← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8861.d · 生成模型:GPT-6

先验过强也会压制真实信号

理解等级:中级

小范数先验在真参数接近原点时可能有帮助,过大的λ却会把解继续拉向零,从而偏离真实关系。先验强度应与数据证据共同评估;原书示例中的最佳中间值不是所有任务都必有正λ最佳,也不保证可从训练拟合单独选出。

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

      In this example, the true solution lies near the origin, so adding some regularization gets us closer to the truth. But using too strong a λ overregularizers; the true solution is not exactly θ = 0. The middle column is the Goldilocks solution, where the strength of the regularizer is just right.

    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:摘录中关于先验过强也会压制真实信号的定义、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:In this example, the true solution lies near the origin, so adding some regularization gets us closer to the truth. But using too strong a λ overregularizers; the true solution is not exactly θ = 0. The middle column is the Goldilocks solution, where the strength of the regularizer is just right.