← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8972.a · 生成模型:GPT-6

聚类同时学习分配与码本

理解等级:中级

k均值最小化各样本与所分配中心的平方距离,需共同选择离散分配ai和中心zj。可视为离散瓶颈编码器与码本解码器。训练样本分配可存成查表,新样本则需另按最近中心等规则编码,不能只靠原查表完成泛化。

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

      There are two free parameter sets to optimize over: the code vectors and the cluster assignments. The cluster assignments can be represented with a clustering function f : : (30.3) We will represent the code vectors for each cluster with a function g : {1, …, k} → ℝM, where the data dimensionality is M: (30.4) Both f and g can be implemented as lookup

    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:摘录中关于聚类同时学习分配与码本的定义、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:There are two free parameter sets to optimize over: the code vectors and the cluster assignments. The cluster assignments can be represented with a clustering function f : : (30.3) We will represent the code vectors for each cluster with a function g : {1, …, k} → ℝM, where the data dimensionality is M: (30.4) Both f and g can be implemented as lookup