Late-acceptance hill-climbing with a grammatical program representation
James McDermott, Miguel Nicolau · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
The late-acceptance hill-climbing (LAHC) metaheuristic is a stochastic hill-climbing algorithm with a simple history mechanism, proposed by Burke and Bykov in 2008, which seems to give a remarkable and reliable performance improvement relative to hill-climbing itself. LAHC is here used for the first time for genetic programming problems, with a grammatical encoding. A novel variant of LAHC with an initial random sampling is also proposed. Performance of both is competitive with full population-based search.