A Neural Network Model for Image Formation of the Face Through Double-Touch Without Visual Information

Abstract

The fact that neonates only 9 minutes old preferentially look at faces suggests that they have already acquired the visual schema of a face. However, the mechanism for learning the schema of ‘face’ during the fetal period is still unknown. In this paper, a neural network model that can acquire visual schema without visual input is proposed, which is based on the fact that fetuses frequently touch their own faces. This model consists of three major functions: (1) transmission of topological information about touched area on the face to visual area through VIP, which is realized by a 2-layered SOM with Gaussian- like local connections, (2) coding of shape information with Gabor filters by finger-touching, and (3) integration of two kinds of information (i.e., topological and shape information) using Hebbian connection. The sum of Hebbian activation appears to be similar to the representation of face in polar- coordinates.