caesarcai / LRPCA

Learned Robust PCA: A Scalable Deep Unfolding Approach for High-Dimensional Outlier Detection

LRPCA

This repository is for our paper:

[1] HanQin Cai, Jialin Liu, and Wotao Yin. Learned Robust PCA: A Scalable Deep Unfolding Approach for High-Dimensional Outlier Detection. In Advances in Neural Information Processing Systems, 34: 16977-16989, 2021.

To display math symbols properly, one may have to install a MathJax plugin. For example, MathJax Plugin for Github.

Problem description

Given $Y = X + S$ where $X$ is the underlying low-rank matrix and $S$ is the sparse ourlier matrix, robust principal component analysis (RPCA) aims to recover $X$ and/or $S$ from the observed $Y$, depending on the application.

Files description

• synthetic_data_exp involves our codes for the synthetic-data experiments¹.

First Time to Run

• Enter synthetic_data_exp and run testing_codes_matlab.m directly.
• The test script will call a trained model stored in synthetic_data_exp/trained_models.

Training the Model

• Enter synthetic_data_exp and run training_codes.py directly.
• The training script will write the model into a .mat file that the test script can load.

Dependencies

• Testing codes: MATLAB (>= 2017b)
• Training codes: CUDA 11.0; pytorch 1.7.1

Footnotes

1. Other parts will be released soon. ↩