Tensor Train Factorization with Spatio-temporal Smoothness for Streaming Low-rank Tensor Completion

Estimating the missing data from an incomplete measurement or observation plays an important role in the area of big data analytic, especially for some streaming data analysis such as video streaming recovery, traffic data analysis and network engineering. In this paper, by making full use of the potential spatio-temporal smoothness and inherent correlation properties in real-world tensor data, we present a low-rank Tensor Train (TT) factorization method for solving the 3-way streaming low-rank tensor completion problems. Extensive numerical experiments on color images, network traffic data and gray scale videos show that our model outperforms many existing state-of-the-art approaches in terms of achieving higher recovery accuracy.