Too strong regularization for an autoencoder (Keras autoencoder tutorial code)

Question

I'm using this tutorial about autoencoders: https://blog.keras.io/building-autoencoders-in-keras.html

All the code is working, however the performance is very bad (the results are blurred) when I set 10e-5 for the regularization parameter, which is the parameter defined in the tutorial code. In fact, I need to decrease the regularization to 10e-8 to have a correct output.

My question is as follows: Why the result is so different from the tutorial? Data is the same and parameters are the same, I didn't expect a large difference.

I suspect that the default behavior of the Keras functions has been changed (automatic batch normalization performed in all cases?) from May 14th, 2016.

Outputs

• With 10e-5 regularization (blurred); val_loss of 0.2967 after 50 epochs and 0.2774 after 100 epochs.
• With 10e-8 regularization: val_loss of 0.1080 after 50 epochs and 0.1009 after 100 epochs.
• With no regularization: val_loss of 0.1018 after 50 epochs and 0.0944 after 100 epochs.

Complete code (for reference)

# Source: https://blog.keras.io/building-autoencoders-in-keras.html
import numpy as np
np.random.seed(2713)

from keras.layers import Input, Dense
from keras.models import Model
from keras import regularizers

encoding_dim = 32

input_img = Input(shape=(784,))
# add a Dense layer with a L1 activity regularizer
encoded = Dense(encoding_dim, activation='relu',
                activity_regularizer=regularizers.l1(10e-5))(input_img)
decoded = Dense(784, activation='sigmoid')(encoded)

autoencoder = Model(input_img, decoded)

# this model maps an input to its encoded representation
encoder = Model(input_img, encoded)

# create a placeholder for an encoded (32-dimensional) input
encoded_input = Input(shape=(encoding_dim,))
# retrieve the last layer of the autoencoder model
decoder_layer = autoencoder.layers[-1]
# create the decoder model
decoder = Model(encoded_input, decoder_layer(encoded_input))

autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')

from keras.datasets import mnist
(x_train, _), (x_test, _) = mnist.load_data()

x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))
print(x_train.shape)
print(x_test.shape)

autoencoder.fit(x_train, x_train,
                epochs=100,
                batch_size=256,
                shuffle=True,
                validation_data=(x_test, x_test))

# encode and decode some digits
# note that we take them from the *test* set
encoded_imgs = encoder.predict(x_test)
decoded_imgs = decoder.predict(encoded_imgs)

# use Matplotlib (don't ask)
import matplotlib.pyplot as plt

n = 10  # how many digits we will display
plt.figure(figsize=(20, 4))
for i in range(n):
    # display original
    ax = plt.subplot(2, n, i + 1)
    plt.imshow(x_test[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

    # display reconstruction
    ax = plt.subplot(2, n, i + 1 + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)
plt.show()

Answer 1

I have the same issue. And it is on GitHub here https://github.com/keras-team/keras/issues/5414 It seems like you were correct in just changing the constant.