知识卡片

普通读书笔记卡 · 7005.j · 生成模型:DeepSeek-V4.1-Flash

标签噪声、错误目标与恶意反馈

理解等级:中级

监督学习里标签是模型的「正确答案」,标签出问题会成为大量下游问题的根源。统计里 noise 即 errors:标签有误,错误会传导进模型;局部系统性偏差更致命,如标注员总把青蛙误标成蟾蜍。另一类是「错误标签目标」:模型极擅长预测所给标签,甚至揭穿「望其所指」与「实其所指」的落差——把「购买」当用户满意,会让它专推看着划算实则劣质或标题党的商品。依赖用户反馈的系统还会被刷评投毒,故须监控标签质量并做安全防护。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • Basic Introduction to Models · 查看原书依据

      Label noise In statistical language, the term noise is a synonym for errors. If our provided labels are incorrect for some reason, these errors can propagate into the model behavior. Random noise can be tolerable in some settings if the errors balance out over time, although it is still important to measure and assess.

    参考来源

    - 位置:《Reliable Machine Learning》第03章《Basic Introduction to Models》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch03.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:Label noise In statistical language, the term noise is a synonym for errors. If our provided labels are incorrect for some reason, these errors can propagate into the model behavior. Random noise can be tolerable in some settings if the errors balance out over time, although it is still important to measure and assess.