Open Access
Table 2
Summary of loss function choice, architecture and learning hyperparameters adopted for training our best-fit ffNN model, compared to those adopted by Scutt et al. (2023).
| Hyperparameter | Our choice | Scutt et al. (2023) |
|---|---|---|
| # of hidden layers | 6 | 6 |
| # of neurons per layer | 128 | 64 |
| Activation function | ReLU | ELU |
| Kernel initializer | GU | |
| Regularization | LN | – |
| Batch size | 512 | 6 × 104 |
| Optimizer | Adam | Adam |
| lr schedule | Exp. decay | Fixed lr |
| lr range | (10−3, 5 × 10−6) | 7 × 10−5 |
| Loss function | Huber loss | MSE |
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.