1: \begin{abstract}

2: 	In this paper, we study inexact high-order Tensor Methods

3: 	for solving convex optimization problems with composite

4: 	objective. At every step of such methods, we use

5: 	approximate solution of the auxiliary problem, defined by

6: 	the bound for the residual in function value. We propose

7: 	two dynamic strategies for choosing the inner accuracy:

8: 	the first one is decreasing as $1/k^{p + 1}$, where $p

9: 	\geq 1$ is the order of the method and $k$ is the

10: 	iteration counter, and the second approach is using for

11: 	the inner accuracy the last progress in the target

12: 	objective. We show that inexact Tensor Methods with these

13: 	strategies achieve the same global convergence rate as in

14: 	the error-free case. For the second approach we also

15: 	establish local superlinear rates (for $p \geq 2$), and

16: 	propose the accelerated scheme. Lastly, we present

17: 	computational results on a variety of machine learning

18: 	problems for several methods and different accuracy

19: 	policies.

20: \end{abstract}

21: