| EPSRC Reference: |
EP/I026827/1 |
| Title: |
Projective Limit Techniques and Representation Theorems in Bayesian Nonparametrics |
| Principal Investigator: |
Orbanz, Dr P |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Engineering |
| Organisation: |
University of Cambridge |
| Scheme: |
Postdoc Research Fellowship |
| Starts: |
01 May 2011 |
Ends: |
30 September 2012 |
Value (£): |
289,421
|
| EPSRC Research Topic Classifications: |
| Statistics & Appl. Probability |
|
|
| EPSRC Industrial Sector Classifications: |
| No relevance to Underpinning Sectors |
|
|
| Related Grants: |
|
| Panel History: |
| Panel Date | Panel Name | Outcome |
|
15 Feb 2011
|
PDRF Maths Interview Panel
|
Announced
|
|
01 Feb 2011
|
PDRF Maths Sift Panel
|
Announced
|
|
|
Summary on Grant Application Form |
|
Statistical models provide the mathematical means to make sense of data -- to analyze and interpret its contents, and to translate it into decisions and predictions. A recent development is a class of models, called nonparametric Bayesian models in statistics, that are capable both of integrating prior knowledge about the data and of adapting to the complexity of a given data set. My research will contribute to our understanding of the mathematical mechanisms underlying these models, and to our ability to apply them to new problems.My approach builds on two mathematical concepts, projective limits techniques and representation theorems derived from symmetry properties of data. Projective limits are mathematical tools that assemble a complex mathematical object from many simple components. Based on the projective limit tools available in probability theory and other branches of mathematics, I will develop projective limit methods for nonparametric Bayesian models. Such methods will allow us to solve problems involving complex models -- studying their mathematical properties, or applying them to data -- by solving the corresponding problems for simpler models and re-assembling the solutions.A complementary question to the properties and evaluation of a specific model is which model to use for given data -- the fundamental question facing every dataanalyst in practice. Representation theorems relate symmetry properties of data to the class of models compatible with these properties. These are deep mathematical results of great practical utility: They translate simple, intuitive properties of data into characterizations of models adequate for such data. The most well-known example, de Finetti's theorem, has long been a corner stone of Bayesian statistics, but many results have only emerged as recently as the past decade.In my research, I will study (1) the construction and analysis of nonparametric Bayesian models by means of projective limits, (2) the derivation of models from recent results on symmetry principles and representations, and (3) the interplay between these two formalisms.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.cam.ac.uk |