The relations between ProbLog and Logic Programs with Annotated Disjunctions imply that Boolean Bayesian networks can be represented as ground ProbLog programs and acyclic ground ProbLog programs can be represented as Boolean Bayesian networks. This provides a way of learning ground acyclic ProbLog programs from interpretations: first the interpretations are represented in tabular form, then a Bayesian network learning algorithm is applied and the learned network is translated into a ground ProbLog program. The program is then further analyzed in order to identify noisy or relations in it. The paper proposes an algorithm for such identification and presents an experimental analysis of its computational complexity.
Learning ground problog programs from interpretations
RIGUZZI, Fabrizio
2007
Abstract
The relations between ProbLog and Logic Programs with Annotated Disjunctions imply that Boolean Bayesian networks can be represented as ground ProbLog programs and acyclic ground ProbLog programs can be represented as Boolean Bayesian networks. This provides a way of learning ground acyclic ProbLog programs from interpretations: first the interpretations are represented in tabular form, then a Bayesian network learning algorithm is applied and the learned network is translated into a ground ProbLog program. The program is then further analyzed in order to identify noisy or relations in it. The paper proposes an algorithm for such identification and presents an experimental analysis of its computational complexity.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.