知识卡片

普通读书笔记卡 · 6558 · 生成模型:GPT-6

重建信息不等于因果信息

理解等级:中级

重建目标鼓励保留有助还原输入的信息,因果表示则需保留目标机制、干预或跨环境关系。两种信息要求可能重叠,却不等价;不能仅凭低重建误差,就认定Z已把因果因素与其他变化解耦。

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

      A good causal representation, however, shouldn’t try to capture as much information as possible. Rather, it should strive to capture only the causal information in the images and ignore everything else. Indeed, the task of “disentangling” the causal and non-causal factors in Z is generally impossible when Z is unsupervised (meaning we lack labels for Z). However, domain knowledge, interventions, and semi-supervision can help. See

    参考来源

    - 位置:《Causal AI》第05章《连接因果与深度学习》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-5.xhtml) - 结论依据:摘录中关于重建信息不等于因果信息的定义、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:A good causal representation, however, shouldn’t try to capture as much information as possible. Rather, it should strive to capture only the causal information in the images and ignore everything else. Indeed, the task of “disentangling” the causal and non-causal factors in Z is generally impossible when Z is unsupervised (meaning we lack labels for Z). However, domain knowledge, interventions, and semi-supervision can help. See