The lecture on "Evolution Strategies" offers an in-depth exploration of optimization techniques that are pivotal in solving complex problems. It begins by introducing basic optimization concepts, setting the stage for more advanced strategies. The lecture delves into the (1+1)-ES, a simple evolution strategy that evolves solutions using one parent and one offspring per generation, illustrating the foundational mechanism of this approach. It further discusses the 1/5 success rule, a method for adapting the step size based on a target success rate, which helps maintain efficient progress. The concept of restarts is explored, emphasizing strategies to escape local optima and improve solution diversity. More complex is the (μ+λ)-ES, which involves multiple parents and offspring, enhancing the robustness and convergence rate of the strategy. Self-adaptation is highlighted as a crucial feature, allowing the algorithm to dynamically adjust its parameters to better suit the problem landscape. The lecture also covers the adaptation of the covariance matrix, a sophisticated technique that helps the algorithm learn and adapt to the shape of the optimization landscape. Experimental results are presented to showcase the practical applications and effectiveness of these strategies. Finally, benchmark functions described in the appendix serve as a standard for evaluating and comparing the performance of evolution strategies.

In practical exercises, participants are introduced to Python and all algorithms are programmed to facilitate hands-on learning and application.

The course is worth 6 ECTS.