Other forms of Bayesian inferencing has been applied in a variety of areas. For instance, Mosteller and Wallace [85] show how Bayesian classification can be used to establish authorship, in particular, for the disputed authorship of the Federalist Papers. Simplifying somewhat, the model can be viewed as combining the likelihood of authorship given keywords that are highly indicative of particular authors. The techniques they developed have been influential in later statistical NLP work [44,126].
Geman and Geman [45] show how Bayesian inference can be used for image restoration. The basic model is that the probability that a pixel is in a given state can be approximated by just considering the states of the pixels in the immediate neighbor. This allows for image restoration through the use of local computations. An important aspect of this work is the illustration of Gibb's Sampler for inferencing. This is a form of stochastic simulation that is similar to simulated annealing. In simulated annealing, the probability that the state of a node changes decays over time. Initially state changes are very likely in order to avoid getting stuck in local minima, but later state changes become unlikely in order to facilitate stablization. See [30] for the use of simulated annealing in word-sense disambiguation.