Path Loss Prediction on LOGO-CV Blind 6 holdouts (test)
7.4RMSE Mean (dB)Heteroscedastic Neural Network (Shared)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Heteroscedastic Neural Network (Shared)Architecture=Shared, Number of hidden layers=2, Neurons per layer=64, Total parameters=45462025.11 | 7.4 | 0.1 | 95.1 | 0.4 | 29.6 | — | |
| ML (MSE loss)Architecture=MSE-optimized three-feature ML model, Loss Function=MSE2025.11 | 7.4 | 0.7 | 93.2 | 2.7 | 27 | — | |
| Heteroscedastic Neural Network (Partial)Architecture=Partial, Internal shared layer=1 hidden layer (45 neurons), Independent head=1 hidden layer (45 neurons), Total parameters=44122025.11 | 7.7 | 0.2 | 94.7 | 0.6 | 30 | — | |
| Heteroscedastic Neural Network (Independent)Architecture=Independent, Independent network depth=2 layers, Neurons per layer=45, Total parameters=45922025.11 | 8.3 | 1.5 | 95 | 1.2 | 32.5 | — |