1: \begin{abstract}

2: 	High-order tensor methods that employ Taylor-based local models of positive integer degree $p$ within adaptive regularization frameworks have recently garnered significant research interest for both convex and nonconvex optimization problems.

3: 	The well-known $p$th-order adaptive regularization (AR$p$) method has demonstrated optimal worst-case global convergence rates and local convergence rates.

4: 	At each iteration of this method, a local model of the objective function, consisting of a $p$th-order Taylor approximation and a $(p+1)$th-order regularization term, is minimized.

5: 	However, for $p \geq 2$, the subproblem typically contains multiple local minima.

6: 	It is recommended to solve the subproblem globally when $p = 2$, as the global minimizer is guaranteed to be a descent direction.

7: 	However, for any $p \geq 3$, it remains an open question how to efficiently minimize the subproblem and which minimizer to select.

8: 	In this context, designing efficient AR$p$ variants is a critical area of interest, especially for $p \geq 3$.

9: 	In this paper, we extend the interpolation-based updating strategies, originally introduced by [Gould, Porcelli and Toint, \textit{Comput Optim Appl}, Vol. 53, 2012, pp. 1–22] for $p=2$, to cases where $p \geq 3$. This updating strategy enables the regularization parameter to adjust adaptively based on interpolation models.

10: 	We propose a novel prerejection mechanism that rejects unwanted subproblem minimizers prior to any function evaluations, thereby reducing computational costs for $p\geq 3$.

11: 	Numerical results illustrate how the proposed interpolation-based updating strategy and the prerejection mechanism for $p = 3$ enhance the overall numerical performance.

12: \end{abstract}

13: