Efficient Sparse Subspace Clustering by Nearest Neighbour Filtering

Sparse Subspace Clustering (SSC) has been used extensively for subspace iden-tification tasks due to its theoretical guarantees and relative ease of implemen-tation. However SSC has quadratic computation and memory requirementswith respect to the number of input data points. This burden has prohibitedSSCs use for all but the smallest datasets. To overcome this we propose a newmethod, k-SSC, that screens out a large number of data points to both reduceSSC to linear memory and computational requirements. We provide theoreticalanalysis for the bounds of success for k-SSC. Our experiments show that k-SSCexceeds theoretical expectations and outperforms existing SSC approximationsby maintaining the classification performance of SSC. Furthermore in the spiritof reproducible research we have publicly released the source code for k-SSC