Table 1
Range of parameters tested for the fully connected NN and optimal value found for PCA with 12 components and using a CAE with a bottleneck layer of 48, 64, and 71 (CAE48, CAE64, and the optimal CAE71 we call iNNterpol).
| nl Layers | nn Neurons/layer | Activation | Batch size |
|---|---|---|---|
| 8–22 | 20–80 | relu/lkrelu/elu | 256–32 |
| Optimal values | |||
| (PCA36,48-NN) | |||
| 12 | 40 | lkrelu | 64 |
| (CAE48-NN) | |||
| 16 | 48 | lkrelu | 64 |
| (CAE71-NN | |||
| or iNNterpol) | |||
| 16 | 71 | elu | 64 |
Notes. This all pertains to ATLAS9 data; for more details on MARCS and PHOENIX, readers can refer to Sect. 3.4. The activation functions referred to are rectified linear unit (relu), leaky rectified linear unit (lkrelu), and exponential linear unit (elu).
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.