AAS 201: Special Session in AstroStatistics
AAS 201: Special Session in Astrostatistics
MAKING IT WORK:
PRINCIPLED MODEL-FREE DECONVOLUTION
VIA MULTISCALE METHODS
We are holding a Post-AAS
Interdisciplinary Workshop
Harvard Smithsonian Center for Astrophysics
Jan 15th and 16th, 2003
It will feature many of our key speakers,
plus other experts from other fields (and from astrophysics).
Web page should be available soon!
Format:
-
HIGHLIGHTS AND CHALLENGES FROM NEW AREAS OF PROGRESS:
-
Practical, Flexible Frameworks for Multiscale 'Deconvolution'
for both Guass-Normal and Poisson Data
Rebecca Willett and Robert Nowak, EE and Statistics, Rice University
-
No More Magic Numbers:
Bayes Multiscale Methods with Chandra Images
Margarita Karovska (CfA), David Esch (Harvard Statistics),
and the Harvard Astrostat Working Group
(Web page).
-
Just confirmed:
Albert Bijaoui, Observatoire de Nice
Web page; and
papers list.
Using multiscale methods (wavelets and their relatives) to extract
the "best representation" of one's image, spectra, or timing
data is always a topic of great practical interest. Recently,
a number of collaborations in Astrophysics, Electrical Engineering,
Statistics, and Medical Imaging have reached very exciting complementary
breakthroughs. These include:
- Handling non-gaussian errors (esp. Poisson)
- Handling instrument response (PSF, energy response, etc...)
- "No More Magic Numbers", e.g. no need to guess the stopping
iteration (in something like Richardson-Lucy), smoothness
parameter, or number of components;
- related to that: the ability to address uncertainties on fluxes and features in an image (or spectrum or light--curve);
- expanding all to a more flexible suite of basis functions
to better represent quirky astronomical images.
- We have a total of an hour and a half. Typically the speakers
leave plenty of time for discussion and commentary by learners
and experts in the audience.
- We plan for time at the end for a one-viewgraph
"advertisement" for each of the posters in an associated Poser
Special Session;
- We plan to develop a "wish-list" of multiscale and `model-free
deconvolution' challenges and problems! Come join us! Voice your
questions and opinions! Problems that were mentioned at last year's
AstroStat Special Session are now being actively worked on. Many
are close to being solved!
-
We also challenge participants to try out here own methods, and compare
them with the kinds of solutions suggested in the Tutorial sessions.
Sponsored in part by the AISR program
- Questions? Comments? Ideas for the next session?
- You are welcome to contact:aconnors@frances.wellesley.edu