Graphical model

A graphical model is a probabilistic model that uses a graph structure to encode conditional independence relationships among random variables. The term encompasses both Bayesian networks (directed acyclic graphs that encode asymmetric, often causal, dependencies) and Markov random fields (undirected graphs that encode symmetric, mutual interactions). More broadly, the concept extends to factor graphs, conditional random fields, and probabilistic graphical models in general — a framework that unifies graph theory, probability theory, and statistical inference into a single computational architecture.

The power of graphical models lies in the factorization they impose on joint probability distributions. By asserting that a variable is independent of all others given its neighbors in the graph, a graphical model replaces an exponentially complex joint distribution with a product of local, tractable factors. This is not merely a computational convenience. It is a claim about the world's structure: that complex systems are locally coupled, that global patterns emerge from neighborhood interactions, and that the map of those interactions — the graph — is the right level of abstraction for understanding the system's statistical behavior. The graphical model is the formal embodiment of self-organization in probability theory: local constraints producing global coherence without centralized control.

The graphical model does not merely describe how variables relate. It prescribes how information must flow for the system to maintain its coherence. The graph is not a map of reality. It is a protocol for computation.