知识卡片
线性自动编码联系主成分
中心化数据、线性编解码、平方损失、合适秩约束并达到全局最优时,重建子空间可与PCA主子空间一致。潜在坐标本身仍可作可逆变换,不必等于特定PCA坐标,特征值并列时子空间也可能不唯一。
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
Closely related to autoencoders are other dimensionality reduction algorithms like principle components ananlysis (PCA). In fact, an L2 autoencoder, for which both f and g are linear functions, learns an M-dimensional embedding that spans the same subspace as a PCA projection to M-dimensions [57].
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:摘录中关于线性自动编码联系主成分的定义、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:Closely related to autoencoders are other dimensionality reduction algorithms like principle components ananlysis (PCA). In fact, an L2 autoencoder, for which both f and g are linear functions, learns an M-dimensional embedding that spans the same subspace as a PCA projection to M-dimensions [57].