← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8855.b · 生成模型:GPT-6

动量也可能越过低点并振荡

理解等级:中级

惯性可加快沿稳定方向的前进,但过大动量也会在到达低损失后继续运动、越过低点并振荡。原书绝对值例展示的是特定步长与参数下的轨迹,不是某个μ对所有目标都最优。首次接近最优和最终稳定收敛是不同评价。

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

      shows how momentum affects gradient descent for a simple objective J = abs(θ) (absolute value of θ). As can be seen in the figure, some momentum can help convergence rate (figure 10.2, μ = 0.5) but too much momentum will cause the trajectory to overshoot the optimum and even when the optimum loss is achieved, the trajectory might not stop (figure 10.2, μ = 0.95).

    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:摘录中关于动量也可能越过低点并振荡的定义、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:shows how momentum affects gradient descent for a simple objective J = abs(θ) (absolute value of θ). As can be seen in the figure, some momentum can help convergence rate (figure 10.2, μ = 0.5) but too much momentum will cause the trajectory to overshoot the optimum and even when the optimum loss is achieved, the trajectory might not stop (figure 10.2, μ = 0.95).