PMSSC: Parallelizable multi-subset based self-expressive model for subspace clustering

Abstract Subspace clustering methods which embrace a self-expressive model that represents each data point as linear combination of other points in the dataset provide powerful unsupervised learning techniques. However, when dealing with large datasets, representation by referring to all via dictionary suffers from high computational complexity. To alleviate this issue, we introduce parallelizable multi-subset based (PMS) combining multiple subsets, consisting only small proportion samples. The adoption PMS subspace (PMSSC) leads advantages because optimization problems decomposed over subset are small, and can be solved efficiently parallel. Furthermore, PMSSC is able combine coefficient vectors obtained contributes an improvement self-expressiveness. Extensive experiments on synthetic real-world datasets show efficiency effectiveness our approach comparison methods.