按年被提到
20142023 · 5 次2026
视觉、语音与图像生成›生成对抗网络·arXiv:1406.2661 · stat.ML

Generative Adversarial Networks

生成对抗网络
Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio · 2014

生成器与判别器博弈训练的生成模型框架

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 15 人首次 2014-06-19 · 最近 2024-09-26发表于 Communications of the ACM
15不同的人
16活跃月份
17讨论串
18,372被引(OpenAlex)

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。

HN 为什么提它

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。

奠基之作难度 3 / 5约 40 分钟经得起时间
读之前最好知道
1神经网络与反向传播
2概率分布与生成模型
3博弈论与极小极大问题
4多层感知机

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
推荐thisisdave · 2014-06

“recently published a neat preprint on using adversarial networks as part of the training process.”

推荐这篇预印本,把对抗网络用作训练过程的一部分
Does Deep Learning Have Deep Flaws?原评论 ↗
推荐vinn124 · 2018-01

“otherwise, the [original paper] is refreshingly easy to read.”

推荐读 GAN 原始论文,讲得清楚易读
How an A.I. ‘Cat-And-Mouse Game’ Generates Believable Fake Photos原评论 ↗
推荐lwneal · 2023-05

“Probably also canonical are Goodfellow's Deep Learning [2], Koller & Friedman's PGMs [3], the Krizhevsky ImageNet paper [4], the original GAN [5]”

把原始 GAN 论文列为 AI 领域的经典必读之一
引用m0g · 2015-10

“your idea is called adversarial networks, and it works quite well”

指出这个想法叫对抗网络,生成器与判别器相互对抗,效果很好
I made a computer pretend to be a computer that’s pretending to be a human原评论 ↗
引用westoncb · 2017-01

“Edit: a more precise description in the abstract here: Generative Adversarial Networks”

补充指出摘要里对这个双网络博弈的描述更精确
Game Theory Reveals the Future of Deep Learning原评论 ↗
引用itchyjunk · 2017-03

“So this is to counter stuff like GAN's (generative adversarial networks) [1] to reverse engineer data out of a black box systems?”

提问这是否用来对抗 GAN 从黑盒系统逆向还原数据
Building Safe AI: A Tutorial on Homomorphically Encrypted Deep Learning原评论 ↗
看全部 15 条 →
引用robryk · 2017-04

“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”

猜测语音生成网络的训练类似 GAN:生成器要骗过检测器
Lyrebird – An API to copy the voice of anyone原评论 ↗
引用crucialfelix · 2018-10

“Should the original GAN paper ( Generative Adversarial Networks ? ) be given artistic credit here ?”

在 AI 艺术归属争论中问是否该给原始 GAN 论文艺术署名
Christie’s sells its first AI portrait for $432k原评论 ↗
引用lwneal · 2020-06

“The idea, the Generative Adversarial Network, was published by Goodfellow et al. [1] in 2014”

指出 Goodfellow 等人 2014 年提出了 GAN 这一想法
AI vs. AI原评论 ↗
引用somethingAlex · 2021-04

“[2] Ian Goodfellow's original GAN paper: Generative Adversarial Networks”

作为参考文献引用 Goodfellow 的原始 GAN 论文
Critique of 2018 Turing Award for Drs. Bengio and Hinton and LeCun原评论 ↗
引用CGamesPlay · 2023-01

“The arms race was first automated in 2014, when the GAN paper was published.”

指出军备竞赛的自动化始于 2014 年 GAN 论文发表
How we’re approaching AI-generated writing on Medium原评论 ↗
引用vikashrungta · 2023-02

“I posted a list of papers on twitter, and will be posting a summary for each of them as well.”

把它列为介绍 GAN 的奠基性 AI 论文之一
What were the 40 research papers on the list Ilya Sutskever gave John Carmack?原评论 ↗
引用godelski · 2023-11

“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.”

用它说明早期 GAN 只能生成模糊的 28x28 人脸和勉强可辨的 32x32 图像
SDXL Turbo: A Real-Time Text-to-Image Generation Model原评论 ↗
引用godelski · 2024-02

“The predecessor to DDPM (the work that kicked off the diffusion model era) was published in 2015[0], only a year after GANs[1].”

指出 GAN 比扩散模型的前身晚一年发表,因画质更好把研究从扩散上吸引走
Mamba Explained: The State Space Model Taking On Transformers原评论 ↗
引用ml_basics · 2024-09

“10 years ago the GAN paper came out and everyone was excited how amazing the generated image quality was”

拿它说明十年前 GAN 的图像质量曾令人惊叹,标准一直在变
Xkcd 1425 (Tasks) turns ten years old today原评论 ↗
被提交为 story · 2 次