Stochastic gradient descent - Wikipedia
5
公开标注数
5
参与人数
2026-07-22 09:39:54
首次 Whisper
本页的公开 Whisper
划选高亮2026-07-22 13:00:54
原文高亮摘录
“Stochastic gradient descent is a popular algorithm for training a wide range of models in machine learning”
Whisper 随想笔记
True, but for geophysics FWI it struggles with local minima sometimes.
划选高亮2026-07-22 12:51:54
原文高亮摘录
“Stochastic gradient descent is a popular algorithm for training a wide range of models in machine learning”
Whisper 随想笔记
Yeah it's basically the workhorse behind every neural net these days.
划选高亮2026-07-22 09:57:54
原文高亮摘录
“A compromise between computing the true gradient and the gradient at a single sample is to compute the gradient against more than one training sample”
Whisper 随想笔记
Yeah, my code always runs faster with mini-batches, but tuning the size is a pain.
划选高亮2026-07-22 09:48:54
原文高亮摘录
“A compromise between computing the true gradient and the gradient at a single sample is to compute the gradient against more than one training sample”
Whisper 随想笔记
So it's just a middle ground, but smoother convergence sounds nice.
划选高亮2026-07-22 09:39:54
原文高亮摘录
“A compromise between computing the true gradient and the gradient at a single sample is to compute the gradient against more than one training sample”
Whisper 随想笔记
Mini-batches are the sweet spot, vectorization makes it fly.
分享本页 Whisper
短链接
https://domwhisper.com/s/90d15970b3f7嵌入代码
<iframe src="https://domwhisper.com/embed/90d15970b3f7" width="100%" height="480" style="border:0;border-radius:16px" loading="lazy"></iframe>