Algorithm selection using transfer learning

Per-instance algorithm selection has been shown to achieve state of the art performance in solving Travelling Salesman Problems (TSP). By selecting optimization algorithms for each TSP instance, significant time savings have been achieved. In this work, we highlight how recent algorithm selection techniques apply to service composition; which is posed as a TSP problem. However, the service composition environment is highly dynamic, which poses unique challenges for algorithm selection. Chief amongst those is the availability of training data for all algorithms on unseen tasks, which is infeasible to obtain. To address this problem, we propose the use of transfer learning techniques to improve classification accuracy in dynamic settings such as service composition.