PhD Studentship in Uncertainty Quantification Project, Heriot-Watt University, Edinburgh, UK

PhD studentship in Uncertainty Quantification project

Subject: PhD studentship at Heriot-Watt University, Edinburgh

Machine learning for uncertainty quantification of reservoir models

Uncertainty of complex natural systems is difficult to describe effectively within traditional parametric modelling approach. Many of relations in natural systems are vague and not transparent to capture and describe analytically. Machine learning provides an alternative way to detect dependencies from uncertain data and propagate them into forecasting models.

Machine learning will enable to integrate oblique data, which are not directly related to the modelling subject but still bear non-parametric relation with the modelled dependencies, into the uncertainty modelling framework. Data-driven statistical learning methods such as Support Vector Machines and Artificial Neural Networks are foreseen to be used to capture and describe dependencies in regression and classification problems, and define informative multidimensional prior models. Machine learning approach fits perfectly to tackle the task to eliciting and modelling prior information in high dimensions from multiple sources with feature selection/extraction techniques. Geomanifold modelling based on a recently emerged theory of semi-supervised learning is capable of integrating different sources of information from high dimensional input space.

The challenge of the project is to link contemporary machine learning algorithms with the state-of-the-art Bayesian framework for uncertainty quantification developed in the group. The work will involve application of advanced machine learning methods and recent achievement in multiple point geostatistics.

The candidate will join a dynamic and diverse team of post-docs and PhDs lead by Prof. Mike Christie. The research carried out by the team addresses aspects of uncertainty quantification including stochastic optimisation methods, geostatistics, machine learning, and employs high-level scientific computation (including a 84 node Linux cluster). The research is funded by a consortium of oil companies, and the skills acquired in while studying for the PhD are likely to be applicable to a wide range of areas, including the oil industry.

Desired skills:
The successful candidate must have a strong background in machine learning and numerical methods as well as advanced computational skills. Knowledge of geosciences modelling, geostatistics and petroleum engineering in particular will be beneficial.

To apply send a CV to Dr V Demyanov, vasily.demyanov[ at ]pet.hw.ac.uk

Closing date: 31st July 2009.