Bounded Fitting for Expressive Description Logics
Bounded fitting is a paradigm for learning logical formulas from labeled data, offering PAC-style guarantees and often implemented via SAT solvers. It has been successfully applied to ALC description logic concepts. This work extends bounded fitting to more expressive description logics with inverse roles, qualified number restrictions, and feature comparisons. The authors investigate conditions preserving favorable theoretical properties and implement the approach using a SAT solver. Their tool is compared against state-of-the-art concept learners, showing encouraging results and demonstrating practical applicability for expressive concept learning.
Key facts
- Bounded fitting offers PAC-style generalization guarantees.
- It has been applied to learning concepts in ALC description logic.
- This work extends bounded fitting to expressive description logics.
- Extensions include inverse roles, qualified number restrictions, and feature comparisons.
- Implementation uses a SAT solver.
- Tool compared with state-of-the-art concept learners.
- Results are encouraging.
- Demonstrates practical approach to expressive concept learning.
Entities
—