NIPS MCMC (Markov Chain Monte-Carlo methods) Workshop update

[Nando de Freitas]

Dear connectionists

The talks, software for the tutorial examples and several related papers
are now available from the NIPS MCMC for machine learning workshop page:

http://www.cs.berkeley.edu/~jfgf/nips99.html

 [ Moderator's note:   Here is the description of the workshop from the
   web page:

  MCMC techniques are a set of powerful simulation methods that may be
  applied to solve integration and optimisation problems in large
  dimensional spaces. These two types of problems are the major
  stumbling blocks of Bayesian statistics and decision analysis.  The
  basic idea of MCMC methods is to draw a large number of samples
  distributed according to the posterior distributions of interest or
  weighted such that it is possible to estimate simulation-based
  consistent estimates. MCMC methods were introduced in the physics
  literature in the 1950's, but only became popular in other fields at
  the beginning of the 1990's. The development of these methods is at
  the origin of the recent Bayesian revolution in applied statistics and
  related fields including econometrics and biometrics.  The methods are
  not yet well-known in machine learning and neural networks, despite
  their ability to allow statistical estimation to be performed for
  realistic and thus often highly complex models.

  Neal (1996) introduced MCMC methods, specifically the hybrid Monte
  Carlo method, into the analysis of neural networks. He showed that the
  approach can lead to many benefits. MCMC methods have also been
  successfully applied to interesting inference problems in
  probabilistic graphical models. However, many recent advances in MCMC
  simulation, including model selection and model mixing, perfect
  sampling, parallel chains, forward-backward sampling and sequential
  MCMC among others, have been overlooked by the neural networks
  community. This workshop will attempt to provide a simple tutorial
  review of these state-of-the-art simulation-based computational
  methods. It will also focus on application domains and encourage
  audience participation. Speakers will be encouraged to keep the
  presentation at a tutorial level.

    -- Dave Touretzky, CONNECTIONISTS moderator ]

Happy New Year !!!
Nando

--

Computer Science Division            | Phone : (510) 642-2038
387 Soda Hall                        | Fax   : (510) 642-5775
University of California, Berkeley   | E-mail: jfgf at cs.berkeley.edu
Berkeley, CA 94720-1776 USA          | URL   : http://www.cs.berkeley.edu/~jfgf