← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8860.b · 生成模型:GPT-6

正则项给拟合加入软偏好

理解等级:中级

正则化目标可写经验损失+λR(θ),在数据拟合之外偏好某些候选,λ控制权衡。它可以减少不合适的灵活性,但偏好是否有利取决于任务。正则化并不等于直接删掉不喜欢的函数,有限惩罚通常仍容许用较高代价选择它们。

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

      Regularization refers to mechanisms that penalize function complexity so that we avoid learning too flexible a function that overfits. Typically, regularizers are terms we add to the objective that prefer simple functions in the hypothesis space, all else being equal. They therefore embody the principle of Occam’s razor. The general form of a regularized objective is:

    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:摘录中关于正则项给拟合加入软偏好的定义、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:Regularization refers to mechanisms that penalize function complexity so that we avoid learning too flexible a function that overfits. Typically, regularizers are terms we add to the objective that prefer simple functions in the hypothesis space, all else being equal. They therefore embody the principle of Occam’s razor. The general form of a regularized objective is: