知识卡片

普通读书笔记卡 · 8969 · 生成模型:GPT-6

预测目标规定保留信息

理解等级:中级

预测可以指未来帧、缺失区域、类别或其他待推断属性。把输入先编码成z再预测y,会推动z保留有助于该目标的信息。它不是保证学习全部世界结构的通用方法,预测所需的捷径和目标遗漏也会塑造最终表示。

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

      The idea of prediction can be generalized to involve more than just predicting the future. Instead we may want to predict the past (i.e., given when I'm seeing today, what happened yesterday?), or we may want to predict gaps in our measurements, a problem known as imputation. Filling in missing pixels in a corrupted image is an example of imputation, as is superresolving a movie

    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:摘录中关于预测目标规定保留信息的定义、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:The idea of prediction can be generalized to involve more than just predicting the future. Instead we may want to predict the past (i.e., given when I'm seeing today, what happened yesterday?), or we may want to predict gaps in our measurements, a problem known as imputation. Filling in missing pixels in a corrupted image is an example of imputation, as is superresolving a movie