Robust PCA 学习笔记

很久没有写学习笔记了，年初先后忙考试，忙课程，改作业，回家刚安定下来，读了大神上学期给的paper，这几天折腾数学的感觉很好，就在这里和大家一起分享一下，希望能够有所收获。响应了Jeffrey的建议，强制自己把笔记做成英文的，可能给大家带来阅读上的不便，希望大家理解，多读英文的东西总没坏处的。这里感谢大神和我一起在本文手稿部分推了一些大牛的“ easily achieved”stuff... 本文尚不成熟，我也是初接触Robust PCA,希望各位能够拍砖提出宝贵意见。

Robust PCA

Rachel Zhang

 

1. RPCA Brief Introduction

1.     Why use Robust PCA?

Solve the problem withspike noise with high magnitude instead of Gaussian distributed noise.

2.     Main Problem

Given C = A*+B*, where A*is a sparse spike noise matrix and B* is a Low-rank matrix, aiming at recoveringB*.

B*= UΣV’, in which U∈R^n*k ,Σ∈R^k*k ,V∈R^n*k

3.     Difference from PCA

Both PCA and Robust PCAaims at Matrix decomposition, However,

In PCA, M = L0+N0,    L0:low rank matrix ; N0: small idd Gaussian noise matrix,it seeks the best rank-k estimation of L0 by minimizing ||M-L||₂ subjectto rank(L)<=k. This problem can be solved by SVD.

In RPCA, M = L0+S0, L0:low rank matrix ; S0: a sparse spikes noise matrix, we’lltry to give the solution in the following sections.

 

2. Conditionsfor correct decomposition

4.     Ill-posed problem:

Suppose sparse matrix A* and B*=e_ie_j^Tare the solution of this decomposition problem.

1)       With the assumption that B* is not only low rank but alsosparse, another valid sparse-plus low-rank decomposition might be A₁= A*+ e_ie_j^T andB₁ = 0, Therefore, we need an appropriatenotion of low-rank that ensures that B* is not too sparse. Conditionswill be imposed later that require the space spanned by singular vectors U andV (i.e., the row and column spaces of B*) to be “incoheren