CS231n学习笔记--13. Generative Models

1. Unsupervised Learning

Supervised vs Unsupervised Learning:

2. Generative Models

概述：

**Generative Models的作用：**

**Generative Models的分类：**

3. PixelRNN and PixelCNN

基本原理：

**PixelRNN：**

**PixelCNN：**

Training is faster than PixelRNN (can parallelize convolutions since context region values known from training images)Generation must still proceed sequentially=> still slow**Generation Samples：**

**PixelRNN and PixelCNN**

4. Variational Autoencoders (VAE)

4.1 与PixelRNN/PixelCNN的比较：

**4.2 Some background first: Autoencoders：**

**Tips:**如果将其用于特征提取，则在训练之后，将decoder部分丢弃！

Autoencoders can reconstruct data, and can learn features to initialize a supervised model！**4.3 Variational Autoencoders**

利用高斯分布随机生成特征Z：

**Variational Autoencoders: Intractability**

pθ(z)跟据高斯分布随机获得，pθ(x|z)根据decoder net获得，而为每个z计算pθ(x|z)并最终积分得到pθ(x)是不可能的！解决办法：

如何进行优化：

**4.4 Generating Data!**

**4.5 性能分析：**

5. Generative Adversarial Networks (GAN)

回顾：

**5.1 Training GANs: Two-player game****Generator network:** try to fool the discriminator by generating real-looking images**Discriminator network:** try to distinguish between real and fake images

网络优化：

优化存在的问题：

解决办法：

GAN training algorithm：

5.2 Generative Adversarial Nets

Generated samples:

**Generative Adversarial Nets: Convolutional Architectures**Generator is an upsampling network with fractionally-strided convolutionsDiscriminator is a convolutional network

Generator网络结构：

Samples from the model look amazing!

**Generative Adversarial Nets: Interpretable Vector Math**

**GANs的优缺点：**

6. 回顾：