Attention shift impairments and novelty avoidance : effects of characteristics of autism on the self-organization of an artificial neural network

We discuss application of Artificial Neural Networks (ANN) in simulation of attention shift impairments and novelty avoidance, common deficits in autism. It has been theorized that cortical feature maps in individuals with autism are inadequate for forming abstract codes and representations, explaining the importance paid to detail, rather than salient features. ANNs known as the Self-Organization Maps (SOM) offer insights into the development of cortical feature maps. We present results of the formation of SOMs in response to stimuli from two sources in four modes, namely, novelty seeking (normal learning), attention shift impairment, novelty avoidance and novelty avoidance in conjunction with attention shift impairment. The SOMs resulting from learning with novelty seeking and with attention shift impairment were, perhaps surprisingly, identical. In the case of learning with novelty avoidance the resulting SOMs were adapted to one of the sources at the expense of the other. The SOMs resulting from learning with novelty avoidance in conjunction with attention shift impairment were strikingly different, ranging from almost normal to poor from one simulation to the next, even with identical initial conditions. Such learning, in many different maps, would result in very uneven capacities, common in individuals with autism.Godkänd; 2001; 20061101 (ysko