TABLE V.
Optimal hyperparameters based on simulated anechoic cyst training CNR for CNR regularized DNN beamforming
| Hyperparameter | Value |
|---|---|
| Batch Size | 5496 |
| Number of Hidden Layers | 6 |
| Layer Width | 638 |
| Input Gaussian Noise | False |
| Input Dropout Probability | 0.3 |
| Hidden Node Dropout Probability | 0.2 |
| Batch Norm | True |
| Loss Function | L2 |
| Weight Decay | 0 |