MMath Mathematics and Statistics / Course details
Year of entry: 2021
- View tabs
- View full page
Course unit details:
Advanced Uncertainty Quantification
|Unit level||Level 4|
|Teaching period(s)||Semester 2|
|Offered by||Department of Mathematics|
|Available as a free choice unit?||No|
This unit introduces theoretical tools and numerical methods for incorporating random inputs into models consisting of differential equations. We begin by introducing stochastic processes and random fields and numerical methods for simulating them. We then introduce the multilevel Monte Carlo method for propagating uncertainty in ODE models with random inputs and sparse grid techniques for estimating intergrals in high dimensions. Finally, we investigate intrusive and non-intrusive surrogate modelling techniques in the form of stochastic Galerkin approximation and Gaussian process regression.
Although the concepts and tools introduced in this module will require a theoretical grounding, the primary intention is to focus on the application of the methods to models consisting of ordinary and partial differential equations, derived from environmental, industrial and biological applications. Computational exercises will reinforce understanding of the methods introduced and their theoretical properties.
|Unit title||Unit code||Requirement type||Description|
|Introduction to Uncertainty Quantification||MATH44071||Pre-Requisite||Compulsory|
- Represent second order random fields as series expansions and explain key theoretical results.
- Describe and implement numerical methods for generating realisations of second order random fields on one and two-dimensional domains.
- Apply multilevel Monte Carlo sampling to ODEs with random inputs in combination with standard time-stepping methods, and analyse the associated error.
- Construct and implement standard tensor product quadrature rules in multiple dimensions and explain their disadvantages.
- Derive sparse grid approximation rules, apply them to the computation of expectations and other statistical quantities of interest, and state key approximation theory results.
- Explain the concept of a surrogate model for differential equations with random inputs and state common intrusive and non-intrusive approaches.
- Define the concept of a weak solution for test problems consisting of differential equations and derive the finite-dimensional problems associated with Galerkin approximation.
- Recognise families of orthogonal polynomials associated with common probability distributions and explain how to construct appropriate spaces of multivariate polynomials for stochastic Galerkin approximation.
- Describe and implement stochastic Galerkin approximation schemes for test problems consisting of differential equations with random inputs, and perform error analysis.
- Explain how to apply Gaussian process regression to approximate a function whose value is known only at a finite set of points and derive the predictive distribution from the prior.
- Implement Gaussian process regression for selected test problems consisting of differential equations with random inputs and analyse the properties of the predictive mean.
1. Representation of Random Inputs 
Stochastic processes/random fields. Stationary and isotropic cases. Covariance functions and regularity results. Mercer's theorem. Hilbert–Schmidt theorem. Karhunen-Loeve expansions. Examples of ODEs and PDEs with random inputs.
2. Numerical Methods for Generating Random Fields 
Cholesky factorisation, singular value decomposition, circulant embedding in one dimension.
3. Sampling-based methods for uncertainty in ODEs 
Multilevel Monte Carlo sampling. Telescoping sums. Error analysis and comparison to standard Monte Carlo sampling.
4. Numerical Integration 
Review of Newton-Cotes and Gauss rules in one dimension. Tensor product rules. Sparse grid integration and interpolation in higher dimensions.
5. Galerkin approximation 
Hilbert spaces. Riesz representation theorem. Lax-Milgram Lemma. Weak solution of differential equations. Galerkin approximation.
6. Stochastic Spectral Methods 
Univariate orthogonal polynomials. Legendre and Hermite polynomials. Multivariate orthogonal polynomomials. Stochastic Galerkin approximation.
7. Gaussian Process Regression. 
Statistical models. Linear regression. Gaussian processes and conditioning. Choice of prior. Approximation theory and link to radial basis functions.
- Mid-semester coursework: 20%
- Written exam : 80%
Feedback tutorials will provide an opportunity for students’ work to be discussed and provide feedback on their understanding. Coursework or in-class tests (where applicable) also provide an opportunity for students to receive feedback. Students can also get feedback on their understanding directly from the lecturer, for example during the lecturer’s office hour.
Ralph Smith, Uncertainty Quantification, SIAM, 2014.
C. E. Rasmussen & C. K. I. Williams, Gaussian Processes for Machine Learning, the MIT Press, 2006
J. Voss, An Introduction to Statistical Computing: A Simulation-based Approach, Wiley, 2013.
T.J. Sullivan, Introduction to Uncertainty Quantification, Springer, 2015.
G.J. Lord, C.E. Powell, T. Shardlow. An introduction to computational stochastic PDEs. Cambridge University Press, 2014.
|Scheduled activity hours|
|Independent study hours|
|Catherine Powell||Unit coordinator|
This course unit detail provides the framework for delivery in 20/21 and may be subject to change due to any additional Covid-19 impact.
Please see Blackboard / course unit related emails for any further update.