← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8856.b · 生成模型:GPT-6

替代损失改变可利用的地形

理解等级:中级

原目标为分段常数或缺少有效梯度时,可构造平滑等替代损失,用其梯度提供搜索方向。优化的是替代目标,因此仍需检查原目标是否改善;有非零梯度不保证替代问题与最终业务指标具有相同最优解。

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 gradient of a surrogate loss function, which is a function, Jsurr, with meaningful (non-zero) gradients, that approximates J. An example might be a smoothed version of J. Another way to get v is to compute it by sampling perturbations of θ, and seeing which perturbation leads to lower loss. In this strategy, we evaluate J(θ + ϵ) for a set of perturbations ϵ, then

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第10章《基于梯度的学习算法》(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/c010.xhtml) - 结论依据:摘录中关于替代损失改变可利用的地形的定义、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 gradient of a surrogate loss function, which is a function, Jsurr, with meaningful (non-zero) gradients, that approximates J. An example might be a smoothed version of J. Another way to get v is to compute it by sampling perturbations of θ, and seeing which perturbation leads to lower loss. In this strategy, we evaluate J(θ + ϵ) for a set of perturbations ϵ, then