 | 5 | 0.0474 | 117 | 0.001 | 64 | 3 | 5.1 | ||
| ✓ | 2.0 | 1 | 0.0118 | 163 | 0.05 | 64 | 3 | 5.1 | ||
| ✓ | 2.0 | 1 | 0.0118 | 96 | 0.01 | 32 | 2 | 5.1 | ||
| ✓ | 2.0 | 5 | 0.0217 | 62 | 0.008 | 32 | 3 | 5.1 | ||
| ✓ | 0.5 | 1 | 0.0193 | 151 | 0.008 | 32 | 2 | 5.1 | ||
| ✓ | 0.5 | 5 | 0.0441 | 69 | 0.001 | 32 | 2 | 5.1 | ||
| ✓ | 2.0 | 1 | 0.0129 | 267 | 0.01 | 64 | 2 | 5.1 | ||
| ✓ | 2.0 | 5 | 0.0268 | 285 | 0.005 | 32 | 1 | 5.1 | ||
| Uniform distribution of Einstein radii of lenses | ||||||||||
| ✓ | ✓ | 0.5 | 1 | 0.0223 | 147 | 0.001 | 32 | 1 | 4.3 | |
| ✓ | ✓ | 0.5 | 5 | 0.0528 | 112 | 0.0005 | 64 | 2 | 4.3 | 7, 8, 10, 11, 13, 14 |
| ✓ | 0.5 | 1 | 0.0288 | 73 | 0.008 | 64 | 2 | 5.1 | ||
| ✓ | 0.5 | 5 | 0.0688 | 56 | 0.001 | 32 | 2 | 5.1 | 12 | |
Notes. The first and second columns indicate if quads and/or doubles are included in the data set. The parameter θE, min represents the lower limit on the Einstein radius in the simulation, and wθE is the weighting factor of the Einstein radius in the loss function. The other parameters (lens center, ellipticity) are always weighted by a factor of 1 and the sum of all five weighting factors is normalized to the number of parameters. The fifth and sixth columns give the value of the loss of the test set and the epoch with the best validation loss. This is followed by the specific hyperparameters: learning rate rlearn, batch size N, and seed for the random number generator. The last two columns list the sections and the figures that present the results of the corresponding network.
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.