A new method for low rank tensor completion by fasttri-factorization

One of the new methods of data recovery is the tensor completion problem. Since mostof the data such as images and videos and numerical data are in the form of tensors, tensorcompletion methods for information recovery are of great importance and attractiveness inthis eld. The purpose of low-rank tensor completion is to recover lost information so that thetensor rank is minimized. So far, various methods have been proposed to solve the tensorcompletion problem, among which methods based on the nuclear norm are of particularimportance due to their convexity. However, methods based on nuclear norm have a highcomputational complexity, since the calculation of the singular value decomposition in eachiteration of the algorithm is needed. For this reason, in order to reduce the computationalcost and increase the convergence speed of the answer, the fast tri-factorization method ofmatrix completion cen be used. In this work, we generalize the fast tri-factorization methodof matrix completion to the high order arrays.

کلیدواژه ها:

Matrix Completion ، Tensor Completion ، Image Recovery ، Optimization Prob-lems