Page 93 - 49A Field Guide to Genetic Programming

9.2 Keeping the Objectives Separate 79

round of comparisons with the rest of the population was used as a tie
breaker. The method successfully evolved queues, lists, and circular lists.
Langdon and Poli (1998b) used Pareto selection with two objectives,
fitness and speed, to improve the performance of GP on the Santa Fe Trail
Ant problem. Ross and Zhu (2004) used MO GP with different variants of
Pareto selection to evolve 2-D textures. The objectives were feature tests
that were used during fitness evaluation to rate how closely a candidate
texture matched visual characteristics of a target texture image. Dimopoulos
(2005) used MO GP to identify the Pareto set for a cell-formation problem
related to the design of a cellular manufacturing production system. The
objectives included the minimisation of total intercell part movement, and
the minimisation of within-cell load variation.
Rossi, Liberali, and Tettamanzi (2001) used MO GP in electronic design
automation to evolve VHDL code. The objectives used were the suitability
of the filter transfer function and the transition activity of digital blocks.
Cordon, Herrera-Viedma, and Luque (2002) used Pareto-dominance-based
GP to learn Boolean queries in information retrieval systems. They used two
objectives: precision (the ratio between the relevant documents retrieved in
response to a query and the total number of documents retrieved) and recall
(the ratio between the relevant documents retrieved and the total number
of documents relevant to the query in the database).
Barlow (2004) used a GP extension of the well-known NSGA-II MOO
algorithm (Deb, Agrawal, Pratap, and Meyarivan, 2000) for the evolution of
autonomous navigation controllers for unmanned aerial vehicles. Their task
was locating radar stations, and all work was done using simulators. Four
objectives were used: the normalised distance from the emitter, the circling
distance from the emitter, the stability of the flight, and the efficiency of
the flight.
Araujo (2006) used MO GP for the joint solution of the tasks of statistical
parsing and tagging of natural language. Their results suggest that solving
these tasks jointly led to better results than approaching them individually.
Han, Zhou, and Wang (2006) used a MO GP approach for the identi-
fication of chaotic systems where the objectives included chaotic invariants
obtained by chaotic time series analysis as well, as the complexity and per-
formance of the models.
Khan (2006) used MO GP to evolve digital watermarking programs. The
objectives were robustness in the decoding stage, and imperceptibility by the
human visual system. Khan and Mirza (2007) added a third objective aimed
at increasing the strength of the watermark in relation to attacks.
Kotanchek, Smits, and Vladislavleva (2006) compared different flavours
of Pareto-based GP systems in the symbolic regression of industrial data.
Weise and Geihs (2006) used MO GP to evolve protocols in sensor networks.
The goal was to identify one node on a network to act as a communication