Neural network regularization

Pemodelan Citra dengan Keras

Ariel Rokem

Senior Data Scientist, University of Washington

Dropout

In each learning step:

Select a subset of the units
Ignore it in the forward pass
And in the back-propagation of error

Dropout

Dropout in Keras

from keras.models import Sequential
from keras.layers import Dense, Conv2D, Flatten, Dropout

model = Sequential()
model.add(Conv2D(5, kernel_size=3, activation='relu', 
              input_shape=(img_rows, img_cols, 1)))

model.add(Dropout(0.25))

model.add(Conv2D(15, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(3, activation='softmax'))

Batch normalization

Rescale the outputs

Batch Normalization in Keras

from keras.models import Sequential
from keras.layers import Dense, Conv2D, Flatten, BatchNormalization 


model = Sequential()
model.add(Conv2D(5, kernel_size=3, activation='relu', 
              input_shape=(img_rows, img_cols, 1)))

model.add(BatchNormalization())

model.add(Conv2D(15, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(3, activation='softmax'))

Be careful when using them together!

The disharmony between dropout and batch normalization

Let's practice!

Pemodelan Citra dengan Keras