abstract = "Backward chaining evolutionary algorithms (BC-EA)
offer the prospect of runtime efficiency savings by
reducing the number of fitness evaluations without
significantly changing the course of genetic algorithm
(GA) or genetic programming (GP) runs.
poli05:_tourn_selec_iterat_coupon_probl described how
BC-EA does this by avoiding the generation and
evaluation of individuals which never appear in
selection tournaments. (Poli,2005) suggested the
largest savings occur in very large populations, short
runs and small tournament sizes. It gave some evidence
of the actual savings in fixed-length binary GAs. Here
we provide a generational GP implementation of BC-EA
and empirically investigate its efficiency, in terms of
both fitness evaluations and effectiveness, with
mutation and two offspring crossover.",