motivation

Conditional GAN需要大量标签
GAN训练不稳定：GANs are typically trained using alternating stochastic gradient descent which is often unstable and lacks theoretical guarantees
instability, divergence, cyclic behavior, or mode collapse
GAN训练过程中D模块的遗忘现象
- In non-stationary online environments, neural networks forget previous tasks
- training may become unstable or cyclic
- This issue is usually addressed either by reusing old samples or by applying continual learning techniques
- These issues become more prominent in the context of complex data sets - > conditioning（labeled data）
- CGAN中标记数据可以帮助D模块训练更稳定的表达，并且CGAN对于单一类别训练易于整体数据集训练

contribution

take a step towards bridging the gap between conditional and unconditional GANs，Under the same conditions, the self-supervised GAN attains a similar performance to state-of-the-art conditional counterparts
unsupervised generative model that combines adversarial training with self supervised learning
add an auxiliary, self-supervised loss to the discriminator

original value function for GAN

训练过程中PG(x)参数更新，因此导致D模块在线学习不稳定
训练过程中D模块往往会根据G模块学习到的局部特征（纹理、边缘、结构等）来进行惩罚，因此难以学习到整体的有效表达（局部陷阱）

discriminator is not incentivised to maintain a useful data representation as long as the current representation is useful to discriminate between the classes
训练后期PG = Pdata，D模块输出为常数0.5，因此D模块将无法继续学习到有效信息。

同时如果训练过程中加入正则化，D模块可能会忽略有效信息而专注于次要信息，因此无法正确辨别
常规Uncond-GAN的遗忘现象（500k后遗忘有效表达）
mnist测试，尽管任务相似，训练类别切换时仍存在遗忘现象

generator and discriminator collaborate on the task of representation learning, and compete on the generative task
D模块通过预测真实数据旋转角度进行训练
促进G模块生成具有可检测旋转的真实图片特征的表达
α>0不能保证收敛到PG = Pdata，因此训练过程中逐渐将α衰减至0
the generator is encouraged to generate images that are rotation-detectable because they share features with real images that are used for rotation classification