Table 5 The impact of different LSTM layers (1, 2, 3) on model performance.
From: Channel equalization in ultraviolet communication based on LSTM-DNN hybrid model
LSTM layers | Training time (seconds/epoch) | The final accuracy of the test set | Model complexity |
|---|---|---|---|
1-layer | 3.5 | 91.8% | Simple structure, limited extraction of temporal information |
2-layer | 4.1 | 93.1% | Moderate, capable of effectively extracting temporal features |
3-layer | 5.7 | 92.3% | Training for too long can lead to overfitting |
