Abstract:
First-order learning systems (e.g., FOlL, FOCL, FORTE) generally rely on hill-climbing heuristics in order to avoid the combinatorial explosion inherent in learning first-order concepts. However, hill-climbing leaves these systems vulnerable to local maxima and local plateaus. We present a method, called relational pathfinding, which has proven highly effective in escaping local maxima and crossing local plateaus. We present our algorithm and provide learning results in two domains: family relationships and qualitative model building.