The AstroStat Slog

I’ve copied below the text of an ISBA announcement for the first workshop in a new series addressing Bayesian methods for machine learning. It builds on the model of the earlier well-known “Bayesian case studies” workshops, where just a few applications are featured at each workshop, with the format tailored to produce a lot of back-and-forth between application scientists and statisticians.

One of the three topics for the October workshop is titled “Calibrating the Universe: a Bayesian Uncertainty Analysis of a Galaxy Simulation.” This sounds a bit reminiscent of the “cosmic calibration” work by a collaboration of astronomers and statisticians at Los Alamos. They are using a combination of parametric and nonparametric Bayesian methods and dimensional reduction and experimental design techniques to infer cosmological parameters from CMB, large scale structure, and Type Ia supernova data. Despite the similarity in nomenclature, this appears to be a different team and a different application. However, from what I can glean from the team, it’s the same kind of problem: implementing a parametric Bayesian analysis with a computationally expensive model, by building a fast nonparametric “emulator” for the model. Should be interesting.

Workshop on Case Studies in Bayesian Statistics and Machine Learning

The First Workshop on Case Studies in Bayesian Statistics and Machine Learning will take place on October 15th — 17th, 2009 at Carnegie Mellon University, Pittsburgh, PA. The Workshop will focus on applications of Bayesian Statistics and Machine Learning to problems in science and technology. It will feature three different tracks: In-depth contributed presentations and discussions of substantial research, shorter presentations by young researchers and poster presentations. The workshop builds upon the Case Studies in Bayesian Statistics Workshop which was held at CMU for the last two decades. In conjunction with the workshop, the Department of Statistics’ Eleventh Morris H DeGroot Memorial Lecture will be delivered by Professor Michael Jordan, University of California at Berkeley.

The invited case studies this year include:

Rigorous Error Analysis for Small Angle Neutron Scattering Datasets using Bayesian Inference
Chip Hogg, Jay Kadane, Jong Soo Lee and Sara Majetich

Decision theoretic Bayesian nonparametric inference for the molecular characterisation and stratification of colorectal cancer using genome-wide arrays
Christopher C. Holmes, Christopher Yau, Ian Tomlinson and Jean-Baptiste Cazier

and

Calibrating the Universe: a Bayesian Uncertainty Analysis of a Galaxy Simulation
Ian Vernon, Richard Bower and Michael Goldstein

YOUNG INVESTIGATOR ABSTRACTS DUE JULY 1

We are soliciting detailed abstracts (1 page) of proposed 15-minute presentations by young researchers (students or completed PhD within five years). These abstracts are due July 1, and should emphasize the scientific problems and how the inferential statistical and/or machine learning work solves the problems.

Contributed paper abstracts for posters are due September 1, 2009.

The organizing committee includes Jay Kadane, Ziv Bar-Joseph, David Blei, Merlise Clyde, Zoubin Ghahramani, David Heckerman, Tommi Jaakkola, Rob Kass, Tony O’Hagan, and Dalene Stangl.

Please submit abstracts via our webpage:

http://bayesml1.stat.cmu.edu/

which contains additional information, including abstracts of previous, successful case studies.

If you have questions, please contact Jay Kadane at kadane@stat.cmu.edu or any of the other organizers.