Convolutional neural network - Wikipedia
8
公开标注数
5
参与人数
2026-07-15 09:39:06
首次 Whisper
本页的公开 Whisper
划选高亮2026-07-17 10:20:00
原文高亮摘录
“A distinguishing feature of CNNs is that many neurons can share the same filter.”
Whisper 随想笔记
Yeah but then how do different filters not end up identical after training?
划选高亮2026-07-17 10:11:00
原文高亮摘录
“A distinguishing feature of CNNs is that many neurons can share the same filter.”
Whisper 随想笔记
Sharing weights is basically the whole trick, right? It's what makes convolution efficient.
划选高亮2026-07-17 10:02:00
原文高亮摘录
“A distinguishing feature of CNNs is that many neurons can share the same filter.”
Whisper 随想笔记
So that's how they keep the model small enough to train on a GPU.
划选高亮2026-07-15 13:00:06
原文高亮摘录
“CNNs use relatively little pre-processing compared to other image classification algorithms”
Whisper 随想笔记
But don't they still need a ton of tuning and data? Not that little pre-processing.
划选高亮2026-07-15 12:51:06
原文高亮摘录
“CNNs use relatively little pre-processing compared to other image classification algorithms”
Whisper 随想笔记
So true, hand-crafting features was such a pain, glad CNNs just learn them.
划选高亮2026-07-15 09:57:06
原文高亮摘录
“CNNs are also known as shift invariant or space invariant artificial neural networks”
Whisper 随想笔记
So it's the pooling layers that break the invariance, right?
划选高亮2026-07-15 09:48:06
原文高亮摘录
“CNNs are also known as shift invariant or space invariant artificial neural networks”
Whisper 随想笔记
That explains why tiny shifts can mess up my image classifier.
划选高亮2026-07-15 09:39:06
原文高亮摘录
“CNNs are also known as shift invariant or space invariant artificial neural networks”
Whisper 随想笔记
Wait, so the name is a lie? They're not actually invariant?
分享本页 Whisper
短链接
https://domwhisper.com/s/9eb3d4b4239a嵌入代码
<iframe src="https://domwhisper.com/embed/9eb3d4b4239a" width="100%" height="480" style="border:0;border-radius:16px" loading="lazy"></iframe>