Paper Info
Title
The Role of Action in Object Categorization
Abstract
The hypothesis underlying the present work is that knowledge organization in the mind/brain reflects the actions with which an organism responds to the input and not the perceptual properties of the input and it can be flexibly adapted to new goals. We present the results of two simulations using artificial neural networks in which an organism must generate various movements of its 2-segment arm in response to visually perceived objects. In the first simulation we find that the internal representations of the organism's neural network are organized in terms of macro- actions, that is, sequences of movements (micro-actions) that allow the organism to correctly respond to the objects. In the second simulation the organism's neural network is able to flexibly organize itself in order to adapt to different tasks. The network's internal representations of objects reflect the current task and not the perceptual similarities among the objects. In absence of task information, however, perceptual similarity is the best predictor of categorization. An account of categorization in terms of action makes it possible to explain the so-called 'shape-bias' in object categorization. Shape is a better predictor of categorization than other object properties such as color and size because we tend to respond in the same way to objects with the same shape rather than to objects with the same color or size.
Year
Venue
Keywords
2002
FLAIRS Conference
object categorization,artificial neural network,neural network
Field
DocType
ISBN
Categorization,Computer science,Property (programming),Artificial intelligence,Artificial neural network,Perception,Knowledge organization,Machine learning,Perceptual similarity,Organism
Conference
1-57735-141-X
Citations 
PageRank 
References 
1
0.98
1
Authors
3
Name
Order
Citations
PageRank
Andrea Di Ferdinando1167.83
Anna M. Borghi2325.95
Domenico Parisi3745101.62