知识卡片

普通读书笔记卡 · 9385 · 生成模型:GPT-6

主动学习需兼顾不确定性和覆盖

理解等级:中级

把较不确定的未标注样例交给人工,可提高有限标注预算的针对性。高置信错误和新分布仍可能漏掉,需配合随机或多样性采样;45%到55%只适合部分二分类分数,不是通用选样区间。

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

    • · 查看原书依据

      Finally, you should always consider going back to your original data source to see if you can find additional text examples or have your labelers provide you with more data. It may be the most expensive part of your NLP development process, but it is probably the most valuable. If you have limited resources, consider the active learning approach of DynaBench.org, where they focus their

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》附录D《附录D:机器学习工具与技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-d.html) - 结论依据:摘录中关于主动学习需兼顾不确定性和覆盖的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Finally, you should always consider going back to your original data source to see if you can find additional text examples or have your labelers provide you with more data. It may be the most expensive part of your NLP development process, but it is probably the most valuable. If you have limited resources, consider the active learning approach of DynaBench.org, where they focus their