Learning concept boundaries from observational data

Learn the convex region boundaries that define abstract concepts within conceptual spaces directly from observational data, replacing heuristic boundary specification with data-driven estimation methods that support temporal, trajectory-based recognition.

Background

In the current implementation, concept regions for chess strategies are defined heuristically using expert knowledge. This approach provides initial feasibility but lacks systematic, scalable boundary estimation.

The authors explicitly state that learning concept boundaries from observational data remains an open research question, crucial for refining region shapes and centroids in multi-domain conceptual spaces and improving recognition reliability.