Computer models, or simulators, have become a commonplace way to study physical systems in areas such astronomy, climate, disease propagation, epidemiology, and engineering. In recent years, attention has turned to how to use the simulators in real-world situations, how to combine simulations with field data or other models to build predictive models, and how to quantify uncertainty in predictions. The need for new statistical methodology for inference using computational models is pressing. The session will be of broad interest because of the variety of applications and methods. Indeed, all talks are impactful insofar as they create new methods for real-world applications In this session, there will be four talks and a panelist to discuss their findings. Each of the talks centers around developing new statistical methodology applied to real-world applications utilizing a computational models. The first talk considers statistical inference for agent-based models to assess potential mitigation and/or public policy strategies. In particular, describing and accounting for uncertainty in analyses impose

d by the use of generated (not true) populations remains a difficult task. This work was heavily used during the global pandemic. The second talk presents methodology for computer model emulation when simulators are used in a sequence (i.e., the output of one simulator is the input of the next). A new deep Gaussian process formulation is presented to model a complex climate simulator, and allow for fast exploration of climate outcomes. The third talk considers the use of auto-encoders (a type of neural net) as a statistical emulator of a climate model. This approach is motivated by the need to have fast prediction of simulator output for large-scale simulator residing on a super computer. Finally, the last talk develops new deep Gaussian process methodology for modeling non-stationary behavior in hurricane models. The methods will be compared and contrasted by an expert panelist. Speakers and abstracts: 1. Dr. Leanna House (lhouse@vt.edu) Department of Statistics, Virginia Tech. University, USA 2. Dr. Nathalie Klein (neklein@lanl.gov) Statistical Sciences Group, Los Alamos National Lab, USA 3. Dr. Dave Higdon (dhigdon@vt.edu) Department of Statistics, Virginia Tech. University, USA 4. Dr. Daniel Williamson (D.Williamson@exeter.ac.uk) Department of Mathematics, University of Exeter, UK 5. Panelist: Dr. Chunfang Lin (devon.lin@queensu.ca), Department of Mathematics and Statistics, Queen’s University, Canada

Computer models, or simulators, have become a commonplace way to study physical systems in areas such astronomy, climate, disease propagation, epidemiology, and engineering.  In recent years, attention has turned to how to use the simulators in real-world situations, how to combine simulations with field data or other models to build predictive models, and how to quantify uncertainty in predictions. The need for new statistical methodology for inference using computational models is pressing. The session will be of broad interest because of the variety of applications and methods. Indeed, all talks are impactful insofar as they create new methods for real-world applications in epidemiology, climate, and hurricane tracking.

Organiser: Dr Derek Bingham 

Chair: Dr Derek Bingham 

Speaker: Dave Higdon 

Speaker: Daniel Williamson 

Speaker: Leanna House 

Speaker: Nathalie Klein 

Discussant: Prof. Chunfang Devon Lin