Approximate Inference in Probabilistic Models
Lecture within the tutorial Statistical Methods in Learning
Speaker: Manfred Opper (Southampton University)
I present a framework for approximate inference in
which generalizes the TAP (Thouless, Anderson & Palmer) approach to the
statistical mechanics of spin-glass models.
Using an approximate expansion of a suitably defined Gibbs free energy
around a tractable model, we can compute certain posterior expectations
for Bayesian models with good accuracy. We give applications and show
that the method
can also be used for a faster computation of resampling averages.
This is joint work with Ole Winther
©Copyright 2004 Authors