SKA Inference Workpackage Meeting

Existing inference algorithms, typically used in radio astronomy, require continued massive increases in computer performance for the next decade to be viable for the SKA. Computational requirements are a significant risk to the SKA project.

NZ has an unusual concentration of international research leaders in theoretical, computational, and practical aspects of Bayesian inference. These researchers have a wealth of experience in implementing both astronomical and industrial applications and can potentially make a significant contribution to the SKA. This research has the potential to lead to advances that  not only applicable to the SKA but also to industrial process monitoring and medical imaging.

Advances in several key areas are required for completing the science goals of the SKA:

Reduction in computation, and network costs

Inferential methods make optimal use of radio data, providing the potential for reduction in data pipelines, correlator network, and computation, without loss of information.

Improved resolution and link to science goals

Limitations in imaging resolution using existing 'signal processing' methods arise in part from the paradigm used to formulate these methods. Model-based inference can provide both improved resolution of images, and a direct evidence-based link between science goals and radio data.

Uncertainty Quantification and 'Hyper Images'

There is an opportunity to improve on traditional image catalogues by also quantifying uncertainty in images, leading to the production of hyper-images that contain this extra information. Post-processing of hyper-images for science objectives can then be built on this firm foundation, by providing estimates with quantified accuracies.