GAN

（两个Noise可以是不同的）

Structured Learning:

output sequence, matrix,....

why challenging? one-shot/zero-shot learning; machine has to learning to do planning

why cannot only generator?

the relation between the components are critical. Although highly correlated, they cannot influence each other.

can discriminator generate?

negative examples are critical.

Conditional GAN:

discriminator要同时看generator的input condition和output！ 

Stack GAN:先产生小张的图，再产生大图

 

unsupervised conditional generation:

 

cycle GAN

 

star GAN: multi-domain

 

图片均出自李宏毅老师的课程。