Open Access
Table 2
BGRem training hyperparameters and values.
| Hyperparameter | Values | Other information |
|---|---|---|
| Start learning rate | 0.001 (0.0008) | MeerLICHT data (Fermi-LAT data) |
| Learning rate decay | 0.95 | − |
| Early stopping | − | Validation loss doesn’t improve after 10 epochs |
| max_signal_rate | 0.0 | θstart = arccos (max_signal_rate) ≈ 0.2 |
| min_signal_rate | 0.98 | θend = arccos (min_signal_rate) ≈ 1.6 |
| Normalization factor | 0.01 | Further expanded in Appendix B |
| Diffusion steps | 6 | Further expanded in Appendix B |
| Optimizer | AdamW | Weight decay = 1e − 4 |
| Batch size | 32 (16) | MeerLICHT data (Fermi-LAT data) |
| Start epochs | 100 (150) | MeerLICHT data (Fermi-LAT data) |
| Total parameters in attention U-Net | 834 801 | Trainable: 834 609 |
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.