1: \begin{abstract}

2: %In this paper, we consider the approximate nonnegative tensor canonical polyadic decomposition problem (NTF),

3: This paper is concerned with improving the empirical convergence speed of block-coordinate descent algorithms for approximate nonnegative tensor factorization (NTF).

4: We propose an extrapolation strategy in-between block updates, referred to as heuristic extrapolation with restarts (HER). HER significantly accelerates the empirical convergence speed of most existing  block-coordinate algorithms for dense NTF,

5: in particular for challenging computational scenarios,

6: while requiring a negligible additional computational budget.

7: \end{abstract}

8: