Table C.2.
Optimization of hyper-parameters.
| HYPERPARAMETER | Setting | Data augmentation | |
|---|---|---|---|
| ResNet, FitNet, IRCNN, EffectiveNet, and FitNet | Majority Voting Ensemble, Simple Averaging Ensemble, and Weighted Averaging Ensemble | ||
| Optimizer | ADAM | ADAM | Both axis side random reflection Rescaling randomly b/w[0. 5 to 1.50] Rotating randomly b/w [−40° 40°] |
| Batch Size | 10 | 10 | |
| Max Epoch | 200 | 100 | |
| Global Learning Rate | 4 | 4 | |
| Dropout rate | 0.5 | 0.8 | |
| Validation Frequency | 68 | 68 | |
| Learn Rate Factor | 10 | 10 | |
| classification layer weight vector | [0.75 0.15 1.18] | [0.75 0.15 1.18] | |