Learning Qualitative Constraint Networks

Temporal and spatial reasoning is a fundamental task in artificial intelligence and its related areas including scheduling, planning and Geographic Information Systems (GIS). In these applications, we often deal with incomplete and qualitative information. In this regard, the symbolic representation of time and space using Qualitative Constraint Networks (QCNs) is therefore substantial. We propose a new algorithm for learning a QCN from a non expert. The learning process includes different cases where querying the user is an essential task. Here, membership queries are asked in order to elicit temporal or spatial relationships between pairs of temporal or spatial entities. During this acquisition process, constraint propagation through Path Consistency (PC) is performed in order to reduce the number of membership queries needed to reach the target QCN. We use the learning theory machinery to prove some limits on learning path consistent QCNs from queries. The time performances of our algorithm have been experimentally evaluated using different scenarios.