Artificial neural networks as models of stimulus control*

Provided by Cognitive Sciences ePrint Archive

Artificial neural networks as models of
stimulus control*

Stefano Ghirlanda and Magnus Enquist

Division of Ethology, Department of Zoology, University of Stockholm

Reprint of December 2, 2006

Abstract

We evaluate the ability of artificial neural network models (multi-layer
perceptrons) to predict stimulus-response relationships. A variety of empir-
ical results are considered, such as generalization, peak-shift (supernormal-
ity) and stimulus intensity effects. The networks were trained on the same
tasks as the animals in the considered experiments. The subsequent general-
ization tests on the networks showed that the model replicates correctly the
empirical results. It is concluded that these models are valuable tools in the
study of animal behaviour.

1 Introduction

Artificial neural networks represent an important advance in the modelling of ner-
vous systems and behaviour (see e.g. Churchland & Sejnowski, 1992). During
the last 10 or 15 years artificial neural networks have been actively researched
in as diverse disciplines as cognitive psychology, neurophysiology, engineering,
artificial intelligence and physics. For some reason, however, these models have
been more or less ignored by scientists working within the field of animal be-
haviour. This is surprising since they offer a number of potentials to ethologists.
Artificial neural networks can show us how small units like nerve cells can exhibit
powerful computational abilities when working together (Hopfield & Tank, 1986;
Mezard et al., 1987). They also provide understanding about memory and men-
tal representations (McClelland & Rumelhart, 1985), and about mechanisms such

*Published in Animal Behaviour, 56, 1383-1389. © 1998 The Association for the Study of
Animal Behaviour. Minor differences are present compared to the published version. Correspon-
dence (as of December 2, 2006): S. Ghirlanda, [email protected].

1   2   3   4   >>

More intriguing information