University of Reading SIAM Student Chapter

CONF            


About

On Friday 7th June 2019 we hosted our annual SIAM-IMA Student Chapter Conference at the University of Reading! The day was a huge success with a range of topics in applied mathematics covered. We had three plenary talks, several student talks and a poster session. We ended the day with a drinks reception and a meal at the Queen's head, giving everyone the opportunity to get to know other PhD students and academics and find out ore about what they are working on. Check out our twitter for photos of the day here and our highlights of the day here


Click here to see our schedule and abstracts


Schedule


9:30 - 10:00 Registration and Tea/Coffee

10:00 - 10:10 Welcome address

10:10 - 11:05 Keynote: Dr. Karl-Mikael Perfekt, University of Reading
The spectrum of double layer potentials for some 3D domains with corners and edges

11:05 - 11:30 Erwin Luesink, Imperial College London
Stochastic Ocean Modelling: Why and How?
11:30 - 11:55 Tsz Yan Leung, University of Reading
Atmospheric predictability: the origins of the finite-time behaviour

11:55 - 1:00 Lunch and poster session

1:00 - 1:55 Keynote: Prof. Valentina Escott-Price, Cardiff University
Predictive modelling from genomic data

1:55 - 2:20 Tobias Schwedes, Imperial College London
Rao-Blackwellisation for parallel Markov Chain Quasi-Monte Carlo
2:20 - 2:45 Adriaan Hilbers, Imperial College London
Improving Power System Planning Under Climate-based Uncertainty

2:45 - 3:15 Tea/Coffee break

3:15 - 4:10 Keynote: Prof. Nick Trefethen, University of Oxford
Random Functions, Random ODES and CHEBFUN

4:10 - 4:35 Mariana Clare, Imperial College London
Where has all my sand gone?: Advanced numerical and statistical techniques to assess erosion risk in the coastal zone
4:35 - 5:00 Maha Kaouri, University of Reading
Globally convergent least-squares optimisation methods for variational data assimilation

5:00-6:00 Drinks reception
6:00 Leaving for dinner

Keynote speakers

CONF Prof. Nick Trefethen, University of Oxford
Professor of numerical analysis and head of the Numerical Analysis Group at Oxford, who has made many contributions in both the theory and applications of numerical analysis

Random Functions, Random ODES and CHEBFUN

CONF What is a random function? What is noise? The standard answers are nonsmooth, defined pointwise via the Wiener process and Brownian motion. In the Chebfun project, we have found it more natural to work with smooth random functions defined by finite Fourier series with random coefficients. The length of the series is determined by a wavelength parameter lambda. Integrals give smooth random walks, which approach Brownian paths as lambda shrinks to 0, and smooth random ODEs, which approach stochastic DEs of the Stratonovich variety. Numerical explorations become very easy in this framework.


CONF Prof. Valentina Escott-Price, Dementia Research Institute, University of Cardiff
Professor in the Medical Research Centre for Neuro-Psychiatric Genetics and Genomics who uses machine learning and bioinformatics for big data problems in genetics and neuroscience

Predictive modelling from genomic data

Psychiatric and neurodegenerative disorders have a complex, polygenic architecture in which a large number of genetic variants spanning a wide spectrum of population frequencies contribute to disease risk. In recent years, specific risk variants have begun to emerge from large-scale genomic studies. The standard approach to genome-wide association study (GWAS) data assumes an additive model, which, in statistical terms, is equivalent to looking for the main effects of variants contributing to disease risk. The assumption of additivity has been an extremely effective approach, but it is also pragmatic, since looking at the effects of many 100,000s of Single Nucleotide Polymorphisms (SNPs) would be rendered computationally expensive if all potential combinations of interactions were considered. In addition, the excessive dimensionality of such an approach would require very severe statistical correction for multiple comparison testing. Although testing for some interactions is now technically possible using Graphical Processing Units (GPU) instead of Central Processing Units (CPU), extremely large sample sizes will be required to achieve sufficient power to detect small genetic interaction effect sizes, as are expected in most complex genetic traits, at the very low significance thresholds dictated by multiple testing correction.
The extent to which genetic interactions contribute to disease risk is unknown. We investigated whether a support vector machine learning (SVM) approach can identify the presence of genetic interactions, without explicitly specifying interaction terms in regression models.
Support Vector Machines were introduced by Vapnik and Chervonenkis (1981) and are widely used due to their flexibility in analysing data with different distributions and their ability to deal with high-dimensional data such as gene expression. Previously SVMs using genetic variants as predictors were employed for the classification of populations. We illustrated the use genetic variants to distinguish schizophrenia patients from controls, where genetic differences between the groups are more subtle than between populations, and compared the results with Polygenic Risk Scores and Multivariate regression approaches.


CONF Dr. Karl-Mikael Perfekt, University of Reading
Lecturer of Pure Mathematics, working in operator theory, complex analysis, and spectral theory, recent winner of 2018 Zemánek prize in functional analysis, awarded by IMPAN
The spectrum of double layer potentials for some 3D domains with corners and edges

I will talk about the spectrum of double layer potential operators for 3D surfaces with rough features. The existence of spectrum reflects the fact that transmission problems across the surface may be ill-posed for (complex) sign-changing coefficients. The spectrum is very sensitive to the regularity sought of solutions. For L^2 boundary data, for domains with corners and edges, the spectrum is complex and carries an associated index theory. Through an operator-theoretic symmetrisation framework, it is also possible to recover the initial self-adjoint features of the transmission problem – corresponding to H^{1/2} boundary data – in which case the spectral picture is more familiar.

Registration

It would be great if you could join us! There is no registration fee, but registration is required and we will provide lunch for all registered participants. Please register by 26th May.

We strongly encourage students to submit an title for a 20 minute talk or a poster presentation by 8th May. We will contact you if your submission is successful. Thanks to our sponsors, there is limited funding to cover the travel (up to £50) and/or childcare expenses for speakers. There might be a possibility to support poster presenters as well.

We will finish the day with a drinks reception and a self-funded conference dinner.

Sponsors

Thank you to our sponsors, SIAM, IMA and LMS for making this conference possible.

Location

The conference will take place in Slingo lecture theatre in the JJ Thompson Building on the Whiteknights campus of the University of Reading. Informations about how to get to Reading and the campus can be found here. A map of the campus can be found here. JJ Thompson is building 3.