← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8968 · 生成模型:GPT-6

好表示取决于后续任务

理解等级:初级

表示学习将原始数据映射到便于后续求解的空间。较低维、简单分布和因素分离都可能有帮助,但不是所有任务共同要求。选择表示前应明确要保留哪些差异、忽略哪些变化,以及后续读出如何使用它。

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 representation is one that makes subsequent problem solving easier. Below we discuss several desiderata of a good representation. 30.3.1 Compression A good representation is one that is parsimonious and captures just the essential characteristics of the data necessary for tasks of human interest. There are at least three senses in which compressed representations are desirable. 1. Compressed representations require less memory to store.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第30章《表示学习》(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/c030b.xhtml) - 结论依据:摘录中关于好表示取决于后续任务的定义、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 representation is one that makes subsequent problem solving easier. Below we discuss several desiderata of a good representation. 30.3.1 Compression A good representation is one that is parsimonious and captures just the essential characteristics of the data necessary for tasks of human interest. There are at least three senses in which compressed representations are desirable. 1. Compressed representations require less memory to store.