ReLU

Sigmoid

Tanh

Softmax

By reducing the size of the dataset

By adding noise to the input data

By randomly deactivating neurons during training

By increasing the number of training epochs

Increasing the size of the network

Using data augmentation

Decreasing the learning rate

Reducing the number of layers

Data augmentation

Dropout

Convolutional layers

Regularization

To prevent overfitting

To preserve spatial dimensions of the input data

To speed up training

To increase the number of filters

L1 Regularization

L2 Regularization

Dropout

Data Augmentation

Reduced training time

Improved performance on a related task

Increased model complexity

Better utilization of pre-trained models

The size of the input data

The size of the convolutional kernel

The portion of the input that each neuron in the feature map is sensitive to

The number of feature maps produced