Reporting on experiments in evolutionary computation

Mike Preuß · Technische Universität Dortmund Eldorado (Technische Universität Dortmund) · 2007

Additionally, the impact of nondeterminism on the outcome of EC algorithms may have been underestimated in the past. The result of a comparison between two deterministic computer programs is sufficiently described by reporting its output values. As soon as any stochastic process is involved, much more data has to be provided and taken into account, including at least the recorded performance value samples or suitable statistics derived from them. For the sake of clarity and reproducibility, it is therefore the more important to properly report an experiments composition and outcome, the more it depends on subjective choices (e.g. of performance criteria) and the less exact its result are.

Read the paper · More papers on PaperTik