Table 2 Estimated parameters and hyperparameter setup of DeepATsers model. Weights shape (x,y,z) where x,y—kernel size, z—# of filters. A/B—A for binary classification, B for multi-class (5) classification.
From: DeepATsers: a deep learning framework for one-pot SERS biosensor to detect SARS-CoV-2 virus
Parameters | Values | Hyperparameters | Values |
|---|---|---|---|
1st layer weights shape | (8, 1, 16) | Hidden layers | 2 |
(weights, bias) | (128, 16) | Kernel size | 8 |
Trainable parameters | 144 | Activation functions | tanh, Softmax |
Batch normalization | 64 | Stride | 1 |
2nd layer weights shape | (8, 16, 32) | Epochs | 100 |
(weights, bias) | (256, 32) | Drop out rate | 0.4 |
Trainable parameters | 4,128 | Learning rate | 0.001 |
Batch normalization | 128 | Optimizer | Adam |
Full connected layer weights shape | (42144, 16) | Batch size | 32 |
(weights, bias) | (674304, 16) | Loss function | SCCE |
Trainable parameters | 674,320 | Input size | 1\(\times\)1,331 |
Output layer weights shape | (16, 2/5) | Weight initialization | Xavier |
(weights, bias) | (32, 2/5) | Dropout layer | 1 |
Trainable parameters | 34/85 | Flatten layer | 1 |
Total trainable parameters | 678,722/678,773 | Full connected layer | 2 |
Total optimizable parameters | 1,357,446/1,357,548 | Feature maps | 16 and 32 |
