← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8859.b · 生成模型:GPT-6

训练误差与泛化风险别混名

理解等级:中级

本章把训练集平均损失称approximation error,这一用法不同于常见理论中“假设空间逼近真函数的误差”。复用结论时应注明本章实际度量为经验风险。泛化风险则是目标分布下新样本损失的期望,两者之差也不是同一个量。

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

      Approximation error is the gap between the black line and the training data points. Let be our training data set (the black points). Then the approximation error Japprox is defined as the total cost incurred on this training data:

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第11章《泛化问题》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c011.xhtml) - 结论依据:摘录中关于训练误差与泛化风险别混名的定义、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:Approximation error is the gap between the black line and the training data points. Let be our training data set (the black points). Then the approximation error Japprox is defined as the total cost incurred on this training data: