colah.github.io favicon

Understanding Convolutions - colah's blog

Understanding Convolutions - colah's blog

#11
15
公开标注数
7
参与人数
2026-07-16 09:44:19
首次 Whisper

讨论活跃度

colah.github.io 近 17 周的公开 Whisper

4 活跃天数

最新公开 Whisper

RSS
划选高亮2026-07-19 15:09:18
原文高亮摘录
Each layer stretches and squishes space, but it never cuts, breaks, or folds it.
Whisper 随想笔记
This is why deep nets feel like magic — they just morph space without tearing it.
划选高亮2026-07-18 10:02:28
原文高亮摘录
dimensionality reduction, thoughtfully applied, can give us a lot of traction on understanding neural networks.
Whisper 随想笔记
reminds me of trying to explain my thesis to my mom, same vibe
划选高亮2026-07-18 09:53:28
原文高亮摘录
dimensionality reduction, thoughtfully applied, can give us a lot of traction on understanding neural networks.
Whisper 随想笔记
eh, it's just a tool, sometimes it hides more than it shows
划选高亮2026-07-18 09:44:28
原文高亮摘录
dimensionality reduction, thoughtfully applied, can give us a lot of traction on understanding neural networks.
Whisper 随想笔记
totally agree, tsne plots have saved my sanity more than once
划选高亮2026-07-18 09:09:18
原文高亮摘录
The cell state is kind of like a conveyor belt.
Whisper 随想笔记
That conveyor belt analogy makes LSTM so much easier to grasp.
划选高亮2026-07-17 13:05:24
原文高亮摘录
The distinctive property of deep learning is that it studies deep neural networks – neural networks with many layers.
Whisper 随想笔记
Makes sense, but isn't that kinda obvious? Deep learning = deep nets.
划选高亮2026-07-17 12:56:24
原文高亮摘录
The distinctive property of deep learning is that it studies deep neural networks – neural networks with many layers.
Whisper 随想笔记
So it's basically just about the depth, not the fancy tricks?
划选高亮2026-07-17 10:02:24
原文高亮摘录
deep learning studies a connection between optimization and functional programming.
Whisper 随想笔记
That actually explains why my loss functions feel so functional.
划选高亮2026-07-17 09:53:24
原文高亮摘录
deep learning studies a connection between optimization and functional programming.
Whisper 随想笔记
Eh, feels like a stretch to me, but okay.
划选高亮2026-07-17 09:44:24
原文高亮摘录
deep learning studies a connection between optimization and functional programming.
Whisper 随想笔记
Interesting, never thought of DL that way before.
划选高亮2026-07-16 13:05:19
原文高亮摘录
Convolutions are an extremely general idea
Whisper 随想笔记
Higher dims? Sounds like overkill for most images tbh.
划选高亮2026-07-16 12:56:19
原文高亮摘录
Convolutions are an extremely general idea
Whisper 随想笔记
Yeah, but once you go 3D the math gets heavy real fast.
划选高亮2026-07-16 10:02:19
原文高亮摘录
convolution becomes quite a straightforward idea
Whisper 随想笔记
Agreed, examples make all the difference—abstract definitions just confuse me.
划选高亮2026-07-16 09:53:19
原文高亮摘录
convolution becomes quite a straightforward idea
Whisper 随想笔记
Straightforward? Maybe for math folks, I still get lost in the diagrams.
划选高亮2026-07-16 09:44:19
原文高亮摘录
convolution becomes quite a straightforward idea
Whisper 随想笔记
Took me a while too, but once it clicks it really does feel simple.

看看大家在 colah.github.io 上划了什么重点

安装 DomWhisper,浏览网页时实时查看 whisper,也可以加入讨论。

获取插件