Treffer: Využití strojového učení pro extrakci doménové znalosti plánovacích problémů

Domain Control Knowledge (DCK) could significantly improve planning efficiency of general planners. Automatic creation of such knowledge with computers could be very helpful. In this thesis, two machine learning techniques were tried for this task, Inductive Logic Programming and Automata Learning. The second technique was more successful with quickly finding an automaton, which was used as Transition-Based Domain Con- trol Knowledge (T-DCK) to improve planning times of a general planner on problems from planning competitions. Learned DCK was also helpful in creating plans for larger problems, where the planner on the original problems failed due to resource constraints. Performance of the learned DCK was compared to a human-designed DCK. The learned DCK usually led to shorter plans and sometimes the planning was even faster.