Today, almost every aspect of space-based research is undergoing a revolution in data volumes and quantities. It is fair to say that we are entering an era of exascale data volumes in astronomy, astrophysics, planetary science and related fields. This interest group aims to address AI modelling approaches, big data and exascale computing challenges in diverse fields ranging from solar system exploration and extrasolar planets to understanding our local environment to deciphering the evolution of galaxies and the universe as a whole.
Though the science is very diverse, approaches to big-data analysis are often similar. This interest group aims to provide a communal space to foster and develop interdisciplinary approaches between astronomy and machine learning communities.
Explaining the science
The science of this interest group encompases a broad domain ranging from surface characterisation of solar system planets, to understanding the evolution of planets, stars, galaxies and our universe as a whole. What all these disciplines have in common is the use of big-data techniques to interpret an ever increasing flow of data. For example, the Square Kilometer Array (SKA) radio telescope alone will provide annual data rates of 600 Pb and is but one of many data-intensive projects and missions of the 2020’s. The techniques required to analyse these observations encompass the full range of contemporary machine learning, from supervised, unsupervised to reinforcement learning, Explainable AI methods, Bayesian inverse modelling and physics informed hybrid models. The aim of this group is to find and build on the commonality between machine learning approaches across a diverse set of space science disciplines.
- Foster collaborative approaches and exchange of ideas within The Turing’s growing astronomy community. Run interactive discussion workshops, hackathons and conferences.
- Provide a unified forum to engage with other interest groups on large astronomical data projects.
- Provide a platform for the wider UK and international AstroML community to engage with the Turing community.
- Organise unified responses to grant calls and sessions at leading conferences.
- Promote early-career researchers (through talks/colloquia) and promote ML in astronomy as a career path in the UK
- How can we make full use of AI and machine learning in the phase spaces of large-scale simulations and observation spaces to fuse the worlds of simulations and observations in a unified principled framework?
- How do we best account for observational measurement errors in AI models?
- How do we best guard against and monitor domain drift in production models?
- What are the best strategies to mitigate class imbalance in astronomical data sets?