Table 1 Hyperparameter configuration and limits for tuning the shallow-CNN model
From: Explainable deep-neural-network supported scheme for tuberculosis detection from chest radiographs
Hyper-parameter | Range |
|---|---|
Kernel Size | [3 to 11] |
Number of filters in convolution layers | [16 to 128] |
Kernel Stride | [1 to 5] |
Pooling Method | [MaxPooling, AveragePooling, GlobalMaxPooling] |
Number of units in dense layer-1 | [128 to 1024] |
Learning rate | [0.1 to 0.001] |
Optimizer | [Adam, AdaGrad, AdaDelta, SGD] |