Though not implemented, one idea is to have the algorithm periodically retry "finished" syntax trees, throwing away important data and trying again. This is similar to randomized restarts for hill-climbing algorithms.

Return to my homepage