Posters at Adap[/MCM]’skiii

Due to a very small number of (declared) submissions—contrasting with the stable number of attendees—for the poster session on the evening of January 4 at Adap’skii, I have decided to cancel this session. The posters will be presented on the evenings of Wednesday, January 5 and Thursday, January 6, during MCMSki III. This will ensure an adequate audience and level of interaction for the poster presenters.

Christophe Andrieu being unfortunately unavailable at the present time, I am also unable to update the webpage prior to the meeting. Besides the above cancellation, the other changes in the program are

  • Jan. 3, 6pm: the discussant of Faming Liang’s talk is Pierre Jacob (Université Paris Dauphine & CREST)
  • Jan. 4, 8:30am: the discussant of Yves Atchadé is Luke Bornn (UBC, Vancouver)

Looking forward to the meetings in a few days! And to the 3 feet of fresh snow!!!

One Response to “Posters at Adap[/MCM]’skiii”

  1. Scott Schmidler sent us his abstract: for the same reason as above, I cannot include it within the webpage so here it is:

    Exploration vs. Exploitation in Adaptive MCMC

    The term “adaptive MCMC” currently encompasses a variety of different algorithms which involve modifying a Markov chain kernel in some manner based on the sample history. Most theory to date has focused on asymptotic convergence, yet adaptive algorithms are of interest precisely due to their potential for improving finite sample behavior. The proper way to characterize this is through the “mixing time”. We review results of Schmidler & Woodard (2010) showing that certain types of adaptation do not improve qualitatively on the mixing times of their non-adaptive Markov chain counterparts. We then argue for classifying algorithms as “exploratory” versus “exploitative”, based on their ability to improve mixing time versus autocorrelation for multimodal target distributions. We argue that a hybrid algorithm combining adaptations of both types is likely to be superior to using any single existing adaptive algorithm, and demonstrate this using an algorithm constructed from AMIS and Wang-Landau components. We conclude with some cautionary discussion about AMIS algorithms, and some results about Wang-Landau style algorithms, based on applications to high-dimensional distributions arising in molecular simulation.

    <Available from http://www.stat.duke.edu/~scs:

    Schmidler & Woodard (2010). Lower Bounds on the Convergence Rates of Adaptive MCMC Methods. (Submitted)

    Ji & Schmidler (2010). Adaptive Markov Chain Monte Carlo for Bayesian Variable Selection. Journal of Computational and Graphical Statistics, (to appear).

    Schmidler & Wiehe (2010). Reservoir Exchange and Adaptive Monte Carlo. (submitted)

    In preparation:

    Wang & Schmidler (2010). Exploration vs Exploitation: Hybrid Strategies for Adaptive MCMC. (in preparation)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.