Preoccupation with a Restricted Pattern of Interest in Modelling Autistic Learning

Abstract. Autism is a developmental disorder in which attention shift-ing is known to be restricted. Self-organization of neural networks, con-ditioned by different attention shifting characteristics is investigated for higher-dimensional stimuli presented to the network from different sources. The attention shifting modes are 1) novelty seeking, 2) attention shift impairment (attention is shifted but with a low probability) and 3) atten-tion is shifted with a preference for a source which has become familiar to the map. The feature maps resulting from self-organization are much the same for modes 1 and 2 but distinctly different for mode 3, where the maps learn the stimuli from the source with the lowest variability in great detail, at the expense of the other source(s). Detailed learning in narrow fields is a known characteristic of autism