| Main Page | | Schedule | |
The workshop will provide a forum to discuss unsolved issues, both practical and theoretical, pertaining to the application of approximate Bayesian inference in continuous variable models. The emphasis of the workshop will be in understanding the particular difficulties in this class of models and the differential strengths and weaknesses of available variational approximation techniques, as opposed to the arguably better understood field of approximate inference in discrete variable systems.
The workshop is now over, but slides and links are still here as an archive.
Many of the most important problems in Machine Learning and related application areas are most naturally and succinctly treated using continuous variable models. Several important continuous latent variable models come to mind, each underlying a host of applications:
While a range of deterministic (variational) approximation techniques have been applied successfully to these problems, many important questions remain open so far. Which properties of a posterior distribution make its approximation hard? Are these properties specific to certain techniques, or universal? How do existing methods compare? Can we explain their biases? Is numerical instability a characteristic of certain techniques only, or is it an indicator for the hardness of a posterior? For some methods, like Expectation Propagation, a convergence proof is still lacking. Others, like Variational (mean field) Bayes, use a "wrong" (exclusive) information divergence. Can we understand when and how much this biases the final results?
We welcome participants to share their experiences on practical problems. However, in contrast to the usual ``success stories'' for an established method, we invite descriptions of practical difficulties in applying approximate inference methods, and how these were analyzed and dealt with in the application in a principled manner.
We especially encourage contributions from related fields, such as (Markov Chain) Monte Carlo or Information Geometry, if an effort is made towards bridging the gap between the fields and addressing goals of the workshop. In this context, we do encourage contributions of tutorial nature or of preliminary ideas.
The workshop is therefore intended to appeal both to practitioners with insight into the difficulties in approximate inference in continuous systems, and to theorists with an interest in characterizing the complexity of posterior distributions or in analyzing properties of approximate inference methods.
The workshop is sponsored by EU FP6 PASCAL Network of Excellence and Microsoft Research, Cambridge, .
Page last updated on Monday 10 Dec 2007 at 17:49