Transfer Learning for Reinforcement Learning Domains: A Survey

By Matthew E. Taylor et al

Published on July 10, 2009

Read the original document by opening this link in a new tab.

Table of Contents

1. Transfer Learning Objectives
2. Evaluating Transfer Learning Methods
2.1 Empirical Transfer Comparisons
2.2 Dimensions of Comparison

Summary

The document discusses the concept of transfer learning in reinforcement learning domains, focusing on how experience gained in one task can improve learning in a related task. It presents a framework classifying transfer learning methods and surveys existing literature. The paper outlines evaluation metrics such as Jumpstart, Asymptotic Performance, Total Reward, Transfer Ratio, and Time to Threshold. It also categorizes transfer learning algorithms based on dimensions like task difference assumptions, source task selection, task mappings, and transferred knowledge.