| Bookmark: |
Full-text
| URL: http://www.csl.sony.fr/Language/Research...apers/web-ecml.ps.gz |
| Cached: PDF-193K PS-418K PS.gz-86K |
| SAVE AS an easy-to-recall long filename: |
| Filename format: author--year--title PDF-193K PS-86K :: About GZip'd PS |
| Filename format: author--year--title--journal|proceedings|...--pages PDF-193K PS-86K |
Related links
| Source: http://www.csl.sony.fr/General/Publicati...ference=steels%3A97b |
| CiteSeer: http://citeseer.ist.psu.edu/steels97constructing.html |
| Web search: Google Web Search :: Google Scholar |
| Within this site: Cited by (29) References (7) |
Paper at a Glance
BibTexconstruct 1 Introduction Constructing and Sharing Perceptual Distinctions VUB Artificial Intelligence Laboratory Pleinlaan 2 1050 Brussels email: steels@arti.vub.ac.beand Sony Computer Science Laboratory 6 Rue Amyot F75005 Paris, France The paper describes a mechanism whereby agents generate perceptual distinctions through a series of adaptive discrimination games and share these distinctions through adaptive language games. Results from computer simulations as well as experiments on robotic agents are presented. Luc Steels Machine learning research on concept formation has mostly concentrated on inducing definitions of concepts from examples (see e.g. [3] It is assumed that the necessary conceptualisations have been performed by a teacher and that the learning system gets positive and negative examples of the concepts to be learned. There are however many situations where examples are not available. Indeed, it could be argued that as long as the examples are supplied by a teacher, the intelligence resides in the teacher and does not originate in the learner. This paper focuses on how an autonomous agent might be able to appropriate conceptualisations himself. It investigates how an agent may con struct new distinctions relevant for discriminating objects in its environment. The construction process is based on the random generation of a variety of distinctions which then undergo selectionist pressure based on how well they discriminate. The system is open in the sense that new objects may always enter into the environment forcing the agent to refine or expand the available repertoire of distinctions. But if every agent creates his own distinctions, we are confronted with the problem how there could ever arise coherence in a group of agents. The second part of the paper addresses this question. It is proposed that language plays a critical role in establishing a shared set of perceptual distinctions. A mechanism is therefore proposed whereby distinctions are ...
@inproceedings{steels97constructingAnd,
author={L. Steels},
title={Constructing and sharing perceptual distinctions},
year={1997},
address={Berlin},
editor={M. van Someren and G. Widmer},
publisher={Springer-Verlag},
booktitle={Proceedings of the European Conference on Machine Learning},
url={http://www.isrl.uiuc.edu/~amag/langev/paper/steels97constructingAnd.html},
keywords={perception, semantics, evolutionary linguistics, language games, categorization}
}
| HOME :: Conference List :: Conference Paper | Comments to: junwang4 you-know-at gmail.com | Last update: 2/2/08 |