The design and configuration of optimization algorithms for computationally hard problems is a time-consuming and difficult task. This is mainly This is in large part due to a number of aggravating circumstances such as the NP-hardness of most of the problems to be solved, the difficulty of algorithm analysis due to stochasticity and heuristic biases, and the large number of degrees of freedom in defining and selecting algorithmic components and settings of numerical parameters. Over the recent years, the development of automatic methods to search large configuration spaces has received significant attention as a possible solution to these challenges. Such automatic algorithm configuration methods have by now proved to be instrumental for developing high-performance algorithms. The presentation will discuss how automatic algorithm configuration tools can be used to develop high-performing evolutionary and other optimization algorithms. After an overview of available tools, I will highlight various successful applications of these such as the automatic configuration of multi-objective optimizers, and the automatic configuration of hybrid stochastic local search algorithms. Finally, I will highlight the impact automatic algorithm configuration has and will have on the algorithm design and development process.
展开▼
