The School main website is HERE
Scientific background and motivation
Bayesian Analysis approaches data based on an understanding of probability theory in its original sense of degree-of-belief, allowing a consistent treatment of inference and prediction. While methods used in the frequentist approach can be recovered under assumptions that do often hold in routine applications, the power of the Bayesian approach lies in its ability to tackle problems which cannot be treated by these conventional statistical methods.
While dating back more than 200 years, Bayesian methods were underutilised because the resulting computations were often too large to be done by hand. Massive increases in data collection and computational power have, however, changed this completely and have opened up Bayesian methods to more opportunities and applications than ever before. They are now used across all disciplines, including meteorology, biostatistics, medicine, economics and, of course, physics and astronomy.
Examples of applications on which Bayesian Analysis can now be brought to bear include
- dealing with uncertainty in measurement,
- determining probabilities for problems or events which are not based on counting or relative frequencies,
- proper handling of small datasets,
- quantitative comparison between different models or hypotheses, even in cases where chisquared tests or p-values fail,
- proper handling of nongaussian distributions.
Bayes systematically incorporates knowledge and facts relevant to the experiment which were available before the data was obtained. Taking proper cognizance of "priors" which formalise this knowledge is a major advantage in that it requires us to think and quantify what we consider relevant. While the mathematical formulation of priors can be difficult, the difficulty is at least "out in the open" rather than buried in hidden assumptions. Beyond technique, Bayesian Analysis also represents a change in world view, in that it places data and theory on an equal footing, and makes explicit provision for the observer.
Target audience
The School is aimed at both established researchers in physics and astronomy who regularly work with datasets and at postgraduate students. No previous knowledge of Bayesian inference as such is required. Participants should have some basic background in statistics as used in data analysis. Some funds for student support are available; details will be announced later.
School structure
- Morning sessions will provide an intensive theoretical introduction to the ideas and methods of Bayesian analysis, followed in the afternoons by theoretical and practical discussions and tutorials and hands-on computational work.
- Three invited speakers will conduct most of the above sessions. Details of the speakers appear below.
- Participants are encouraged to bring their own laptops and datasets in order to facilitate direct interaction and advice.
- The Bayesian Analysis Toolkit, an add-on for the CERN root package will be used as well as other Bayes-related software.
- Talks by regular participants on Bayesian analysis topics can be scheduled within limits.
- Posters by regular participants on their own existing data analysis are strongly encouraged in order to facilitate efficient communication on the problems and opportunities of analysis.
Venue
The School will be held in the Department of Physics, University of Stellenbosch and in the National Institute of Theoretical Physics housed in the Stellenbosch Institute for Advanced Studies. Stellenbosch boasts a unique combination of academia, vineyards, mountainous wilderness, endangered ecosystems, company headquarters, technological initiative and history. The Southern Ocean begins 20km south of Stellenbosch; the nature areas are accessible by bicycle and on foot from the town itself. The Coetzenburg sports centre and nearby schools have excellent facilities and have provided scores of national sports stars. Stellenbosch is endowed with strong civil society institutions and a functioning local government. Cape Town and its international airport are within easy reach.
Invited speakers
- Allen Caldwell is co-director of the Max-Planck-Institut für Physik in Munich active in neutrinoless double beta decay and future accelerator technology. He is co-author of the Bayesian Analysis Toolkit which will be used at the School.
- Giulio d'Agostini, is an experimental particle physicist at INFN and Rome University "La Sapienza", working on CERN and DESY detectors and data analysis. He has written a book on Bayesian Reasoning
- Roberto Trotta is a cosmologist and astroparticle physicist at Imperial College London with a string of Bayes-based articles; see also the compilation Bayesian Methods in Cosmology.
SKA conference:
The Bayes School overlaps in part with the annual Postgraduate Bursary Conference of the Square Kilometre Array project, which is being held at the same Stellenbosch Institute for Advanced Studies November 24 - 29. Due to the close proximity of venues, it will be possible for participants to register for both the SKA conference and the Bayes School.
