Definition
A method that represents dependencies between variables using a directed acyclic graph and conditional probabilities, and updates beliefs after observing evidence.
Research by Pearl and others combined probabilistic knowledge with graph structures, using conditional independence to factor complex distributions and studying efficient local message passing.
Results for tree-structured or singly connected networks cannot be generalized directly into a claim that efficient exact inference is possible for arbitrary graphs. An ordinary Bayesian network represents probabilistic dependencies; assigning it causal meaning requires additional assumptions.
Principles and impact
Problems solved
- Organizes knowledge and supports inference when information is incomplete and uncertain.
Principles
- Factors a joint distribution using conditional probabilities, then computes posterior probabilities from evidence.
Improvements
Not recorded
Significance
- Provides systematic tools for reasoning under uncertainty in AI.
History and applications
People
Judea Pearl
Organizations
Not recorded
Events
Not recorded
Applications
Reasoning under uncertainty、Diagnostic support
Limitations
- Network structures and probabilities need supporting evidence, and inference in complex networks may be expensive; correlation does not automatically imply causation.
Upstream and downstream relations
Upstream
No direct upstream relations recorded