[Developers] FW: [Fisheries Think Tank] Cole Monnahan and Jim Thorson, Feb 27th, 3:30-5pm, FSH106

[Mark Maunder]

From: miniworkshop-bounces at mailman2.u.washington.edu [mailto:miniworkshop-bounces at mailman2.u.washington.edu] On Behalf Of Kotaro Ono
Sent: Monday, February 25, 2013 11:07 AM
To: miniworkshop at u.washington.edu
Subject: [Fisheries Think Tank] Cole Monnahan and Jim Thorson, Feb 27th, 3:30-5pm, FSH106

Hi all!

Cole Monnahan (UW-SAFS) and Jim Thorson (NOAA-NWFSC) will be our next Think Tank speaker and they
will be talking on Wednesday February 27th, from 3:30-5:00pm in Fish 106 about:

"Next-generation MCMC: theory, options, and practice for Bayesian inference in ADMB"

If you'd like to attend remotely, please use Meeting ID #627-931-861,
by going to www.GoToMeeting.com<http://www.GoToMeeting.com>, clicking "Join a Meeting," and
entering the Meeting ID number.  For additional information, please
consult the document on the Think Tank website (see below).

The current schedule for the Winter 2013 series is posted at the Think
Tank website (http://fish.washington.edu/news/miniworkshop/index.html).

If you wish to subscribe/unsubscribe from this mailing list, you can
do so by visiting
https://mailman2.u.washington.edu/mailman/listinfo/miniworkshop and following
the directions.  There is also now a link to add a Google Calendar for
the Think Tank series, available on the Think Tank website.

Thanks,

Kotaro


Abstract:
Bayesian inference is a powerful framework for analysis and relatively simple to implement in ADMB. The Metropolis MCMC algorithm is typically efficient, however the posterior surfaces for some models present challenges and the standard method is highly ineffective and convergence extremely slow. In this mini-workshop we will detail the theory and application of the suite of tools available to ADMB users for conducting Bayesian analyses. We begin by reviewing the Metropolis algorithm currently implemented by ADMB for MCMC chains, including the -mcrb and -mcgrope runtime options. Building on this knowledge we detail how an analyst can force ADMB to use any covariance matrix and demonstrate the advantage of this approach on the notoriously difficult Schaefer model and a simple four parameter age-structured whale model. Moving beyond the standard Metropolis algorithm, there are two next-generation MCMC methods immediately available to the user: (1) the Metropolis-coupled MCMC (MCMCMC) and (2) the "hybrid" method. The theory of these two methods will be discussed and their application demonstrated.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/developers/attachments/20130225/7203d7d0/attachment.html>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ATT00001.txt
URL: <http://lists.admb-project.org/pipermail/developers/attachments/20130225/7203d7d0/attachment.txt>