1: \begin{abstract} \small\baselineskip=9pt 

2: The cubic regularization method (CR) 

3: and its adaptive version (ARC)

4: are popular Newton-type methods in solving unconstrained non-convex optimization problems, due to its global convergence to local minima under mild conditions. 

5: The main aim of this paper is to  

6: %Cartis et al. proposed the adaptive cubic regularization method (ARC) which accelerates %the empirical convergence of CR. In this paper, we 

7: develop a momentum accelerated adaptive cubic regularization method (ARCm)

8: to improve the convergent performance.

9: %that further improves the numerical performances of ARC. 

10: With the proper choice of momentum step size, we show the  

11: global convergence of ARCm and the local convergence can also be guaranteed under the \KL property. Such global and local convergence can also be established when 

12: inexact solvers with low computational costs are employed in the iteration procedure. 

13: %The global and local convergence is also developed for ARCm with inexact cubic %regularized subproblem solutions as exact solutions are computationally expensive, %especially in high-dimensional problems. 

14: Numerical results for non-convex logistic regression and robust linear regression models 

15: %on three high-dimensional and large-scale datasets 

16: are reported to demonstrate that

17: %the effectiveness of the proposed ARCm. In particular, 

18: %ARCm 

19: %investigate the performances of the proposed ARCm. Experimental results show that 

20: the proposed ARCm significantly outperforms state-of-the-art cubic regularization methods (e.g., CR, momentum-based CR, ARC) and the trust region method. In particular, 

21: the number of iterations required by ARCm is less than 

22: 10\% to 50\% required by the most competitive method (ARC) in the experiments.

23: %The proposed ARCm is 2 to 10 times faster than CRm and $10$\%-$50$\% faster than %ARC in terms of iterations for convergence.

24: \end{abstract}

25: