“recently published a neat preprint on using adversarial networks as part of the training process.”
Generative Adversarial Networks
生成器与判别器博弈训练的生成模型框架
提出通过对抗过程估计生成模型的新框架:同时训练生成模型 G 来捕捉数据分布,以及判别模型 D 来估计样本来自训练数据而非 G 的概率;G 的目标是让 D 犯错,这对应一个极小极大两人博弈。在任意函数空间中存在唯一解,即 G 复原训练数据分布且 D 处处等于 1/2;当二者是多层感知机时,整个系统可用反向传播训练,无需马尔可夫链或展开的近似推断。实验从定性和定量上展示了该框架的潜力。
评论者把它当作 GAN 的原始出处,常在讨论 AI 生成图像、深度伪造、AI 艺术署名和对抗训练时引用,称这篇论文读起来出奇地容易。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“otherwise, the [original paper] is refreshingly easy to read.”
“Probably also canonical are Goodfellow's Deep Learning [2], Koller & Friedman's PGMs [3], the Krizhevsky ImageNet paper [4], the original GAN [5]”
“your idea is called adversarial networks, and it works quite well”
“Edit: a more precise description in the abstract here: Generative Adversarial Networks”
“So this is to counter stuff like GAN's (generative adversarial networks) [1] to reverse engineer data out of a black box systems?”
看全部 15 条 →
“This is likely part of how the speech-generating NNs are trained (ie. there's a generated-speech-detector and the network is trained to fool it, while it is also trained): Generative Adversarial Networks”
“Should the original GAN paper ( Generative Adversarial Networks ? ) be given artistic credit here ?”
“The idea, the Generative Adversarial Network, was published by Goodfellow et al. [1] in 2014”
“[2] Ian Goodfellow's original GAN paper: Generative Adversarial Networks”
“The arms race was first automated in 2014, when the GAN paper was published.”
“I posted a list of papers on twitter, and will be posting a summary for each of them as well.”
“They were doing fuzzy black and white 28x28 faces, poorly, and 32x32 color images that if you squint hard enough you could see a dog.”
“The predecessor to DDPM (the work that kicked off the diffusion model era) was published in 2015[0], only a year after GANs[1].”
“10 years ago the GAN paper came out and everyone was excited how amazing the generated image quality was”
- Generative Adversarial Nets [pdf] ↗peterkshultz · 2016-11-06 · 1 分
- Generative Adversarial Networks ↗Anon84 · 2023-05-30 · 2 分