xiangyongcao / PMoEP

This is the implementation of "Low rank matrix factorization under general mixture noise distributions."

README

Version 1.0, 29-Dec-2015

This package contains the MATLAB implementation of "Low-Rank Matrix Factorization Under General Mixture Noise Distributions".

The code has been tested with MATLAB 2014b on a PC with 64-bit windows 7.

================================================================================================

Use of this code is free for research purposes only.

================================================================================================

Reference:

Xiangyong Cao, Yang Chen, Qian Zhao, Deyu Meng, Yao Wang, Dong Wang and Zongben Xu, Low-Rank Matrix Factorization Under General Mixture Noise Distributions, 15th International Conference on Computer Vision (ICCV), Chile, Dec. 2015 (Oral)

================================================================================================

Installation:

1. Unpack the contents of the compressed file to a new directory.

2. Run the Demos

================================================================================================

Demos:

Demo_EP.m % EP 0.2 noise Demo_Gauss.m % Gaussian noise Demo_Laplace.m % Laplace noise Demo_Sparse.m % Sparse noise Demo_Mixture1.m % Mixture1 noise: Sparse noise + Gaussian noise + Gaussian noise Demo_Mixture2.m % Mixture2 noise: EP 0.5 noise + Gaussian noise + Laplace noise

================================================================================================

Main Routine

[label,model,TW,OutU,OutV,llh,llh_BIC,p] = EM_PMoEP(InW,InX,r,param,p,lambda) %Input: InW: d x n x param.k indicator matrices InX: d x n input data matrix r: the rank param: --param.maxiter: maximal iteration number --param.OriX: ground truth matrix --param.InU: initialized factorized matrice U --param.InV: initialized factorized matrice V --param.k: the number of mixture components --param.display: display the iterative process --param.tol: the tolerance for stop p: the candidate components lambda: the tuning parameter

%Output: label: the labels of the noises model: model.eta, the precisions of the different EPs model.Pi,the mixing coefficients W: d x n weighted matrix OutU: the final factorized matrix U OutV: the final factorized matrix V llh: the log likelihood llh_BIC: the log likelihood used in BIC criterion p: the selected components

• =========================================================================
  If you have any quesion, please contact Xiangyong Cao([e-mail])