Sequential Rule Mining for Automated Design of Meta-heuristics

Abstract

With a recently defined AutoGCOP framework, the design of local search algorithms can be defined as the composition of the basic elementary algorithmic components. These compositions into the best algorithms thus retain useful knowledge of effective algorithm design. This paper investigates effective algorithmic compositions with sequential rule mining techniques to discover valuable knowledge in algorithm design. With the collected effective algorithmic compositions, sequential rules of basic algorithmic components are extracted and further analysed to automatically compose basic algorithmic components within the general AutoGCOP framework to develop new effective meta-heuristics. The sequential rules present superior performance in composing the basic algorithmic components for solving the benchmark vehicle routing problems with time window constraints, demonstrating its effectiveness in designing new algorithms automatically.

Download paper here

Recommended citation:

@inproceedings{meng2023sequential, title={Sequential Rule Mining for Automated Design of Meta-heuristics}, author={Meng, Weiyao and Qu, Rong}, booktitle={Proceedings of the Companion Conference on Genetic and Evolutionary Computation}, pages={1727–1735}, year={2023} }