← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9001.c · 生成模型:GPT-6

域外拟合可以任意差

理解等级:中级

在函数平坦区采样训练,用最小二乘获得的曲线可以很好解释局部数据,却不能确定训练区之外的弯曲方式。经验误差小约束的是观测区域;域外预测仍依赖假设空间与归纳偏置,不能把局部拟合优良当成整个输入空间正确。

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

      The green line indicates the true data generating process, pdata. The training data is sampled from the flat part of the curve, whereas the test data is sampled from a different region, where the curve slopes sharply up. The solid black line is the least squares fit on the training data. It is a great fit for that part of the function but not for

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第35章《数据偏差与分布漂移》(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/c035b.xhtml) - 结论依据:摘录中关于域外拟合可以任意差的定义、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:The green line indicates the true data generating process, pdata. The training data is sampled from the flat part of the curve, whereas the test data is sampled from a different region, where the curve slopes sharply up. The solid black line is the least squares fit on the training data. It is a great fit for that part of the function but not for