基於變分的自編碼器(variational Autoencoder)是2014由Kingma等人提出的一種基於變分貝葉斯推斷的生成網絡。與傳統的生成網絡相比它具有兩點優勢:
1。避免了複雜的邊界似然概率(太難了,誰會算啊)
2。避免了馬爾科夫鏈的採樣過程(好像最近很火生成對抗網絡也避免了這個)
1. 生成器網絡
對於一堆我們收集到的樣本,這些樣本在默認情況下來自一個獨立的同分布,也就是我們常說的iid。我們不妨把這個概率分佈記作
2. 邊分邊界
當我有了兩個概率分佈的評價模型以後,需要做的就是優化。首先將KL散度進行拆分:
根據貝葉斯公式,
其中,中間含有
假設我們的概率優化的夠好,讓我們的變分下界儘可能的大,那麼我們的KL散度就會越來越小,兩個概率分佈的相似性就會越來越好。
3. 變分自編碼器
上面的式子可以簡寫成
針對自編碼器的結構,我們採用多維標準高斯進行逼急z的後驗分佈
結合keras中的vae代碼進行分析:
@author: sky
"""
'''This script demonstrates how to build a variational autoencoder with Keras.
Reference: "Auto-Encoding Variational Bayes" https://arxiv.org/abs/1312.6114
'''
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
from keras.layers import Input, Dense, Lambda
from keras.models import Model
from keras import backend as K
from keras import objectives
from keras.datasets import mnist
batch_size = 100
original_dim = 784
latent_dim = 2
intermediate_dim = 256
nb_epoch = 50
epsilon_std = 1.0
#定義模型初始化
x = Input(batch_shape=(batch_size, original_dim))
h = Dense(intermediate_dim, activation='relu')(x)
z_mean = Dense(latent_dim)(h)
z_log_var = Dense(latent_dim)(h)
#建立一個連接模型,這裏的z_mean代表了高斯分佈的均值,z_log_var代表了高斯分佈方差的對數表達。
def sampling(args):
z_mean, z_log_var = args
epsilon = K.random_normal(shape=(batch_size, latent_dim), mean=0.,
std=epsilon_std)
return z_mean + K.exp(z_log_var / 2) * epsilon
#建立一個採樣方程,最終返回一個生成的z
# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])
decoder_h = Dense(intermediate_dim, activation='relu')
decoder_mean = Dense(original_dim, activation='sigmoid')
h_decoded = decoder_h(z)
#將採樣後的z映射到隱含層中(可以參考AE)
x_decoded_mean = decoder_mean(h_decoded)
#對隱含層的h進行解碼得到重構樣本x
def vae_loss(x, x_decoded_mean):
xent_loss = original_dim * objectives.binary_crossentropy(x, x_decoded_mean)
kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
return xent_loss + kl_loss
#定義整個模型的損失函數,包括了讓採樣儘可能的靠近標準高斯分佈和隱含層的解碼值與原始樣本更加接近。
vae = Model(x, x_decoded_mean)
vae.compile(optimizer='rmsprop', loss=vae_loss)
# train the VAE on MNIST digits
(x_train, y_train), (x_test, y_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:])))
vae.fit(x_train, x_train,
shuffle=True,
nb_epoch=nb_epoch,
batch_size=batch_size,
validation_data=(x_test, x_test))
這樣就可以實現VAE的訓練