A Selection Hyper-Heuristic for Transfer Learning in Genetic Programming
Jeffrey Russell, Nelishia Pillay · 2023
Transfer learning has recently gained popularity in the evolutionary algorithms space. Determining a good criteria for how to transfer knowledge can be very challenging. It has been a focus of research to optimize the process of deciding what, when and how to transfer knowledge. Genetic programming is an area of evolutionary algorithms that has benefited greatly from transfer learning. In this paper, a selection hyper-heuristic (SHeTL) has been used to optimize transfer learning, i.e. what to transfer, when to transfer and how to transfer, in genetic programming. The proposed approach has been applied to benchmark and real-world symbolic regression problems. SHeTL produced significant improvements over canonical genetic programming. Furthermore, the hyper-heuristic produced better results than previous transfer learning approaches used in genetic programming.