SOAR: an architecture for general intelligence
Artificial Intelligence
A Heuristic Approach to the Discovery of Macro-Operators
Machine Learning
Learning subgoal sequences for planning
IJCAI'89 Proceedings of the 11th international joint conference on Artificial intelligence - Volume 1
Learning teleoreactive logic programs from problem solving
ILP'05 Proceedings of the 15th international conference on Inductive Logic Programming
Acquisition of hierarchical reactive skills in a unified cognitive architecture
Cognitive Systems Research
| Hi-index | 0.00 |
Decomposing a difficult problem into simpler subproblems is a classic problem solving technique. Unfortunately, the most difficult subproblems can be as difficult, if not more difficult, than the original problem. This is not an obstacle to problem solving if the difficult subproblems recur in other problems. If the difficult subproblems recur often, then its solution need only be learned once and reused. SteppingStone is a learning problem solver that decomposes a problem into simple and difficult-but-recurring subproblems. It solves the simple subproblems with an inexpensive constrained problem solver. To solve the difficult subproblems, SteppingStone uses an unconstrained problem solver. Once it solves a difficult subproblem, it uses the solution to generate a sequence of subgoals, or steppingstones, that can be used by the constrained problem solver to solve this difficult subproblem when it occurs again. In this paper we provide analytical evidence for SteppingStone's capabilities as well as empirical results from our work with the domain of logic synthesis.