The ExAMPLER project held their third workshop in Glasgow 29 – 30th April entitled Developing a roadmap for exascale modelling. ExAMPLER is all about exploring the role of exascale supercomputing for Agent Based Modelling (ABM). There are many situations, especially in the social sciences, where agents provide a highly accurate prediction of events and can be used where more traditional simulation techniques are not appropriate. However, many users of ABMs currently run these on their own desktops and would benefit greatly from being able to scale their simulations by leveraging supercomputing.
COVID-19 is an example here because, whilst it demonstrated the key challenge around understanding a wide range of factors and ABM was used to some extent, if it were possible to run these codes at scale then insights with improved accuracy could be delivered in a more timely fashion. As we move further into the exascale era, there are significant new capabilities that can be unlocked if ABMs are able to leverage these future supercomputers effectively.
This workshop followed from two visioning workshops which were held in 2023, where the participants discussed and identifed transformative long-term visions for agent-based modelling underpinned by exascale computing, along with associated software and institutional requirements. This workshop took those visions and explored what steps need to be undertaken to bring these into reality.
The event was attended by over 30 delegates including academics, stakeholders and practitioners. Talks on day 1 focused around the challenges of developing ABM for exascale computing (Richard Milton, UCL), the use of emulators from the SEAVEA ExCALIBUR project (Peter Challenor, Exeter) and potential case uses from Brunel and UCL. This was then followed on day 2 by discussions that further highlighted and explored various aspects of each of these challenges for agent-based modelling for exascale, ultimately building up a set of challenges associated with the visions that have been already developed.