Adaptive parameterisation in Markov chain Monte Carlo methods
The International Centre for Numerical Methods in Engineering (CIMNE) is a research centre, created in 1987 by consortium between the Catalan Government and the Universitat Politècnica de Catalunya (UPC-BarcelonaTech), devoted to the development and application of numerical methods to a wide range of areas in engineering. CIMNE has been selected as a Severo Ochoa Centre of Excellence for the period 2019-2023. This is the highest level of recognition of excellence and leadership awarded to a research centre in Spain.
Sergio Zlotnik and J.C. Afonso
Number of vacancies: 1
Category: PhD (PHD2)
Salary (gross): 17.563,14 EUR
Weekly working hours: Full time
Duration: 3 years
Starting date: No later than Sept 2021
CIMNE is looking for a PhD Researcher to be part of the Research and Technical Development (RTD) Group on Credible data-driven models.
The functions assigned to the candidate will be:
During the last 20 years the Solid Earth community pursues massive data-driven simulations and joint inversions for the physical state of the Earth's interior with unprecedented complexity and resolution. Traditional inversion techniques applied to the problem of characterising the thermal and compositional structure of the upper mantle are not well suited to deal with the nonlinearity of the problem. Probabilistic inversions, on the other hand, offer a powerful formalism to cope with these difficulties. The parameterisation usually used in geophysical inversions is based on very fine structured grids, where the properties of each cell is considered independent. This simple parameter space is chosen as the structures to be recovered include sharp contrasts, discontinuities and complex spatial shapes. Although, the grid based parameterisation in three dimensional domains produce an enormous number of parameters to determine, exceeding 500.000 unknowns in practical cases. Several drawbacks arise: first, solving an inverse problem to determine 500.000 parameters is computationally exhausting. Second, this kind of parameterisation reduces the sensitivity of each parameter while the grid is refined and, therefore, increases the difficulty of the inversion. In this thesis we want to explore model reduction techniques that seek low-dimensional representations of parameters. The goal is producing an adaptive parameterisation effectively reducing the number of parameters. This is expected to accelerate the probabilistic inversion algorithm and facilitate efficient sampling in the reduced parameter space. The result will be a tractable procedure for the solution of statistical inverse problems involving partial differential equations with high-dimensional parametric input spaces.
The requirements and merits will be evaluated with a maximum mark of 100 points. Such maximum mark will be obtained by adding up the points obtained in the following items:
Candidates must complete the "Application Form" form on our website, indicating the reference of the vacancy and attaching the following documents in English:
The deadline for registration to the offer ends on 31st May, 2021 at 12 noon.
The shortlisted candidates may be called for an interview. They may also be required to provide further supporting documentation.
CIMNE is an equal opportunity employer committed to diversity and inclusion. We are pleased to consider all qualified applicants for employment without regard to race, colour, religion, sex, sexual orientation, gender identity, national origin, age, disability or any other basis protected by applicable state or local law. CIMNE has been awarded the HRS4R label.