Affordable Access

Publisher Website

Operator learning for a problem class in a distributed peer-to-peer environment

Authors
Publisher
Springer
Publication Date
Identifiers
DOI: doi: 10.1007/3-540-45712-7_17
Disciplines
  • Computer Science

Abstract

all.dvi Operator Learning for a Problem Class in a Distributed Peer-to-Peer Environment⋆ Ma´rk Jelasity1, Mike Preuß2, and A. E. Eiben1 1 Free University of Amsterdam, Amsterdam, The Netherlands [email protected], [email protected] 2 University of Dortmund, Dortmund, Germany [email protected] Abstract. This paper discusses a promising new research direction, the auto- matic learning of algorithm components for problem classes. We focus on the methodology of this research direction. As an illustration, a mutation operator for a special class of subset sum problem instances is learned. The most impor- tant methodological issue is the emphasis on the generalisability of the results. Not only a methodology but also a tool is proposed. This tool is called DRM (distributed resource machine), developed as part of the DREAM project, and is capable of running distributed experiments on the Internet making a huge amount of resources available to the researcher in a robust manner. It is argued that the DRM is ideally suited for algorithm learning. 1 Introduction This paper discusses a promising new research direction, the automatic learning of al- gorithm components for problem classes. The main contribution of this paper is three- fold. First, we emphasize the importance of an appropriate methodology that allows researchers to produce generalizable knowledge over a problem class rather than a prob- lem instance. Second, we propose a tool that might be ideally suitable for generating such knowledge automatically. Finally, both the methodology and our proposed tool are illustrated via an example: a search operator for a special class of subset sum problem instances is learned. In the recent years much research effort has been devoted to methods that try to im- prove heuristic search through some form of learning. To motivate our approach, let us elaborate on its relationship with these methods. The way different approaches generate knowledge can be categorized along at least two dimensi

There are no comments yet on this publication. Be the first to share your thoughts.