Heuristic Learning Based on Genetic Programming

In this paper we present an approach to learning heuristics based on Genetic Programming (GP). Instead of directly solving the problem by application of GP, GP is used to develop a heuristic that is applied to the problem instance. By this, the typical large runtimes of evolutionary methods have to be invested only once in the learning phase. The resulting heuristic is very fast. The technique is applied to a field from the area of VLSI CAD, i.e. minimization of Binary Decision Diagrams (BDDs). We chose this topic due to its high practical relevance and since it matches the criteria where our algorithm works best, i.e. large problem instances where standard evolutionary techniques cannot be applied due to their large runtimes. Our experiments show that we obtain high quality results that outperform previous methods, while keeping the advantage of low runtimes.