原问题1:What is the advantage of generative adversarial networks compared with other generative models?
生成对抗网络相比其他生成模型的优点?
Ian Goodfellow回答:
—————————————————————————————————–
相比其他所有模型,我认为
从实际结果来看,GAN看起来能产生更好的生成样本
GAN框架可以训练任何生成网络(在理论实践中,很难使用增强学习去训练有离散输出的生成器),大多数其他架构需要生成器有一些特定的函数形式,就像输出层必须是高斯化的. 另外所有其他框架需要生成器整个都是非零权值(put non-zero mass everywhere),然而,GANs可以学习到一个只在靠近真实数据的地方(神经网络层)产生样本点的模型( GANs can learn models that generate points only on a thin manifold that goes near the data.)
原问题2: What are the pros and cons of using generative adversarial networks (a type of neural network)?
生成对抗网络(一种神经网络)的优缺点是什么?
It is known that facebook has developed a means of generating realistic-looking images via a neural network. They used “GAN” aka “generative adversarial networks”. Could this be applied generation of other things, such as audio waveform via RNN? Why or why not?
facebook基于神经网络开发了一种可以生成现实图片的方法,他们使用GAN,又叫做生成对抗网络,它能应用到其他事物的生成吗,例如通过RNN生成音频波形,可以吗?为什么?
Ian Goodfellow回答:
—————————————————————————————————–
优势