Self-organizing task modules and explicit coordinate systems in a neural network model for 3-D saccades

Citation
Ma. Smith et Jd. Crawford, Self-organizing task modules and explicit coordinate systems in a neural network model for 3-D saccades, J COMPUT N, 10(2), 2001, pp. 127-150
Citations number
74
Language
INGLESE
art.tipo
Article
Categorie Soggetti
Neurosciences & Behavoir
Journal title
JOURNAL OF COMPUTATIONAL NEUROSCIENCE
ISSN journal
0929-5313 → ACNP
Volume
10
Issue
2
Year of publication
2001
Pages
127 - 150
Database
ISI
SICI code
0929-5313(200103)10:2<127:STMAEC>2.0.ZU;2-V
Abstract
The goal of this study was to train an artificial neural network to generat e accurate saccades in Listing's plane and then determine how the hidden un its performed the visuomotor transformation. A three-layer neural network w as successfully trained, using back-prop, to take in oculocentric retinal e rror vectors and three-dimensional eye orientation and to generate the corr ect head-centric motor error vector within Listing's plane. Analysis of the hidden layer of trained networks showed that explicit representations of d esired target direction and eye orientation were not employed. Instead, the hidden-layer units consistently divided themselves into four parallel modu les: a dominant "vector-propagation" class (similar to 50% of units) with s imilar visual and motor tuning but negligible position sensitivity and thre e classes with specific spatial relations between position, visual, and mot or tuning. Surprisingly, the vector-propagation units, and only these, form ed a highly precise and consistent orthogonal coordinate system aligned wit h Listing's plane. Selective "lesions" confirmed that the vector-propagatio n module provided the main drive for saccade magnitude and direction, where as a balance between activity in the other modules was required for the cor rect eye-position modulation. Thus, contrary to popular expectation, error- driven learning in itself was sufficient to produce a "neural" algorithm wi th discrete functional modules and explicit coordinate systems, much like t hose observed in the real saccade generator.