P_Y_L_U的博客

两个常用不等式的用法

总结了一些常用矩阵的迹的性质。

截断正态分布的期望的推导过程

Gram-Schmidt orthogonalization 格拉姆-施密特正交化Gram-Schmidt orthogonalization 格拉姆-施密特正交化Gram-Schmidt orthogonalization 格拉姆-施密特正交化在空间Sk\mathcal{S}^kSk中，向量组 α1\alpha_1α1​, α2\alpha_2α2​, …,αk\alpha_kαk​ 线性无关，寻找一组正交向量基 β1\beta_1β1​, β2\beta_2β2​,…, βk\beta_kβk​,

Glivenko-Cantelli class & Donsker classGlivenko-Cantelli classDonsker classX1,...,XnX_1,...,X_nX1​,...,Xn​是来自于可测空间 (X,A)(\mathcal{X},\mathcal{A})(X,A) 上概率分布 PPP 的随机样本。经验分布函数是观测值的离散均匀测度，记δx\delta_xδx​ 是 xxx 处的概率分布，则令 Pn=n−1∑i=1nδX\mathbb{P}_n = n^{-1}\

Slutsky's TheoremLet Xn,Yn{\displaystyle X_{n},Y_{n}}Xn​,Yn​ be sequences of scalar/vector/matrix random elements. If Xn{\displaystyle X_{n}}Xn​ converges in distribution to a random element X{\displaystyle X}X and Yn{\displaystyle Y_{n}}Yn​ converges i

矩阵对标量求导： U=U(x,y) U=U(x,y) U = U(x,y) U is the matrix, x,y is the scalar.\ partial derivat\begin{equation} \frac{\partial |U|}{\partial x}=|U|tr(U^{-1}{\frac{\partial U}{\partial x..

Kullback-Leibler(KLKL\mathrm {KL}) loss （离散）For discrete probability distributions F(x)F(x)F(x) and G(x)G(x)G(x), the Kullback-Leibler (KLKL\mathrm {KL}) loss from F(x)F(x)F(x) to G(x)G(x)G(x) is def...

Stirling’s formula n!∼2πn−−−√(n/e)nn!∼2πn(n/e)n n! \sim \sqrt{2\pi n}(n/e)^n Sums of powers of integers Let S(k)n=1k+2k+...+nkSn(k)=1k+2k+...+nk S_n^{(k)} = 1^k+2^k + ...+n^k then S(k)n∼nk+1k+1...