Bayesian Network¶
Represent a joint probability law by a directed acyclic graph and one conditional distribution per variable given its parents.
Core Idea¶
A Bayesian network models uncertain variables with a directed acyclic graph and a conditional probability distribution for each variable given its parents. Multiplying these local distributions yields the joint law. The graph also specifies conditional independences implied by this particular factorization. An arrow is a probabilistic parent relation, not automatically a causal claim.[^ref-c55474a3e0be]
Scope of Application¶
The pattern works with very different variables: Shwe and Cooper modeled diseases and findings in QMR-DT, while NASA's ADAPT research modeled electrical-power components and sensor evidence for fault isolation. In both cases the same DAG-plus-local-conditionals identity remains, though the variables, probabilities, and validation differ. Neither example is medical or engineering operational advice.[ref-3b8409ef7138][ref-13dc79b9bad9]
Clarity¶
Do not equate Bayesian network with every probabilistic graphical model: undirected models and factor graphs use different local forms. Nor is d-separation a synonym; it is a test for independence statements guaranteed by a DAG's model. If two nodes are not d-separated, the graph alone does not prove they are dependent in every numerical parameterization.[ref-c55474a3e0be][ref-fd7340e0dd28]
Manages Complexity¶
Local conditional tables may be much smaller than a full joint table when each node has few parents. A dense graph need not save storage, and a compact description does not guarantee easy posterior inference. QMR-DT's large connected network required investigation of approximate computation.[ref-c55474a3e0be][ref-3b8409ef7138]
Abstract Reasoning¶
First state the variables and their parents, then specify each conditional law. The model's joint law is \(P(x_1,\ldots,x_n)=\prod_i P(x_i\mid\mathrm{pa}(X_i))\). One can condition that joint on observed variables to query others. D-separation establishes only graph-implied independences; actual extra independences can arise from particular parameter values.[^ref-c55474a3e0be]
Knowledge Transfer¶
Medical diagnosis and engineering fault isolation reuse the same formal roles—variables, parent DAG, local laws, product joint, and conditional queries—without reusing one another's probabilities. Inference algorithms may change without changing the network identity. A bare directed diagram with no probability model is not a transferable instance.
[^ref-c55474a3e0be]: Carnegie Mellon University, “10-708 PGM, Lecture 2: Bayesian Networks”, “Factorization Theorem,” “I-maps,” “Local Markov assumptions,” and “D-separation criterion.” Definition, factorization, and parameter-specific independence boundary.
[^ref-fd7340e0dd28]: Dan Geiger, Tom S. Verma, and Judea Pearl, “d-Separation: From Theorems to Algorithms”, author-posted manuscript. Graphical independence criterion.
[^ref-3b8409ef7138]: Michael Shwe and Gregory F. Cooper, “An Empirical Analysis of Likelihood-Weighting Simulation on a Large, Multiply-Connected Belief Network”, Proceedings of the Sixth Conference on Uncertainty in Artificial Intelligence (UAI 1990), 498–508, §§2–3. QMR-DT disease/finding model; this is the conference paper, distinct from the authors’ 1991 journal article.
[^ref-13dc79b9bad9]: Ole J. Mengshoel et al., Efficient Probabilistic Diagnostics for Electrical Power Systems, NASA/TM-2008-214589 (2008). ADAPT fault-diagnosis model.
Neighborhood in Abstraction Space¶
Bayesian Network sits in a sparse region of the domain-specific corpus (76th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Formal Models & Logical Foundations (33 abstractions)
Nearest neighbors
- Bayesian Programming — 0.84
- Chow–Liu Tree — 0.83
- Exponentially Modified Gaussian Distribution — 0.83
- Gram Matrix — 0.83
- Equicontinuity — 0.83
Computed from structural-signature embeddings · 2026-10-08