Structured problems
From GusWiki
i.e. statistical problems in which dependencies exist in the data (structure input or structured data), or in the parameters to be inferred (structured output or structured prediction). These dependencies can be logical (e.g. see Markov Logic, FOPI), or probabilistic, depending on the scientific problem being addressed.
A simple example of a structured problem is performing a regression on each node, while knowing that neighboring nodes tend to be similar to each other. This can often be viewed in a hierarchical Bayesian framework, in which inference can be interpreted as shrinkage.
In case the structure is "continuity" (i.e. a bias towards small derivatives), inference amounts to smoothing. See -:Kalman filter.
Continuous Structure
Discrete Structure
Rather than taking real numbers, your variables can take as values general data structures.
This is common in NLP (grammars with complex dependencies), bioinformatics.
Noah Smith's tutorial - Structured Prediction for Natural Language Processing
