基于张量环分解的非精确的低秩填充算法

A Inexact Low Rank Tensor Completion Algorithm Based on Tensor Ring Decomposition

文章提出一种基于张量环分解的低秩填充算法.利用张量核因子决定存储信息的2-模展开来代替控制结构的1-模和3-模展开.虽然每次迭代不是最优下降,但保证了整体下降.从而减少了计算花费,提高了张量填充效率.最后通过实验验证了新算法的可行性.在精度一致的情况下,文章算法较之前算法快了近3倍.

Based on tensor ring(TR)decomposition,a inexact low rank tensor completion algorithm is proposed.By employing the mode-2 unfolding matrix of core tensors to represents mode-1 unfolding matrix and mode-3’s,it ensures the global descending rather than optimal descent every iteration.The computational cost is reduced and efficiency is improved by employing proposed method.Finally,the experiments on real-world datasets were conducted to evaluate the performance of algorithm.The simulation experiment shows that the proposed algorithm is about 3 times faster than traditional algorithm without almost loss of accuracy.