XCS's Strength-Based Twin. Part II

Tim Kovacs

Research output: Chapter in Book/Report/Conference proceedingChapter in a book

Abstract

This sequel continues the comparison of the twins XCS and SB--XCS. We find they tend towards different representations of the solution and distinguish three types of representations which rule populations can form, namely complete maps, partial maps, and default hierarchies. Following this we evaluate the respective advantages and disadvantages of complete and partial maps at some length. We conclude that complete maps are likely to be superior for sequential tasks. For non-sequential tasks, partial maps have the advantage of parsimony whereas complete maps can take advantage of subsumption deletion. It is unclear which is more significant. We also conclude that partial maps are likely to be suitable for Pittsburgh classifier systems and supervised learning systems.
Translated title of the contributionXCS's Strength-Based Twin. Part II
Original languageEnglish
Title of host publicationLearning Classifier Systems
PublisherSpringer
VolumeLNCS 2661
ISBN (Print)3540205446
Publication statusPublished - 2003

Bibliographical note

Other page information: 81-98
Other identifier: 2000114

Fingerprint Dive into the research topics of 'XCS's Strength-Based Twin. Part II'. Together they form a unique fingerprint.

Cite this