← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8995 · 生成模型:GPT-6

多解预测需要区分决策与分布

理解等级:中级

灰度输入可能对应多种合理颜色,单一点预测无法表示全部可能性。若任务需要不确定性与多样输出,应学习条件分布;若只需按明确损失采取一个决策,点估计仍可合理。问题不是单点永远无效,而是它表达不了完整多峰分布。

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

      What if there are multiple equally valid answers? Taking a color image and making it grayscale is a many-to-one mapping: for any given grayscale value, there are many different color values that could have projected to that value of gray. This means that the colorization problem is fundamentally ambiguous, and there may be multiple valid color solutions for any grayscale input. A point estimate is

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于多解预测需要区分决策与分布的定义、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:What if there are multiple equally valid answers? Taking a color image and making it grayscale is a many-to-one mapping: for any given grayscale value, there are many different color values that could have projected to that value of gray. This means that the colorization problem is fundamentally ambiguous, and there may be multiple valid color solutions for any grayscale input. A point estimate is