Zero-shot learning - Wikipedia
5
公开标注数
5
参与人数
2026-07-23 10:02:28
首次 Whisper
本页的公开 Whisper
划选高亮2026-07-23 13:23:28
原文高亮摘录
“Zero-shot methods generally work by associating observed and non-observed classes through auxiliary information that encodes observable distinguishing properties of objects”
Whisper 随想笔记
Sounds like how I learn new words from context clues, makes sense even if the model never saw a zebra.
划选高亮2026-07-23 13:14:28
原文高亮摘录
“Zero-shot methods generally work by associating observed and non-observed classes through auxiliary information that encodes observable distinguishing properties of objects”
Whisper 随想笔记
So it's basically using side info to bridge what you know and what you don't, clever but fragile if the descriptions are
划选高亮2026-07-23 10:20:28
原文高亮摘录
“Zero-shot learning (ZSL) is a problem setup in machine learning where, at test time, a learner observes samples from classes which were not observed during…”
Whisper 随想笔记
We used this for rare animal species, it actually works better than I thought.
划选高亮2026-07-23 10:11:28
原文高亮摘录
“Zero-shot learning (ZSL) is a problem setup in machine learning where, at test time, a learner observes samples from classes which were not observed during…”
Whisper 随想笔记
Sounds cool but how does it even know what to look for then?
划选高亮2026-07-23 10:02:28
原文高亮摘录
“Zero-shot learning (ZSL) is a problem setup in machine learning where, at test time, a learner observes samples from classes which were not observed during…”
Whisper 随想笔记
So basically the model guesses classes it's never seen? Wild.
分享本页 Whisper
短链接
https://domwhisper.com/s/3d56608ee949嵌入代码
<iframe src="https://domwhisper.com/embed/3d56608ee949" width="100%" height="480" style="border:0;border-radius:16px" loading="lazy"></iframe>