知识卡片

普通读书笔记卡 · 6677.c · 生成模型:GPT-6

截断分解近似保留主要矩阵结构

理解等级:中级

保留较大的奇异值及其方向,可得到低秩近似,减少表示和计算规模。若被舍弃方向本来为零,仍可精确重构;否则有信息损失。非中心化SVD关注整体矩阵近似,不宜直接等同PCA最大化中心化方差。

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

      At this point, you might ask, “But wait, can’t I do the full SVD and just choose the dimensions with the most variance and keep those?” And you’d be completely right—you can indeed—that’s the SVD algorithm in a nutshell! However, there are other benefits to using truncated SVD. In particular, there are several algorithms that allow computing truncated SVD decomposition of a large matrix efficiently,

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第04章《从词数到语义分析》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-4.html) - 结论依据:摘录中关于截断分解近似保留主要矩阵结构的定义、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:At this point, you might ask, “But wait, can’t I do the full SVD and just choose the dimensions with the most variance and keep those?” And you’d be completely right—you can indeed—that’s the SVD algorithm in a nutshell! However, there are other benefits to using truncated SVD. In particular, there are several algorithms that allow computing truncated SVD decomposition of a large matrix efficiently,