Variational Auto-Encoder Example

最新推荐文章于 2024-03-04 12:50:32 发布

原创

最新推荐文章于 2024-03-04 12:50:32 发布 · 360 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#tensorflow

本教程通过构建变分自编码器(VAE)，利用TensorFlow从噪声分布中生成MNIST手写数字图像。该模型包含编码器和解码器，能够学习数据的潜在表示并从中采样生成新的图像。

Variational Auto-Encoder Example

Build a variational auto-encoder (VAE) to generate digit images from a noise distribution with TensorFlow.

Author: Aymeric Damien
Project: https://github.com/aymericdamien/TensorFlow-Examples/

在这里插入图片描述

VAE Overview

References:

Auto-Encoding Variational Bayes The International Conference on Learning Representations (ICLR), Banff, 2014. D.P. Kingma, M. Welling
Understanding the difficulty of training deep feedforward neural networks. X Glorot, Y Bengio. Aistats 9, 249-256

MNIST Dataset Overview

This example is using MNIST handwritten digits. The dataset contains 60,000 examples for training and 10,000 examples for testing. The digits have been size-normalized and centered in a fixed-size image (28x28 pixels) with values from 0 to 1. For simplicity, each image has been flattened and converted to a 1-D numpy array of 784 features (28*28).

MNIST Dataset
Tensorflw快速入门——运用逻辑回归算法手写数字识别

from __future__ import division, print_function, absolute_import

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
import tensorflow as tf

Import MNIST data

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

Parameters

learning_rate = 0.001
num_steps = 30000
batch_size = 64

# Network Parameters
image_dim = 784 # MNIST images are 28x28 pixels
hidden_dim = 512
latent_dim = 2

# A custom initialization (see Xavier Glorot init)
def glorot_init(shape):
    return tf.random_normal(shape=shape, stddev=1. / tf.sqrt(shape[0] / 2.))

Variables

weights = {
   
   
    'encoder_h1': tf.Variable(glorot_init([image_dim, hidden_dim])),
    'z_mean':

最低0.47元/天解锁文章