Surrogate-Assisted Multitask Genetic Programming for Learning Scheduling Heuristics

Abstract

This chapter shows how to use surrogate techniques to improve the transfer effectiveness between tasks in multitask learning. This chapter introduces how surrogates are built and used to help share knowledge between tasks. The results show that the proposed surrogate-assisted multitask genetic programming can learn effective scheduling heuristics for all tasks simultaneously. In addition, the results show that the proposed algorithm can improve the diversity of individuals for tasks and the proportions of individuals from different tasks are different. Last, this chapter studies the sizes of the learned scheduling heuristics and finds that the proposed algorithm has a bigger effect on the learned routing rule than the sequencing rule. The analyses of the structures of the learned rules show that the learned scheduling heuristics for different tasks share common knowledge.