按年被提到
20172017 · 11 次2026
机器学习与深度学习›神经科学启发的 AI·arXiv:1710.09829 · cs.CV

Dynamic Routing Between Capsules

胶囊之间的动态路由
Sara Sabour, Nicholas Frosst, Geoffrey E Hinton · 2017

Hinton 的胶囊网络:以一致性路由取代池化

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 16 人首次 2017-10-27 · 最近 2023-02-08
16不同的人
10活跃月份
18讨论串
4,080被引(OpenAlex)

胶囊是一组神经元,其活动向量的长度表示某类实体存在的概率,方向表示实例化参数;低层胶囊通过变换矩阵为高层胶囊的实例化参数做预测,多个预测一致时高层胶囊才被激活。作者用迭代的「一致性路由」机制训练多层胶囊系统,在 MNIST 上达到当时最优,在识别高度重叠的数字时明显优于卷积网络。

HN 为什么提它

2017 年评论者把它视为 Hinton 期待已久的新架构,同时讨论它在 CIFAR-10 等数据集上较弱、只在玩具数据集上验证,之后多被当作胶囊网络的出处引用。

方法难度 4 / 5约 45 分钟经得起时间
读之前最好知道
1卷积神经网络(CNN)
2反向传播
3向量与点积相似度
4视角不变性与池化

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leecarraher · 2017-12

“the long awaited paper from Geoffrey Hinton (although Sara Sabour is principal author, Hinton has been hinting at this for a while now) on an alternative to back-propagation”

推荐 Hinton 期待已久的反向传播替代方案论文
Ask HN: What was the best CS paper you read in 2017?原评论 ↗
批评Smerity · 2017-11

“Capsule networks[1] have literally not been used on anything larger than what is considered a toy dataset for computer vision yet.”

批评胶囊网络还没用在比玩具数据集更大的任务上
A Year in Computer Vision原评论 ↗
引用nl · 2017-10

“(where they finally got Hinton's capsules to work)”

称 Hinton 的胶囊终于被做出来了
RCN is much more data efficient than traditional Deep Neural Networks原评论 ↗
引用SophosQ · 2017-11

“The paper however details their experiments on Cifar-10 and other datasets in the discussion section.”

指出论文在讨论部分还给出了 Cifar-10 等数据集的实验
A New Twist on Neural Networks原评论 ↗
引用skierscott · 2017-11

“Geff Hinton created an entire new architecture to get 0.25% error, which is far better than the baseline 0.39% error [1].”

说明 Hinton 的新架构错误率 0.25%,优于 0.39% 的基线
Deep Learning Is Eating Software原评论 ↗
引用arketyp · 2017-11

“What, for instance, is your opinion about Hinton's recently published capsules idea?”

询问对 Hinton 新发表的胶囊想法有何看法
Researchers create “neuromorphic chip” that is modeled after the brain原评论 ↗
看全部 13 条 →
引用babak_ap · 2018-01

“Is the mechanism of 'Arc' gene related to the dynamic routing process of Capsule networks [1, 2]?”

询问 Arc 基因机制是否与胶囊网络的动态路由有关
Brain Cells Share Information with Virus-Like Capsules原评论 ↗
引用xvedejas · 2018-02

“Link to the capsule network paper, for those who haven't heard of it:”

为没听说过的人贴出胶囊网络论文链接
Keras reimplementation of "One pixel attack for fooling deep neural networks"原评论 ↗
引用dr_zoidberg · 2018-06

“the idea of Capsule Networks (Hinton et al)[1, 2] seems to tackle some of these issues”

认为胶囊网络似乎能解决其中一些问题
Why do CNNs generalize so poorly to small image transformations?原评论 ↗
引用cs702 · 2019-06

“The main example the authors use to illustrate these issues is capsule networks, first proposed two years ago.[a]”

说明文中主要例子是两年前提出的胶囊网络
Machine learning systems are stuck in a rut原评论 ↗
引用cs702 · 2019-07

“As an example, they use capsule networks[c]”

以胶囊网络为例
A Differentiable Programming System to Bridge ML and Scientific Computing原评论 ↗
引用cs702 · 2020-05

“In certain new capsule-routing algorithms -- e.g., EM routing[a], Heinsen routing[b], dynamic routing[c]”

把动态路由列为胶囊路由算法之一,与 EM routing 并列
Explainable Deep Learning: A Field Guide for the Uninitiated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.”

把它列入自己在推特发布的 AI 论文清单
What were the 40 research papers on the list Ilya Sutskever gave John Carmack?原评论 ↗
被提交为 story · 5 次