半监督的GAN算法

ImprovedGAN

$
Loss = Loss_{supervised}   + \lambda * Loss_{unsupervised}
$
第二项形式与原始的GAN模型类似。

参考： Improved Techniques for Training GANs
代码： https://github.com/Sleepychord/ImprovedGAN-pytorch/blob/master/ImprovedGAN.py