← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8954.b · 生成模型:GPT-6

差分幂次决定补洞偏好

理解等级:中级

在[1,1,1,1,1,a,0,0,0,0,0]的单点补洞中,优化仅剩|1−a|ʳ+|a|ʳ。r>1时最优a=0.5,r=1时[0,1]均最优,0时最优为0或1。这个例子说明不同导数先验偏好平滑过渡或稀疏突变,不保证复杂图像都能唯一恢复。

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

      For r = 2 the best value of a is a = 0.5, and this solution is the best for any value of r > 1. For r = 1, any value a ∈ [0, 1] is equally good, and p(ℓ) decreases for values of a outside that range. And for r < 1, the best solution is for a = 0 or a =

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第27章《统计图像模型》(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/c027b.xhtml) - 结论依据:摘录中关于差分幂次决定补洞偏好的定义、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:For r = 2 the best value of a is a = 0.5, and this solution is the best for any value of r > 1. For r = 1, any value a ∈ [0, 1] is equally good, and p(ℓ) decreases for values of a outside that range. And for r < 1, the best solution is for a = 0 or a =