知识卡片
交替优化两组聚类变量
固定中心时将每个样本分配给最近中心,固定分配时将中心更新为该簇样本均值。每步求解当前参数块的最优值,在明确处理并列和空簇后不会增加平方失真。但整体问题非凸,最终固定点可能依赖初始化,不能保证全局最优。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 表示学习
原书出处 · 1 处
- · 查看原书依据
K-means uses an optimization algorithm called block coordinate descent. This algorithm splits up the optimization parameters into multiple subsets (blocks). Then it alternates between fully minimizing the objective with respect to each block of parameters. In k-means there are two parameter blocks: and . So, we need to derive update rules that find the minimizer of the k-means objective with respect to each block. It
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:摘录中关于交替优化两组聚类变量的定义、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:K-means uses an optimization algorithm called block coordinate descent. This algorithm splits up the optimization parameters into multiple subsets (blocks). Then it alternates between fully minimizing the objective with respect to each block of parameters. In k-means there are two parameter blocks: and . So, we need to derive update rules that find the minimizer of the k-means objective with respect to each block. It