Benchmark problems for evaluating non-linear optimization techniques. Most of these problems are multimodal and non-separable and many are non-differentiable. Since these are intended to evaluate optimization techniques that would apply to noisy functions and measurements, each problem includes a noisy version (where repeated evaluation of f(x*) produces different results).
The problems are taken from [1], which in turn modified the benchmark problems used by [2]. I have scaled them so that all problems are attempting to minimize an objective function defined over [-1, 1]^d; in most but not all cases the minimum value of f(x*) = 0.
[1] Ortiz-Boyer, Domingo, César Hervás-Martınez, and Nicolás Garcıa-Pedrajas. "Cixl2: A crossover operator for evolutionary algorithms based on population features." Journal of Artificial Intelligence Research 24.1 (2005): 1-48.
[2] Eiben, Agoston E., and Thomas Bäck. "Empirical investigation of multiparent recombination operators in evolution strategies." Evolutionary Computation 5.3 (1997): 347-365.