| | |
Summary: Infant and Child Development, Vol.10, No. 1, pages 1920. 2001. c
FACE RECOGNITION BY MYOPIC BABY NEURAL
NETWORKS
Dominique Valentin
& Hervé Abdi
Université de Bourgogne ŕ Dijon
The University of Texas at
Dallas
Do we need to assume that face recognition represents an innate ability to explain
the fact that newborns are able to discriminate mother from stranger? Indeed if
we consider that faces are highly similar, that newborns possess poor visual acuity
and contrast sensitivity, and that they cannot resolve high spatial frequencies, this
ability seems paradoxical. However, this paradox might come from the fact that
we evaluate the complexity of the task performed by newborns through our adult
experience. A less biased way of evaluating the complexity of a perceptual task is
to simulate the task via an articial neural network (Abdi, Valentin & Edelman,
1999; Abdi, Valentin, Edelman & O'Toole, 1995). The performance of such a
model depends both on the learning algorithm implemented and on the number
and perceptual characteristics of the stimuli on which the learning algorithm is
|
