Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Bayesian Network
Probabilistic graphical model representing conditional independence relationships between random variables through a directed acyclic graph. It allows for the calculation of complex conditional probabilities by factorizing the joint distribution.
Conditional Probability Table (CPT)
Matrix associated with each node in a Bayesian network specifying the probability of each possible state given the combinations of states of its parents. The CPT quantifies the local probabilistic relationships between connected variables.
Latent Variable
Variable not directly observed in a belief network but influencing the observed variables. It allows for modeling hidden factors or abstractions within the probabilistic system.
Hybrid Network
Bayesian network combining both discrete and continuous variables in its structure. It requires specialized methods to represent and compute the mixed conditional distributions.
Conditional Gaussian Distribution
Probability distribution for continuous variables in hybrid networks, conditioned by discrete variables. It allows for modeling Gaussian linear relationships between continuous and discrete variables.
Variable Elimination
Exact inference algorithm that sequentially eliminates variables from the network to compute marginal probabilities. The elimination order significantly influences the computational complexity.
Dynamic Bayesian Network
Temporal extension of Bayesian networks modeling time series with interconnected time slices. It captures temporal dependencies between variables at different time instants.
Canonical Parameterization
Alternative mathematical representation of probability distributions in Gaussian networks using canonical potentials. It facilitates algebraic operations in exact inference.
Potential Factor
Unnormalized function associated with cliques of a belief network, representing local interactions between variables. Factors combine multiplicatively to form the joint distribution.
Parameter Learning
Process of estimating conditional probability tables from observed data. It uses maximum likelihood or Bayesian methods to calibrate the network.
Moral Separator
Set of variables blocking all paths between two sets of variables in a moralized graph. It formally characterizes conditional independence in Bayesian networks.