Six PhD studentships in Statistical Methodology and Its Application at University College London
The studentships are attached to the Department of Statistical Science at University College London, and a subset of them are UCL Impact awards. Impact awards support collaborative studentship projects with
organisations such as charities, companies, government institutions and social enterprises. The impact awards are joint with Lloyds bank, Xerox Research Centre Europe (www.xrce.xerox.com), and NCR Labs, respectively.
UCL is a member of the London Taught Course Centre (www.ltcc.ac.uk) that provides additional training in foundations of Mathematics and Statistics. UCL also offers training via its graduate school.
UCL is among the top-ten research institutions in the world and the Department of Statistical Science is one of the three largest statistics groups in the UK having a unique combined strength in Statistical Methodology and Machine Learning. The studentships are based in the Department of Statistical Science which has over twenty full time members of staff, including Professors Tom Fearn, Mark Girolami, Valerie Isham, Sofia Olhede and Trevor Sweeting. Together with other groups at UCL the department forms the Centre for Computational Statistics and Machine Learning (CSML) (www.csml.ucl.ac.uk), which is part of the European Network of Excellence PASCAL (www.pascal-network.org).
For informal inquiries please contact Professor Mark Girolami girolami@stats.ucl.ac.uk or Professor Sofia Olhede, sofia@stats.ucl.ac.uk. Candidates should complete the general UCL PhD application form, available at www.ucl.ac.uk/prospective-s
1. Advanced Monte Carlo Methods for Images and Text (with Xerox) – Prof. M.Girolami The Bayesian framework for statistical inference is largely dependent on numerical simulation for all but the most straightforward of statistical models. In the probabilistic representation of digital documents comprised of texts, images and embedded information, sophisticated statistical models are often required. It is hugely challenging to perform simulation based inference over these classes of models due to a variety of factors such as (1) exceedingly high number of parameters in the model, (2) the discrete nature of the configuration space, (3) lack of strong
likelihood-based identifiability and (4) strong posterior correlation of parameters. This project will seek to develop generic Monte Carlo sampling methods that addresses some of the issues listed above. The research will be carried out in close collaboration with Dr Cedric Archambeau (cedric.archambeau@xerox.com) and Dr Guillaume Bouchard (guillaume.bouchard@xerox.com)
2. Evolving Lead and Lag Times of Credit Cycles (with Lloyds) – Prof. S. Olhede
For policymakers and companies constructing accurate business and credit cycle indicators is pivotal for future planning, as well as for managing risk and troughs in cycles. Such indicators are constructed from observations of multiple time series, such as gross domestic product, production for certain sectors, employment, spread of interest rates, that is from collections of multiple observations of many different processes at several time instances. It is particularly important to identify leading credit cycle indicators in such data, as these will show effects of a changing financial climate ahead of other variables changing and thus they will allow for prior warning of a worsening or improving climate. This studentship will develop time series methods to estimate leads and lags, as well as the common cyclical structure in multiple time series, in particular accounting for evolving structure in the relationships to identify evolving leading indicators.
3. Geometric Markov chain Monte Carlo – Prof. M.Girolami
A recent paper read before the Royal Statistical Society developed sampling methodology based on Riemannian geometric principles and provided a way forward in systematically addressing some of the biggest challenges
faced in modern day computational statistics. The ability to design proposal mechanisms for Markov chain Monte Carlo (MCMC) that traverse geodesics and transform in a covariant manner across the statistical manifold brings great potential to what problems can conceivably be addressed. In this project the student will work on the further
development and analysis of this methodology from a number of possible perspectives such as considering alternative geometries.
4. Probabilistic Models for Adaptive Content Creation (with Xerox) – Prof. M.Girolami
This research project will focus on the development of structured prediction models to build document templates and learn to customize texts or sentences according to user preferences and habits. Conditional language models to generate human readable text based on the specific target application and device appropriate algorithms for the generation of small pieces of text, such as introductory sentences will also be developed. This project will draw upon recent advances in Natural Language Processing tools, Machine Learning algorithms and Stochastic Optimization techniques, in developing intelligent document creation tools. The research will be carried out in close collaboration with Dr Cedric Archambeau (cedric.archambeau@xerox.com) and Dr Guillaume Bouchard
(guillaume.bouchard@xerox.com)
5. Spatio-Temporal Statistical Models of Banknote Ageing (with NCR) – Prof. M.Girolami
A practical challenge to fully realising automated currency validation in Automated Teller Machines (ATM) is the variable quality of banknotes presented to the machine. It is desirable that a probabilistic generative model, and associated inferential machinery, of the ageing effects on banknote images be made available. This project will adopt advanced Bayesian modeling and inferential methodology in developing note ageing process models. NCR Labs collections of machine readable banknotes will be employed in formally assessing model adequacy as well as the ability to generate sample ageing profiles of banknotes. The theory, analysis, and methodology developed within this project will push the boundaries of spatio-temporal statistical modelling and presents a superb opportunity in making important advances in computational statistics in general.
6. Statistical Machine Learning methods for fMRI Analysis - Prof. M.Girolami
Functional magnetic resonance imaging (fMRI) is providing the means for both early detection of a number of neurodegenerative diseases and to study their origin and the mechanisms underlying them. Multivariate statistical methods show great promise in the systematic study and analysis of fMRI images and associated genetic data. There are still many methodological statistical challenges to be addressed in this research and the student will have the opportunity to work in a cross-disciplinary group seeking to develop appropriate statistical models and associated methods for this ongoing research.
------------------------------
Professor M.A. Girolami FRSE FIET
EPSRC Advanced Research Fellow
Chair of Statistics
Department of Statistical Science
University College London
1-19 Torrington Place
London, WC1E 7HB
Tel : +44 (0)20 7679 1861
Fax: +44 (0)20 3108 3105
email: girolami@stats.ucl.ac.
[ phd studenships full-scholarships.com ]
