AI 与机器学习2021arxiv.org编辑精选
扩散模型在图像合成上胜过 GAN
Diffusion Models Beat GANs on Image Synthesis
进阶 · 约 90 分钟读完前置知识：扩散模型基础、GAN 基础、U-Net 架构
导读
这篇 OpenAI 的论文（即 ADM 论文）研究扩散概率模型在图像合成上的表现，证明通过改进架构（如使用带分类器引导的 U-Net），扩散模型在多个基准上超越了当时最强的 GAN（如 BigGAN），首次确立了扩散模型在图像生成质量上的领先地位。
HN 关注点：扩散模型与 GAN 两大生成模型路线之争，讨论焦点集中在生成质量与推理速度的权衡、标题是否夸大，以及对模型不开源的批评。
后来怎么样
这篇 ADM 论文确立了扩散模型在图像生成质量上超越 GAN 的地位，直接引出了 Stable Diffusion、DALL-E 2、Imagen 等后续工作，使扩散模型成为图像生成的主流范式。
HN 评论精选
这几条评论里：支持 1 质疑 2 补充 1
支持扩散模型质量惊艳但推理比 GAN 慢 25 倍以上，且批评论文不开源模型
扩散模型最大的缺点是速度：GAN 在普通 nVidia GPU 上单核不到半秒就能渲染（StyleGAN 甚至能到 10fps 以上），只需一次前向传播；而扩散模型需要 25 次以上前向传播，至少慢 25 倍，我们实测往往要几秒到一分钟才出结果。不过这篇论文的样本是我见过最逼真、最复杂的照片级图像，即使有挑选成分也瑕不掩瑜。最后我很失望模型又没开源——囤积 ML 模型的风气该停止了。这些研究模型没有商业价值，但对 ML 圈外的人有巨大好处，我自己就是因为能玩 GPT-2 和 StyleGAN 才入行的。不发布的各种借口都站不住脚，据我所知没有任何模型造成过威胁社会的实际影响。
sillysaurusx查看英文原文
The biggest downside with diffusion models is that GANs can be rendered in much less than half a second (sometimes 10fps or higher) on one core of a standard device you probably have. (Many of you have an nVidia GPU, and any ol’ nVidia GPU will render stylegan quickly.) With diffusion models, you need to do >25 forward passes to achieve a result. It’s kind of like an O(1) algorithm vs O(N): stylegan has one pass, diffusion models have N. And N is currently 25 or more, which means it tends to be 25x slower than stylegan at a minimum. (In our experience it was often many seconds to a full minute before we saw results, but we didn’t try very hard to make it fast, and this paper shows advances in speed since then.) The flipside: this paper has the most beautiful photorealistic complex samples I’ve ever seen. I don’t really care if they’re cherry picked; it’s hard to pick cherries on a rotten tree. I know. The last thing I want to say is that I am disappointed the model is, yet again, not released. The hoarding of ML models needs to stop. These research models aren’t commercially useful, but they have incredible benefits for people outside of ML. I myself got into ML thanks to being able to play with GPT-2 and Sfylegan. I would leap at the chance to play with AlphaZero or OpenAI’s old Dota 2 models; no reason to keep those locked up. I also don’t care about the excuses for not releasing: none of them hold water. In my experience, no model has ever had a society-threatening impact, and it’s usually a euphemism for “we’re worried there’s a small chance we might look bad.” Releasing your model is as simple as scp’ing to your server; just do it and quit worrying so much. Love the work. Diffusion models are a really interesting way of thinking about generative approaches in ML.
质疑质疑标题以偏概全：只是特定实现胜出，不代表整类模型的优势
为什么要笼统地宣称一整类生成模型胜过另一类？他们只是展示了自己特定的扩散模型实现在某些指标上优于当前最好的 GAN 实现。几乎可以肯定（很可能很快），会有人投入时间调优，从 GAN/VAE/flow 里再榨出一些性能，届时这个标题就过时了。
ansk查看英文原文
Why broadly claim an entire class of generative model outperforms another? They simply show that their particular implementation of a diffusion model attains better performance (for some metric) than the current best implementation of a GAN. Almost inevitably (and likely in the very near future) someone will invest the time and tuning to squeeze out some more performance from a GAN/VAE/flow, and this title will be outdated.
补充请求推荐学习扩散模型的入门资料
也许要求有点多，但有人能给我讲解或推荐一些材料，帮我理解扩散概率模型背后的原理吗？我的统计学知识太基础，连这些论文都读不懂。
mateo1查看英文原文
Maybe that's a lot to ask, but can someone explain to me or guide me to material to gain a better understanding of principles behind diffusion probabilistic models? My knowledge of statistics is just too basic to even read these papers.
质疑主观感受：生成图像的瑕疵反而更诡异
我不太懂为什么他们的模型更好，但他们图像中的瑕疵比其他算法的更让人不安。
colordrops查看英文原文
I don't know enough about this to understand why their model is better, but the imperfections in their images are much more unsettling that other algorithms.
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2021Hacker News 讨论帖209 分 · 50 评
同领域还值得读
DeepSeek-R1：通过强化学习激励 LLM 的推理能力
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
研究如何用强化学习激励大语言模型的推理能力，提出 DeepSeek-R1，并探讨 RL 训练对模型推理表现的影响。
AI 与机器学习2025 · 1,351 分 · 1,056 评 · 引用 30进阶 · 约 80 分钟编辑精选1-bit LLM 时代：面向低成本计算的三值参数
The Era of 1-bit LLMs: ternary parameters for cost-effective computing
提出用三值参数 {-1, 0, 1} 替代 LLM 中的浮点权重，即所谓“1-bit LLM”。
AI 与机器学习2024 · 1,040 分 · 447 评 · 引用 19进阶 · 约 45 分钟编辑精选DeepSeek-V3.2：推进开放大语言模型的前沿
DeepSeek-v3.2: Pushing the frontier of open large language models
这是 DeepSeek-V3.2 的技术报告，延续了 DeepSeek 开源权重模型的路线，目标是让开放模型继续逼近甚至追平前沿闭源模型。
AI 与机器学习2025 · 982 分 · 465 评进阶 · 约 90 分钟编辑精选通过自我对弈与通用强化学习掌握国际象棋和将棋（AlphaZero）
Mastering Chess and Shogi by Self-Play with General Reinforcement Learning
DeepMind 的 AlphaZero 论文：用通用强化学习算法（自我博弈 + MCTS + 神经网络），从零开始仅数小时训练便在国际象棋和将棋上击败了世界顶级引擎 Stock…
AI 与机器学习2017 · 539 分 · 270 评 · 引用 13进阶 · 约 90 分钟编辑精选用强化学习发现更快的矩阵乘法算法
Discovering faster matrix multiplication algorithms with reinforcement learning
论文提出 AlphaTensor，用强化学习搜索矩阵乘法算法。
AI 与机器学习2022 · 518 分 · 111 评专业 · 约 90 分钟编辑精选