按年被提到
20142016 · 4 次2026
机器学习与深度学习›科学机器学习·arXiv:1410.3831 · stat.ML

An exact mapping between the Variational Renormalization Group and Deep Learning

变分重整化群与深度学习之间的精确映射
Pankaj Mehta, David J. Schwab · 2014

证明变分重整化群与基于 RBM 的深度学习之间存在精确映射

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 12 人首次 2014-12-04 · 最近 2025-05-20
12不同的人
10活跃月份
13讨论串
195被引(OpenAlex)

作者指出深度学习与理论物理中的重整化群(RG)密切相关,RG 是一种迭代的粗粒化方案,可在不同尺度上提取相关特征。他们构造了 Kadanoff 提出的变分重整化群与基于受限玻尔兹曼机(RBM)的深度学习架构之间的精确映射,并用一维和二维最近邻 Ising 模型加以说明,提示深度学习算法可能在用一种广义的类 RG 方案从数据中学习相关特征。

HN 为什么提它

它常被引用来说明深度学习与统计物理之间的深层联系,有人说它让自己真正理解了神经网络,也有人觉得它不过很可爱、并没有把问题变成物理,并质疑论据与对比散度训练的比较不足。

其他难度 5 / 5约 50 分钟经得起时间
读之前最好知道
1重整化群(renormalization group)与粗粒化
2受限玻尔兹曼机(RBM)
3Ising 模型与统计物理基础
4深度学习基础(逐层特征学习)

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
强烈推荐riemannzeta · 2025-05

“I "woke up" to this sort of approach to understanding deep neural networks when I read this paper”

强烈推荐,称读它才开窍:RBM 与重整化群精确对应,由此理解深度网络
Deep Learning Is Applied Topology原评论 ↗
推荐cs702 · 2017-08

“PS. You will enjoy the following papers too:”

推荐它,说明变分重整化群与深度学习存在精确映射
Deep Learning and Quantum Entanglement [pdf]原评论 ↗
批评thisisdave · 2014-12

“I don't see any comparison between the approximations that physicists normally use versus the contrastive divergence for training RBM-based networks, or evidence that their results are more similar in practice than any other technique.”

批评它缺少物理近似与对比散度训练 RBM 的比较,也没证据显示更相似
Renormalization: A Common Logic to Seeing Cats and Cosmos原评论 ↗
引用T-A · 2015-07

“It's cute, but it does not turn the subject into physics.”

承认这个映射有趣,但认为它并没有把该领域变成物理学
Strike for physics funding原评论 ↗
引用adityab · 2016-09

“Additionally, a paper that has everyone excited about deep connections between the mathematical analysis of physical systems and the hierarchical feature learning paradigm speaks of the connection in terms of the Renormalization Group [2].”

拿它说明物理系统分析与层次化特征学习的联系,以重整化群来表述
Link Between Deep Neural Networks and the Nature of the Universe原评论 ↗
引用ssivark · 2016-11

“In particular, I've spent the last several months in trying to understand the following two papers in increasing levels of depth”

提到自己花了数月深入研究这篇物理与深度学习的论文
Ask HN: What projects are you currently working on?原评论 ↗
看全部 9 条 →
引用epberry · 2016-12

“there are also many researchers poking and prodding at deep learning theory [1, 2].”

把它列为研究者探究深度学习理论的例子之一
OpenAI Universe原评论 ↗
引用mlevental · 2019-07

“it's quite plausible that at the fringe/edge there is research being done that involves differential geometry - hell there's this paper”

举它为例,说明前沿研究可能涉及微分几何,指出变分重整化群与深度学习的映射
Discrete Differential Geometry: An Applied Introduction [pdf]原评论 ↗
引用riemannzeta · 2024-06

“The pyramid strategy loosely tracks with renormalization group theory, which has been formally studied for years as a method of interpreting machine learning models:”

引用它说明重整化群理论已被正式研究多年,用来解释机器学习模型
How Does GPT-4o Encode Images?原评论 ↗
被提交为 story · 4 次