Using Experimental Design to Find Effective Parameter Settings for Heuristics
Journal of Heuristics
A Racing Algorithm for Configuring Metaheuristics
GECCO '02 Proceedings of the Genetic and Evolutionary Computation Conference
Comparison of Multiobjective Evolutionary Algorithms: Empirical Results
Evolutionary Computation
A GA-based method to produce generalized hyper-heuristics for the 2D-regular cutting stock problem
Proceedings of the 8th annual conference on Genetic and evolutionary computation
Experimental Research in Evolutionary Computation: The New Experimentalism (Natural Computing Series)
Design and Analysis of Experiments
Design and Analysis of Experiments
Fine-Tuning of Algorithms Using Fractional Experimental Designs and Local Search
Operations Research
Multiobjective optimization problems with complicated Pareto sets, MOEA/D and NSGA-II
IEEE Transactions on Evolutionary Computation
ParamILS: an automatic algorithm configuration framework
Journal of Artificial Intelligence Research
A new dispatching rule based genetic algorithm for the multi-objective job shop problem
Journal of Heuristics
Automatic configuration of multi-objective ACO algorithms
ANTS'10 Proceedings of the 7th international conference on Swarm intelligence
Parameter tuning boosts performance of variation operators in multiobjective optimization
PPSN'10 Proceedings of the 11th international conference on Parallel problem solving from nature: Part I
A fast and elitist multiobjective genetic algorithm: NSGA-II
IEEE Transactions on Evolutionary Computation
Performance assessment of multiobjective optimizers: an analysis and review
IEEE Transactions on Evolutionary Computation
Hi-index | 0.00 |
A genetic algorithm heuristic that uses multiple rank indicators taken from a number of well established evolutionary algorithms including NSGA-II, IBEA and SPEA2 is developed. It is named Multi-Indicator GA (MIGA). At every iteration, MIGA uses one among the available indicators to select the individuals which will participate as parents in the next iteration. MIGA chooses the indicators according to predefined probabilities found through the analysis of mixture experiments. Mixture experiments are a particular type of experimental design suitable for the calibration of parameters that represent probabilities. Their main output is an explanatory model of algorithm performance as a function of its parameters. By finding the point that provides the maximum we also find good algorithm parameters. To the best of our knowledge, this is the first paper where mixture experiments are used for heuristic tuning. The design of mixture experiments approach allowed the authors to identify and exploit synergy between the different rank indicators. This is demonstrated by our experimental results in which the tuned MIGA compares favorably to other well established algorithms, an uncalibrated multi-indicator algorithm, and a multi-indicator algorithm calibrated using a more conventional approach.