数据科学矩阵基础演示文稿.ppt

* * * 表形象地描述了一组人类基因组的SNP数据（Single Nucleotide Polymorphism data）。一组研究人员在全世界挑选出1064个志愿者，并把他们的SNP数据数字化，也就是把每个位置上可能出现的10种碱基对用数字来代表.  Jun Z. Li et al, Worldwide humanrelationships inferred from genome-wide patterns of variation，Science, 22, Feburary， 2008  * Ref: The Singular Value Decomposition, Applications and Beyond  MATLAB中矩阵分解 (1)特征值分解: AV=VD %[V,D]=eig(A) (2) 奇异值分解: A=USVH % [U,S,V] = SVD(A) Original image (1000×622?) 现实世界的应用I: K=1 K=5 K=20 K=50 K=100 图像压缩 Demo 1 I=double(rgb2gray(imread(monalisa.png))); [U,S,V]=svd(I, econ); s=diag(S); n1=5; Snew=diag([s(1:n1);zeros(size(s,1)-n1,1)]); figure,imshow(U*Snew*V,[]) n2=20; Snew=diag([s(1:n2);zeros(size(s,1)-n2,1)]); figure, imshow(U*Snew*V,[]) A Singular Value Thresholding Algorithm for Matrix Completion rank(A) ||x||0 This number (i.e., the number of linearly independent rows or columns) is simply called the?rank?of A. 参考：信号与数据处理中的低秩模型——理论、算法与应用 Ref: 1. Learning the parts of objects by non-negative matrix factorization, Nature (Citation: 5990) 2. Deconstructing Recommender Systems 3.实例 /entry/209226 [W,H] = nnmf(A,k) Latent Factor Models Learning the parts of objects by non-negative matrix factorization, Nature (Citation: 5990) Start with: we use the following model. ……. Approximation since there are lots of effects we don’t model. * * * * * * /iaiti/article/details/p/blog/3770-xianxingyiweijicunqijiami.html Image formation process = ? Blurry image Sharp image Blur kernel Input to algorithm Desired output Convolutionoperator = ? Blurry image Sharp image Input to algorithm Desired output Blurry image * 卷积 周期边界 循环矩阵 * 梯度域 离散Fourier变换 * 离散Fourier变换 《乐舞图》韩休墓中的一幅壁画《乐舞图》，左侧地毯前的男子可以隐约发现改动的痕迹（上图），经过对高光谱图像的进一步分析发现，左侧地毯前的男子处原先画了一个小孩（下图），后来被修改成了大人。 字典，冗余 Matrix Factorization Jungle 矩阵分解将一般矩阵化为多个简单的因子矩阵之和或之积(因子矩阵揭示了一般矩阵的重要信息)。 * MATLAB中矩阵分解 (1)特征值分解: AV=VD %[V,D]=eig(A) A Tutorial on Principal Component Analysis (PCA，主成分分析) If you have never heard of PCA, don’t worry: for our purposes