Exploration of configural representation in landmark learning using working memory toolkit

Wang, Xiaochun, Tugcu, Mert, Hunter, Jonathan E. and Wilkes, D. Mitch (2009) Exploration of configural representation in landmark learning using working memory toolkit. Pattern Recognition Letters, 30 1: 66-79. doi:10.1016/j.patrec.2008.09.002


Author Wang, Xiaochun
Tugcu, Mert
Hunter, Jonathan E.
Wilkes, D. Mitch
Title Exploration of configural representation in landmark learning using working memory toolkit
Journal name Pattern Recognition Letters   Check publisher's open access policy
ISSN 0167-8655
1872-7344
Publication date 2009-01-01
Sub-type Article (original research)
DOI 10.1016/j.patrec.2008.09.002
Volume 30
Issue 1
Start page 66
End page 79
Total pages 14
Place of publication Amsterdam, Netherlands
Publisher Elsevier
Language eng
Formatted abstract
The mechanisms by which humans and animals use visually-acquired landmarks to find their way around have proved fascinating. Considerable evidence suggests that animals navigate not only on the basis of the overall geometry of the space but also on the basis of a configural representation of the cues. In contrast to earlier linear models of elemental feature representation, configural representation requires individual stimulus be represented in the context of other stimuli and is typified by non-linear learning tasks such as the transverse patterning problem.

This paper explores the suitability of configural representation for automatic scene recognition in robot navigation by conducting experiments designed to infer semantic prediction of a scene from different configurations of its stimuli. The main contribution of this work is that it provides a methodology for automatic landmark-based scene identification with the aid of a reinforcement learning based software package, called the working memory toolkit (WMtk), which allows reward associations between a target location and the conjunctive representations of its stimuli. Experimental results obtained with two different target locations are presented and compared with those of two other classification mechanisms, a support vector machine approach and a simple linear two-class classifier, perceptron.
Keyword Visual landmarks
Scene recognition
Configural representation
Feature representation
Q-Index Code C1
Q-Index Status Provisional Code
Institutional Status Non-UQ

Document type: Journal Article
Sub-type: Article (original research)
Collection: School of Information Technology and Electrical Engineering Publications
 
Versions
Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 2 times in Thomson Reuters Web of Science Article | Citations
Scopus Citation Count Cited 4 times in Scopus Article | Citations
Google Scholar Search Google Scholar
Created: Fri, 10 May 2013, 15:32:37 EST by System User on behalf of Scholarly Communication and Digitisation Service