Subspace clustering using temporal information and subspace-dictionary update

Abstract

Subspace Clustering has been widely used to cluster data into some subspaces. Ordered Subspace Clustering (OSC), one of representative methods, reflects temporal characteristics of sequence data. However, OSC suffers from scalability to a large-scale data. For this issue, Stream Sparse Subspace Clustering (StreamSSC) can handle stream data of which entire subspace structure is unclear at each time, and overcome this problem updating adaptively representative sets of subspaces. % We present a proposal of a novel subspace clustering algorithm for sequence data, which aims to reduce the computational complexity of OSC building upon the framework of StreamSSC. The preliminary numerical experiments demonstrate that our proposed method reduces more processing time than OSC does.