I run the notMNIST dataset on 7 layers DNN and test with different activation functions. The confusion matrix, classification report and AUC score are generated.
Activation functions:
- ReLU
- Swish
- Tanh
- LReLU
- LReLU (0.25)
- PReLU
- Softplus
- ELU
- FReLU (initialized at -0.398)
- FReLU
- Flatten-T Swish