← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9101.b · 生成模型:GPT-6

乘积玩具揭示反问题歧义

理解等级:中级

只观察y=ab,一条观测通常对应多个(a,b),例如y=1时正值解沿双曲线分布。这个模型帮助理解反射率与照明相乘时的不可辨识性,以及先验和决策损失的作用;增加优化步数不能消除数据本身缺失的信息。

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

      Here's an example. The color constancy problem is to estimate surface reflectance colors when we only get to observe the wavelength-by-wavelength product of the each surface reflectance spectrum and the unknown illuminant spectrum. A toy model for that problem is to try to estimate the scalars a and b from only observing their product, y = a × b. There's a surprising richness even to

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第52章《如何做研究》(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/c052b.xhtml) - 结论依据:摘录中关于乘积玩具揭示反问题歧义的定义、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:Here's an example. The color constancy problem is to estimate surface reflectance colors when we only get to observe the wavelength-by-wavelength product of the each surface reflectance spectrum and the unknown illuminant spectrum. A toy model for that problem is to try to estimate the scalars a and b from only observing their product, y = a × b. There's a surprising richness even to