TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Diffusion Normalizing Flow

43 点作者 lnyan超过 3 年前

2 条评论

igorkraw超过 3 年前
Need to read this one, if you are the authors can I ask you to delineate it against Song et al. &quot;Score-Based Generative Modeling through Stochastic Differential Equations&quot; <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2011.13456" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2011.13456</a> ?
marmaduke超过 3 年前
Pretty cool in fact all the methods they demo there show good approximation of difficult distributions (if that looks easy, take a look at scikit learn’s manifold doc page). The shoe that hasn’t dropped in the article is behavior in high dimensions. For instance, I seem to recall that backwards integration of high dimensional DEs is unstable (not to mention memory issues).