Advanced Monte Carlo Methods for Inference in Complex Dynamic Models

Lead Research Organisation: University of Oxford
Department Name: Statistics

Abstract

Many physical phenomena and much data can be accurately modelled using statistical or proba-
bilistic models. Examples include the volatility of the stock market, gene expressions, radar signals,
relational data or images. However, even if it is possible to obtain realistic physical models or satisfactory
statistical models, only the simplest can be solved without the use of numerical methods.
Examples of the need for such numerical methods include non-linear non-Gaussian time series models,
Markov random fields, social networks models and so on. Thanks to the advent of enormous,
cheap computational power and the development of a plethora of complex inference mechanisms, it
is now possible, and in many real world systems it is becoming increasingly common, to employ
sophisticated simulation-based techniques to provide solutions to problems previously deemed
insoluble. The intention behind the research program discussed herein is to extend current, and devise
novel, simulation-based architectures to attack and efficiently solve problems that are still deemed
intractable.

Planned Impact

The development of advanced Monte Carlo methods for inference has multiple applications in a wide range of fields.
The following applications will be addressed during this research programme

* the development of more powerful methods for data assimilation problems arising in marine and atmospheric contexts.

* the development of efficient Monte Carlo inference methods in financial econometrics (stochastic volatility models) and structural econometrics (auctions models widely used in e-commerce).

* the development of Monte Carlo methods for inference for very large data sets which is becoming crucial in this era of "Big Data".

* the application to social networks analysis. Social networks are increasingly used to help understand phenomena as distinct as the spread of diseases, analyze friendship or corporate networks.

However Monte Carlo methods are already employed in many areas: computer graphics, data assimilation, ecology, econometrics, genetics, robotics, vision, signal processing, tomography, tracking, etc. Any significant development, properly disseminated, in this area should have attract a lot of interest and could have a large impact.


The longer term benefits of this project are also closely linked to the RCUK "Digital Economy" programme. For example the `digital hospital' component of this programme involves the real-time accurate data fusion and tracking of patients. This could directly benefit from the development of the techniques I plan to develop.

Publications

10 25 50
publication icon
Bardenet Remi (2017) On Markov chain Monte Carlo methods for tall data in JOURNAL OF MACHINE LEARNING RESEARCH

publication icon
Bishop A (2014) Distributed Nonlinear Consensus in the Space of Probability Measures in IFAC Proceedings Volumes

publication icon
Bishop A.N. (2014) Distributed nonlinear consensus in the space of probability measures in IFAC Proceedings Volumes (IFAC-PapersOnline)

publication icon
Bouchard-Côté A (2018) The Bouncy Particle Sampler: A Nonreversible Rejection-Free Markov Chain Monte Carlo Method in Journal of the American Statistical Association

publication icon
Bouchard-Côté, A. (2017) Particle gibbs split-merge sampling for Bayesian inference in mixture models in Journal of Machine Learning Research

publication icon
Cuturi M. (2014) Fast computation of Wasserstein barycenters in 31st International Conference on Machine Learning, ICML 2014

publication icon
Del Moral P (2015) Uniform Stability of a Particle Approximation of the Optimal Filter Derivative in SIAM Journal on Control and Optimization

publication icon
Deligiannidis G (2018) The correlated pseudomarginal method in Journal of the Royal Statistical Society: Series B (Statistical Methodology)

 
Description I have worked on new statistical methods to estimate the state and parameter of complex stochastic systems arising in a wide range of scientific fields such as econometrics, engineering and computational biology.
Exploitation Route Some of the methods I have developed are very general and could be used in a very wide range of applications.
Sectors Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Electronics,Energy,Environment,Financial Services, and Management Consultancy