Regularização em redes neurais

Modelagem de imagens com Keras

Ariel Rokem

Senior Data Scientist, University of Washington

Dropout

Em cada passo de aprendizado:

Selecione um subconjunto das unidades
Ignore no forward pass
E na retropropagação do erro

Dropout

Dropout no 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'))

Normalização em lote

Reescalonar as saídas

Batch Normalization no 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'))

Cuidado ao usar juntos!

A desarmonia entre dropout e batch normalization

Vamos praticar!

Modelagem de imagens com Keras