最近刚开始学《机器学习与信息内容安全》,用的是吴恩达的机器学习讲义。在cs229-notes1的2.1 Matrix derivatives一小节中,出现了许多矩阵运算公式,这里我试图给出其中几个公式的推导证明。
迹交换律公式的证明
公式: tr A B = tr B A \operatorname{tr} A B =\operatorname{tr} B A trAB=trBA
证明:
令 A = [ a 11 … a 1 m ⋮ ⋱ ⋮ a n 1 … a n m ] A=\left[\begin{array}{ccc}a_{11} & \dots & a_{1 m} \\ \vdots & \ddots & \vdots \\ a_{n 1} & \dots & a_{n m}\end{array}\right] A=⎣⎢⎡a11⋮an1…⋱…a1m⋮anm⎦⎥⎤, B = [ b 11 … b 1 n ⋮ ⋱ ⋮ b m 1 … b m n ] B=\left[\begin{array}{ccc}b_{11} & \dots & b_{1 n} \\ \vdots & \ddots & \vdots \\ b_{m 1} & \dots & b_{m n}\end{array}\right] B=⎣⎢⎡b11⋮bm1…⋱…b1n⋮bmn⎦⎥⎤
其中 A为 n × m n \times m n×m矩阵(即 A ∈ R n × m A \in \mathbb{R}^{n \times m} A∈Rn×m), B 为 m × n m \times n m×n矩阵(即 B ∈ R m × n B \in \mathbb{R}^{m \times n} B∈Rm×n)
则有
A B = [ ∑ i = 1 m a 1 i b i 1 ⋯ ∑ i = 1 m a 1 i b i n ⋮ ⋱ ⋮ ∑ i = 1 m a n i b i 1 ⋯ ∑ i = 1 m a n i b i n ] A B=\left[\begin{array}{ccc}\sum_{i=1}^{m} a_{1 i} b_{i 1} & \cdots & \sum_{i=1}^{m} a_{1 i} b_{i n} \\ \vdots & \ddots & \vdots \\ \sum_{i=1}^{m} a_{n i} b_{i 1} & \cdots & \sum_{i=1}^{m} a_{n i} b_{i n}\end{array}\right] AB=⎣⎢⎡∑i=1ma1ibi1⋮∑i=1manibi1⋯⋱⋯∑i=1ma1ibin⋮∑i=1manibin⎦⎥⎤,
B A = [ ∑ i = 1 n b 1 i a i 1 ⋯ ∑ i = 1 n b i i a i m ⋮ ⋱ ⋮ ∑ i = 1 n b m i a i 1 ⋯ ∑ i = 1 n b m i a i m ] B A=\left[\begin{array}{ccc}\sum_{i=1}^{n} b_{1 i} a_{i 1} & \cdots & \sum_{i=1}^{n} b_{i i} a_{i m} \\ \vdots & \ddots & \vdots \\ \sum_{i=1}^{n} b_{m i} a_{i 1} & \cdots & \sum_{i=1}^{n} b_{m i} a_{i m}\end{array}\right] BA=⎣⎢⎡∑i=1n