Benchmarking Parameter-free AMaLGaM on Functions With and Without Noise

Estimation-of-distribution algorithms (EDAs) are optimization algorithms at the frontier of genetic- and evolutionary computation (GEC) research. Characteristic of EDAs is the iteration of selecting promising solutions, estimating a probability distribution from the selected solutions and subsequently generating new solutions by drawing samples from the estimated distribution. Probability distributions provide a principled way of modelling dependencies between problem variables. Contrary to classic GEC methods, this allows EDAs to successfully and automatically identify and exploit problem structures with respect to dependencies between problem variables. EDAs are therefore able to solve a much larger class of problems efficiently without requiring prior knowledge. In this article, we present an EDA for real-valued optimization, benchmark it within the BBOB framework and discuss our findings. The EDA is known as the AdaptedMaximum-Likelihood Gaussian Model Iterated Density-Estimation Evolutionary Algorithm (AMaLGaM-IDEA, or AMaLGaM for short). The BBOB (Black-Box Optimization Benchmarking) framework for real-valued optimization, introduced in 2009, consists of both an experimental setup and an extensive, but carefully chosen, set of functions to be optimized.