Table B.1
Best found hyperparameters of the network NN1 (SEDs) and NN2 (correlated fluxes).
| Hyperparameter | NN1 for SEDs | NN2 for corr. Flux |
|---|---|---|
| Hidden layers | 4 | 6 |
| Neurons in | 360, 350, | 6440, 1430, 1540, |
| hidden layers | 350, 200 | 3840, 5030, 19080 |
| Output neurons | 18 | 18 × 171 |
| Activation of hidden layers | First layer: tanh; then ReLU | ReLU |
| Activation of output | Linear | tanh |
| Epochs | 180 | 250 |
| Learning rate | 150 epochs: 2.8 × 10−4; next 30 epochs: 1.8 × 10−7 | 200 epochs: 5.1 × 10−5; next 50 epochs: 1.1 × 10−5 |
Notes. The number of artificial neurons is arranged in ascending order of the layer numbers. The input layers of both networks have four neurons (number of model parameters). The output layer of the network NN1 has twelve neurons (number of simulated wavelengths) and the output layer of the NN 2 has 18 × 171 output neurons: one for each combination of observing wavelength and baselength.
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.