FranceCountry of destination:
United States of America
Project Statistical validation of simulator
Simulators (a.k.a. computer models) are mathematical models of complex real-world processes. They are a crucial ingredient in most fields of science, engineering, medicine and business. Running the simulator at a chosen set of inputs on a computer is sometimes named a numerical experiment or an experiment in silico. These experiments replace wet lab experiments or physical experiments when they are too costly or impracticable. From a statistical point of view, the main issues when working with simulators is uncertainty quantification (UQ). Uncertainties are present at different levels of the simulators. The uncertainty quantification task consists in taking into account all these sources of variability and to determine to what extent the simulator is reliable to describe the real-world process.
My project is particularly concerned with the statistical validation of simulator. To validate a simulator, field data (often obtained as noisy measurements of the real-world process) are confronted with the simulator outputs. The goal of statistician validation is to provide decision tools to simulator users to decide whether or not the simulator is accurate enough for intended use. The two main axes for developing new methodologies in statistics are the model comparison approach for validation and the selection of (static and adaptive) design of experiments adapted to the validation task. Then, collaborations with disciplinary sciences such as agronomy or environmental science will be the occasion to apply developed methods on real case simulators.
The project plan spreads on 18 months. The first nine months are the outgoing phase at SAMSI which would be funded thanks to AgreenSkills fellowship and the next 9 months are the incoming phase at AgroParisTech. The outgoing phase is mainly dedicated to statistical developments and the incoming phase is more centered on applications in collaboration. I already work with simulators for which we have field data where it will be interesting to apply the new developed techniques. It concerns a simulator of nitrogen transfer in the landscape, a simulator of wheat growing and a simulator of photovoltaic power plant.
Since my Ph.D. entitled Kernel interpolation for expensive black box functions which dealt with the emulation of timeconsuming simulator (defended in 2010), I have still worked within the field of uncertainty quantification of simulator. I had several works dedicated to this subject in different contexts in several collaborations with EDF (French electricity supplier), HydroQuebec (Quebecian energy supplier) and with academic colleagues in genetics and agroecology. I was involved in the cosupervision of 5 Ph.D. students and 2 post-doctoral fellows. I am the main coordinator of one funded project MIRES dedicated to the interdisciplinary study of seed sharing networks and the local coordinator of another. I met Pr. James Berger in 2016 during the poster session of a conference in Warwick. I was presenting a work on validation of simulators in a simplified context. He told me about the thematic year in SAMSI in 2018-2019 and proposed me to attend this thematic year during this year as a visiting professor. I then applied to the AgreenSkills fellowship to fund my mobility.
Damblin, G., Barbillon, P., Keller, M., Pasanisi, A. et Parent, E., 2018. Adaptive numerical designs for the calibration of computer codes. Journal of Uncertainty Quantification, 6(1). Doi: 10.1137/15M1033162.
Courbariaux, M., Barbillon, P., & Parent,E., 2017. Water flow probabilistic predictions based on a rainfall-runoff simulator: a two-regime model with variable selection. Journal of Agricultural, Biological and Environmental Statistics, 22(2). Doi: 10.1007/s13253-017-0278-5.
Barbillon, P., Barthelemy, C., et Samson, A., 2016. Parameter estimation of complex mixed models based on meta-model approach. Statistics & Computing 27(4). Doi: 10.1007/s11222-016-9674-x.
Damblin, G., Keller, M., Barbillon, P., Pasanisi, A., et Parent, E., 2016. Bayesian Model Selection for the Validation of Computer Codes. Quality and Reliability Engineering International, 32(6). Doi: 10.1002/qre.2036.
Auffray, Y., Barbillon, P., Marin, J.-M., 2014. Bounding rare event probabilities in computer experiments. Computational Statistics & Data Analysis, 80. Doi: 10.1016/j. csda.2014.06.023.