engleski

Experimental Research Challenges in Swarm Intelligence and Evolutionary Computation

Many practical optimization problems have high computational time complexity and cannot be solved by exact methods in reasonable time. A viable approach in these situations is to use metaheuristics from the field of swarm intelligence and evolutionary computation. Although experimental research in this field is very common there are many possible pitfalls and open questions regarding appropriate methodology. Recently this has led to the increased interest of the scientific community about this subject. The source of problems lies in stochastic nature of such algorithms, the inability to guarantee an optimal solution in polynomial time, and sometimes very different levels of success for problem instances of equal size. One of the first choices a researcher (and practitioner) needs to address is how to choose stopping criteria for the algorithm. Possible approaches and practical recommendations regarding this issue will be presented and discussed. One of the basic questions, that often is not addressed at all, concerns the method of summarizing the entire distribution of solutions into a single value. Some deficiency of usual choices for measures or indicators of performance will be discussed and practical recommendations will be given. It is usually taught that smaller variability in solution quality between repeated executions of the stochastic optimization algorithm is a good property of such algorithm. This common misconception will be discussed and cases when the opposite is true will be presented. Finally, to use more rigors statistical tools appropriate statistical methods will be considered.

experimental methodology ; swarm intelligence ; evolutionary computation ; performance ; quantiles

nije evidentirano