Greedy Approach for Subspace Clustering from Corrupted and Incomplete Data

TL;DR

This paper introduces the GSSC algorithm, a greedy method that improves subspace clustering performance on incomplete, corrupted, and noisy data, significantly enhancing existing sparse subspace clustering techniques.

Contribution

The paper presents a novel greedy algorithm for subspace clustering that effectively handles incomplete and corrupted data, outperforming current state-of-the-art methods.

Findings

GSSC significantly improves clustering accuracy on corrupted data.

The greedy approach enhances the capability of existing SSC algorithms.

Numerical experiments demonstrate the effectiveness of GSSC.

Abstract

We describe the Greedy Sparse Subspace Clustering (GSSC) algorithm providing an efficient method for clustering data belonging to a few low-dimensional linear or affine subspaces from incomplete corrupted and noisy data. We provide numerical evidences that, even in the simplest implementation, the greedy approach increases the subspace clustering capability of the existing state-of-the art SSC algorithm significantly.