Unverified50% confidenceFactExact time
Early research found that even when using examples with randomly incorrect labels, few-shot performance was sometimes not much worse than with correct labels.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified先前的难负样本挖掘工作大多忽略非干扰项，只关注与查询的相似度，因而更易误选假负样本73% similarUnverified预标注会引入锚定效应，标注员倾向于接受而非推翻已有预标注结果，即使存在微小错误，低质量预标注会系统性地将判断拉向错误方向70% similarUnverifiedThe missingness patterns of missing data are often not random but systematic, and ignoring them can lead to serious inferential bias.68% similarUnverified在关系抽取中，标签出现的频率几乎不能预测模型在该标签上的表现，真正决定泛化能力的是模型是否见过相似的证据表达方式67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924783API
curl https://kongchang.com/api/v1/knowledge/claims/924783MCP
get_claim(id=924783)