Probabilistic models like Bayesian Networks due their success to the possibility of performing inferences in large domains. Anyhow, to achieve such computational benefit, the model designer (or the decision maker) is forced to introduce stochastici ndependencec onditions not always required by the problem in hand. But this way is obliged only if the "false myths" of uniqueness of the infered values and of a fully detailed model are followed. With the more general models of probability partial assessment(sc onditionalo r not), unnaturalc onstraintsc an be avoided and evaluation on only relevant events (macro situations) is required. In this framework inference is reduced to linear programming problems and a skilful use of the "measurefree" logical conditions can help on complexity reduction.
Utilizzo delle componenti logiche nei processi inferenziali basati su assegnazioni di probabilità parziali
CAPOTORTI, Andrea
2001
Abstract
Probabilistic models like Bayesian Networks due their success to the possibility of performing inferences in large domains. Anyhow, to achieve such computational benefit, the model designer (or the decision maker) is forced to introduce stochastici ndependencec onditions not always required by the problem in hand. But this way is obliged only if the "false myths" of uniqueness of the infered values and of a fully detailed model are followed. With the more general models of probability partial assessment(sc onditionalo r not), unnaturalc onstraintsc an be avoided and evaluation on only relevant events (macro situations) is required. In this framework inference is reduced to linear programming problems and a skilful use of the "measurefree" logical conditions can help on complexity reduction.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
