We show that the adoption of a three-valued setting for inductive concept learning is particularly useful for learning in single and multiple agent systems. Distinguishing between what is true, what is false and what is unknown can be useful in situations where decisions have to be taken on the basis of scarce infor- mation. Such situation occurs, for example, when an agent incrementally gathers information from the sur- rounding world and has to select its own actions on the basis of such acquired knowledge. In a three-valued setting, we learn a denition for both the target concept and its opposite, consider- ing positive and negative examples as instances of two disjoint classes. To this purpose, we adopt Extended Logic Programs (ELP) under a Well-Founded Seman- tics with explicit negation (WFSX) as the representa- tion formalism for learning. Standard Inductive Logic Programming techniques are then employed to learn the concept and its opposite. The learnt denitions of the positive and negative concepts may overlap, both when learning con icting rules for a predicate and its explicit negation by a sin- gle agent or when combining the knowledge learned by multiple agents. In the paper, we handle the issue of strategic combination of possibly contradictory learnt denitions.
Agents learning in a three-valued logical setting
LAMMA, Evelina;RIGUZZI, Fabrizio;
1999
Abstract
We show that the adoption of a three-valued setting for inductive concept learning is particularly useful for learning in single and multiple agent systems. Distinguishing between what is true, what is false and what is unknown can be useful in situations where decisions have to be taken on the basis of scarce infor- mation. Such situation occurs, for example, when an agent incrementally gathers information from the sur- rounding world and has to select its own actions on the basis of such acquired knowledge. In a three-valued setting, we learn a denition for both the target concept and its opposite, consider- ing positive and negative examples as instances of two disjoint classes. To this purpose, we adopt Extended Logic Programs (ELP) under a Well-Founded Seman- tics with explicit negation (WFSX) as the representa- tion formalism for learning. Standard Inductive Logic Programming techniques are then employed to learn the concept and its opposite. The learnt denitions of the positive and negative concepts may overlap, both when learning con icting rules for a predicate and its explicit negation by a sin- gle agent or when combining the knowledge learned by multiple agents. In the paper, we handle the issue of strategic combination of possibly contradictory learnt denitions.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
