GAN

GAN

生成器 generator：G
判别器 discriminator：D
if G and D have enough capacity, they will reach a point at which both cannot improve because p_g = p_data. The discriminator is unable to differentiate between the two distributions, i.e. D(x) = 1/2

MLE = minimize KL divergence

=》

1. 固定G，update D with gradient ascent（最优D：最大D）
   

update：
通过JS divergence 度量p_data 和p_G 之间相似度（距离）

2. 寻找minimize JS

训练Generator

update G: gradiant descent

Problem

Don’t update G too much

JS divergence is log2 if two distributions do not overlap

add noise to the inputs of discriminator (decay over time)

mode collapse

Least Square GAN (LSGAN): Linear

replace sigmoid with linear (replace classification with regression)

WGAN: Earth Mover’s Distance