A graphical model is a graph in which nodes represent variables, and edges represent conditional probabilities. Graphical models include Bayesian networks (also known as belief networks, probabilistic networks, causal network, and knowledge maps). Hidden Markov models and neural networks are also graphical models.
Because graphical models are graphs, evaluating graphical models is a form of Graph Traversal. Graphical models include a probabilistic aspect that can require a small amount of computation per node, but the basic processing is otherwise similar.
Constructing graphical models usually involves processing many observations and variables. As each observation is processed, the relevant variables within the model is updated.
In practice, a single graphical model may be evaluated many times for a single problem, or many graphical models can be evaluated for a single input. For example, in speech recognition, sound may be broken into frames of discrete, short segments of time; each frame may be evaluated against many models to derive a distribution of probabilities that the frame matches particular phonemes. Because independent graphical models or inputs can be evaluated independently, a graphical model may be able to be parallelize very simply.
Parallel construction of graphical models can be complicated by the possibility of conflicts in the part of the graph that is being updated.
Graphical models are often used in artificial intelligence and machine learning applications. Typical examples that make use of graphical models are speech and image recognition.