AMH: a new Framework to Design Adaptive Metaheuristics

Metaheuristics should be configured to perform well on a given problem. Their configuration is either made off-line by automatic algorithm configuration tools or on-line with control mechanisms to adapt their behaviour. The former requires a flexible structure that may be modified during the execution. Therefore, the implementation of such a structure is not straightforward to enable modifications of optimisation strategies and not of parameter values only. In this work, we present AMH, a framework dedicated to the design of configurable metaheuristics. This framework is based on controlling the execution flow of metaheuristics to enable the implementation of flexible structures.