Jakub Bijak, Jason Hilton and myself attended the Uncertainty in Computer Models conference in Sheffield earlier this month (http://www.mucm.ac.uk/UCM2012.html).
Our poster was well-received, and attracted a good number of interested colleagues. Among those coming to speak to us were some colleagues from the National Oceanography Centre, also from Southampton, who had a number of posters and presentations at the conference, mostly to do with wind and wave models. Peter Challenor has been a major part of the MUCM project which spawned the conference, and he gave the initial presentation here in Southampton which inspired us to use the project’s tools on our agent-based model. We also spoke with Stephen Chick from INSEAD, several folks from Sheffield and UCL, and some gentlemen from industry including Ball Aerospace and Pratt & Whitney, and also a fellow from the US Air Force (he had some remarkable pictures of lasers mounted on fighter jets).
The attendees were very mixed, and presented work from a huge variety of fields. The commonality was a focus on quantifying uncertainty in models, whether those models were of climate, mechanical components, ocean currents, or in our case populations. Some of the talks were very intensely statistical, and demonstrated that there is a lot of cutting-edge work being done in this area — work that involves techniques we have yet to learn. But through attending the talks and speaking with other presenters, we now have a number of opportunities to learn more, and potential links for collaborative efforts.
What was most striking was to see the huge impact of this work on modeling techniques across this huge variety of fields. We saw how quantifying uncertainty and altering modelling techniques in response produced huge gains in efficiency for industrial applications, and how Gaussian emulators can allow for innovative modelling methods that save incredible amounts of computing time. We also heard some conceptual arguments regarding the effects of uncertainty in climate models; in particular, the talk on this topic by Jonty Rougier (delivered without slides, with him standing in front of the room barefoot) was extremely thought-provoking and one of the best talks of the conference. His talk was based on a book chapter which was recently completed for an upcoming volume entitled Conceptual Issues in Climate Modeling (U Chicago Press), and a draft of the chapter is available here:
While the talk mainly referred to climate models, the content was quite applicable to any discipline using highly complex computational models, so I recommend giving it a look.
All told, the conference was intense and challenging. We came away with some useful new contacts and new ideas, and with the distinct sense that our modelling efforts could stand to gain substantial legitimacy and novelty by applying these techniques of quantifying uncertainty. The response to our early-stage efforts was very encouraging, and we think that continuing with this will produce great dividends in our efforts to make our models both more powerful and more useful.