Benchmarks are important for comparing performance of optimisation algorithms, but we can select instances that present our algorithm favourably, and dismiss those on which our algorithm under-performs. Also related are automated design of algorithms, which use problem instances (benchmarks) to train an algorithm: careful choice of instances is needed for the algorithm to generalise. We sweep parameter settings of differential evolution to applied to the BBOB benchmarks. Several benchmark functions are highly correlated. This may lead to the false conclusion that an algorithm performs well in general, when it performs poorly on a few key instances. These correlations vary with the number of evaluations.
CHRISTIE, L.A., BROWNLEE, A.E.I. and WOODWARD, J.R. 2018. Investigating benchmark correlations when comparing algorithms with parameter tuning. In Aguirre, H.E. (ed.) Proceedings of the 2018 Genetic and evolutionary computation conference companion (GECCO'18 companion), 15-19 July 2018, Kyoto, Japan. New York: Association for Computing Machinery [online], pages 209-210. Available from: https://doi.org/10.1145/3205651.3205747