A Parametric Framework for Genetic Programming with Transfer Learning for Uncertain Capacitated Arc Routing Problem
Lists
Only you see this
Abstract
The Uncertain Capacited Arc Routing Problem (UCARP) is an important variant of arc routing problems that is capable of modelling uncertainties of real-world scenarios. Genetic Programming is utilised to evolve routing policies for vehicles to enable them to make real-time decisions and handle environment uncertainties. However, when the properties of a solved problem change, the trained routing policy becomes ineffective and a new routing policy is needed to be trained. The training process is time-consuming. Nevertheless, by extraction and transfer of some knowledge learned from the previous similar problem, the retraining process can be improved. Transfer learning is a challenging task that entails many aspects to decide about, which can influence the degree by which knowledge transfer can be effective. Consequently, in this paper we propose a parametric framework to formalise these details so that it can facilitate studying different aspects of using transfer learning for handling scenario changes of UCARP. Conducting a large number of experiments, we utilise this framework to analyse different transfer learning mechanisms and demonstrate how it can help with understanding dynamics of knowledge transfer for UCARP.
BibTeX
@inproceedings{ardeh2020parametric,
author = {Mazhar Ansari Ardeh and Yi Mei and Mengjie Zhang},
title = {{A Parametric Framework for Genetic Programming with Transfer Learning for Uncertain Capacitated Arc Routing Problem}},
booktitle = {Australasian Conference on Artificial Intelligence},
year = {2020},
doi = {10.1007/978-3-030-64984-5\_12},
}Loading related content...