Anti-correlation Measures in Genetic Programming R I (

We compare th ree diversity-preserving mechanisms, implicit fitness sharing, negative co rrelation learning, and a new fo rm, root-quartic negative co rrelation learning, on a standard genetic prog amming problem, the 6multiplexer. On this problem, root-quartic negative co rrelation learning significantly outpe rforms standard negative co rrelation learning, and marginally outpe rforms implicit fitness sharing. We analyse the diffe rence between standard and rootquartic negative co rrelation learning, and provide a partial explanation fo r the improved pe rformance.