Realizing GANs via a Tunable Loss Function
Realizing GANs via a Tunable Loss Function
复制标题
DOI：
10.1109/itw48936.2021.9611499
复制
发表时间：
2021-06
期刊：
影响因子：
--
通讯作者：
Gowtham R. Kurri;Tyler Sypherd;L. Sankar
中科院分区：
文献类型：
--
作者：
Gowtham R. Kurri;Tyler Sypherd;L. Sankar
我们引入了一个可调的GAN，称为$\alpha$-GAN，由$\alpha\in$（0，$\infty$]参数化，它在各种f-GAN和基于积分概率度量的GAN之间插值（在约束条件下）。我们使用监督损失函数构建$\alpha-$ GAN，即$\alpha-$ loss，这是一个可调的损失函数，可以捕获几个典型的损失。我们证明了$\alpha-$ GAN与Arimoto散度密切相关，Arimoto散度首先由Österriecher（1996）提出，后来由Liese和Vajda（2006）研究。我们认为，$\alpha-$ GAN引入的整体理解将具有解决消失梯度和模式崩溃问题的实际好处。
We introduce a tunable GAN, called $\alpha$-GAN, parameterized by $\alpha\in$(0, $\infty$], which interpolates between various f-GANs and Integral Probability Metric based GANs (under constrained discriminator set). We construct $\alpha-$ GAN using a supervised loss function, namely, $\alpha-$ loss, which is a tunable loss function capturing several canonical losses. We show that $\alpha-$ GAN is intimately related to the Arimoto divergence, which was first proposed by Österriecher (1996), and later studied by Liese and Vajda (2006). We posit that the holistic understanding that $\alpha-$ GAN introduces will have practical benefits of addressing both the issues of vanishing gradients and mode collapses.