Training | Parallel Algorithms | Uncertainties
Required Education : M2 or equivalent
Start date : 1 March 2023
Mission duration : 6 months
Deadline for applications : 1 May 2023
Salary : 650€/month
General information
- Starting date: March – May 2023
- Duration: 6 months
- How to apply: Send a CV and a cover letter to mycek@cerfacs.fr and mohamed-reda.el-amri@ifpen.fr.
Description
Multifidelity techniques for statistical estimation constitute an emerging and rising branch of numerical methods for uncertainty quantification [Peherstorfer et al., 2018]. In an industrial context where the numerical simulators are expensive to evaluate (in terms of CPU time), pure Monte Carlo estimation would lead to prohibitively long estimation. On the other hand, in high-dimensional uncertain input spaces, surrogate models constructed from a reasonable (limited) number of training data points typically lead to significantly high model error (bias). Recent approaches based on the variance reduction technique called control variates propose to combine Monte Carlo sampling of the high-fidelity (expensive) numerical simulator with low-fidelity surrogate models, to make the most of both worlds [Yang et al., 2022]. This internship aims to explore alternative formulations of the approach proposed in [Yang et al., 2022].
Further information: detailed internship subject
