FAIME: A Feature based Framework to Automatically Integrate and Improve Metaheuristics via Examples
Metaheuristics tackle hard computational problems by sampling a tiny subset of the possible solutions, using knowledge gained from previous solutions to generate new ones. Much human effort has been spent on tuning metaheuristics for different problems. Recent work has explored ways to achieve this automatically. A separate strand of work has looked at the use of features to predict the performance of a given metaheuristic on a problem: such features might include the variance in city-city distances in the Travelling Salesman Problem, or statistics on the number and type of clauses in Maximum Satisfiability.
This project is investigating the use of features to automatically design new metaheurstics for a class of problems. Features of problems and features of algorithms will be matched. The idea is that given a set of existing problem instances as training data, a metaheuristic will be tailor-made to outperform off-the-shelf algorithms on similar problem instances.
This ARCHIE-WeSt project forms part of a broader research project of the same title, funded by the EPSRC under grant number EP/N002849/1, led by PI Dr. John Woodward.
For more information about the project contact Dr Alexander Brownlee (email@example.com), Senior Research Assistant at School of Computing Science and Mathematics at the University of Stirling.
For a list of the research areas in which ARCHIE-WeSt users are active please click here.