TENSOR FACTORIZATION WITH TOTAL VARIATION FOR INTERNET TRAFFIC DATA IMPUTATION

摘要

Recovering network traffic data from incomplete observed data becomes increasingly critical in network engineering and management. To fully exploit the spatial-temporal features of the Internet traffic data, this paper presents a new tensor completion model which combines the T-product-based tensor factorization with total variation (TV) regularization. To tackle the proposed model, an effective Proximal Alternating Minimization (PAM) algorithm with guaranteed convergence is designed. Extensive experiments on the real-word traffic datasets show that the proposed method has superiority over the existing state-of-the-art methods.