Automated parameter selection of scheduling algorithms using machine learning techniques

The work describes the effort to automatically select scheduling algorithms and generate corresponding parameters for new problem instances based on the results obtained for similar problem instances that have been extensively investigated. The effort tries to vastly reduce the development cycle of optimization algorithms as parameter tuning is usually more time consuming that implementing the algorithm or model. We investigated various heuristic methods for hyper-parameter selection and evaluated different machine learning methods. The results are very promising as selecting the top 5% combination of algorithms and parameters manages to consistently achieve results that are in the top 10% of the generated solutions, if full parameter and algorithm execution is performed.