Competition on Testing Evolutionary Algorithms on Real-world
Numerical Optimization Problems
Over the past few decades several efficient Evolutionary Computing (EC) algorithms have been devised for handling bound-constrained single-objective, multi-objective, non-linearly constrained, dynamic, large scale, and multimodal optimization problems. Performances of many of these algorithms have mainly been demonstrated in the corresponding journal articles on benchmark function sets with previously known optima and bounds on each decision variable. However, outstanding performances on a set of benchmarks may not guarantee a similar performance on every practical optimization problem too as it is well-nigh impossible to model all the complexities of the real world problems by a limited set of benchmarks. Very often due to the lack of communications among researchers from EC and other domains of science and engineering like power systems, computational electromagnetics, computational chemistry, signal processing, communication engineering and so on, the non-EC researchers try classical techniques on hard optimization problems and continue with poor results, which could have been substantially improved by applying an EC algorithm. Similarly on the other side, EC researchers miss the opportunity to test the strength of their algorithms on challenging real world problems and have to remain satisfied with results on benchmarks. Often results on these benchmarks become so close for different EC techniques, that it becomes really tough to appreciate the distinguishing features or merits of a newly proposed technique.
Taking into consideration all such facts, this competition aims at bridging the gap between EC algorithms and real-world problems from diverse domains, which are often way more relentless than the model benchmarks. Real-world problems are characterized by the fact that the objective function is derived from an engineering problem or a practical problem from other domain. We will also include fitness function approximation, if the real-world objective function is too time consuming to evaluate. Researchers from academia and industries are invited to submit unpublished manuscripts by applying one or more of the state-of-the-art nature-inspired optimizers to solve the selected 15 – 20 well-documented and tough problems. In particular, we encourage the researchers to submit their problems belonging to (but definitely not limited to) one of more of the following categories and sub-categories:
1) Real-parameter optimization problems (based on nature of the objective function):
Constrained (linear and non-linear) single and multi-objective
Dynamic, noisy, and uncertain objectives
Expensive and Large Scale (Typically involving 500+ decision variables)
Multi-modal (with an objective of locating all global and local objectives)
2) Real-parameter optimization problems (based on problem domains):
Bioinformatics and Computational biology
Power Systems Engineering
Sensors and Networks
Pattern recognition and Data Mining
The authors are encouraged to highlight the main features of the problems such as uni/multi-modality, ill-conditioned, epistasis, scalability, etc. We discourage submission of problems that require usage of commercial software or require excessive computation time to perform each function evaluation. As an example, please see the documentations on molecular clustering problems available from the Cambridge Cluster Database: http://www-wales.ch.cam.ac.uk/CCD.html.
We hope to have the codes of the test functions available by late November 2010 from http://www.ntu.edu.sg/home/EPNSugan.
Dr. Swagatam Das
Dr. P. N. Suganthan