Stochastic Optimisation for Large-Scale Inverse Problems

This repository contains the implementation of the numerical experiments for the preprint

M. Ehrhardt, Z. Kereta, J. Liang, J. Tang (2024) A Guide to Stochastic Optimisation for Large-Scale Inverse Problems

If you have any questions please contact the authors

Usage

We use the Core Imaging Library for tomographic imaging and as a baseline for reconstruction algorithms. To install CIL you may run

conda create --name cil -c conda-forge -c https://software.repos.intel.com/python/conda -c ccpi cil=24.2.0 ipp=2021.12

or follow the installation instructions on the linked github repository, where you may find further details, documentation and demos. This installation includes the supporting packages (e.g numpy)

The walnut dataset can be accessed by downloading 20201111_walnut_sinogram_data_res_280.mat in the 202001111_walnut_sinograms directory here

Contents

Results in the paper can be reproduced by running code as follows:

• Figure 4a-b: SheppLogan_200Epochs.py
• Figure 4c: SheppLogan_10Epochs.py
• Figure 5: SheppLogan_Sampling.py
• Figure 6: SheppLogan_ADAM.py and SheppLogan_SGDStepsize.py
• Figure 8-9: Walnut_Comparison.py
• Figure 10: Walnut_SVRGStepsizes.py

Stochastic optimisation yields faster CT reconstruction