1: \begin{abstract}

2: Model-based iterative reconstruction plays a key role in solving inverse problems.

3: However, the associated minimization problems are generally large-scale, ill-posed, nonsmooth,

4: and sometimes even nonconvex, which present challenges in designing efficient iterative solvers

5: and often prevent their practical use.

6: Preconditioning methods can significantly accelerate the convergence of iterative methods.

7: In some applications, computing preconditioners on-the-fly is beneficial.

8: Moreover, forward models in image reconstruction are typically represented as operators,

9: and the corresponding explicit matrices are often unavailable,

10: which brings additional challenges in designing preconditioners.

11: Therefore, for practical use, computing and applying preconditioners

12: should be computationally inexpensive.

13: This paper adapts the randomized \NY approximation

14: to compute effective preconditioners that accelerate image reconstruction

15: without requiring an explicit matrix for the forward model.

16: We leverage modern GPU computational platforms to compute the preconditioner on-the-fly.

17: Moreover, we propose efficient approaches for applying the preconditioner

18: to problems with nonsmooth regularizers.

19: Our numerical results on image deblurring, super-resolution with impulsive noise,

20: and computed tomography reconstruction

21: demonstrate the efficiency and effectiveness of the proposed preconditioner.

22: \end{abstract}

23: