Online graph learning for time‐varying graphs

Abstract

Abstract In this paper, the is focus on the online graph learning problems in time‐varying environment. Traditional graph learning methods always assume that the underlying graph is static and that enough training data are available. However, in many real world applications, the underlying graph always changes slowly and only a small batch of data can be accessed at each time step. Under the assumption that the variation of the adjacency matrix is smooth, an online graph learning method is proposed, which improves the performance by leveraging the prior from the previous estimations. Experiments show that the proposed method performs best in both dynamic and static settings.