arXiv Board & Classic Game AI· Stefan Edelkamp·· 25 d agoAI score32
Inner-Learning: Turning hand-coded Skat belief reasoning into declarative knowledge
Inner-Learning: From Hand-Coded Beliefs to Declarative Knowledge in Trick-Taking Games
AI summary
Inner-Learning automatically extracts the hand-coded belief reasoning rules of a top-level Skat player into a declarative DSL, with the core primitive ¬holds(P,C). Round-trip code generation verification shows that the extracted rules reproduce the intended behavior of the original engine and fix 3 bugs found in the original engine through auditing; the rule base is expanded through symmetry derivation and annotation from expert strategy documents, and adds at-most-k / at-least-k cardinality constraints and confidence parameters.
Source: arXiv Board & Classic Game AI · arxiv.org