Improving subspace clustering by combining self-expressive and greedy models

Subspace clustering is a vital task in machine learning that aims to identify clusters within different subspaces. Inspired by two famous models, the self-expressive subspace clustering model (SEM) and greedy subspace clustering model (GSM), in this study, we propose to combine them for possible improvement. Specifically, the proposed method alternates between learning SEM and GSM to improve the connectivity between data points. Depending on the adaption order, two alternatives are provided to achieve the final affinity matrix: (a) SEM→GSM→SEM or (b) GSM→SEM→GSM. Both alternatives are evaluated in the experiment and show some possible improvements.