Evolutionary Computation - Special issue on magnetic algorithms
Towards cortex sized artificial neural systems
Neural Networks
| Hi-index | 0.00 |
Current understanding of the effects of damage on neuralnetworks is rudimentary, even though such understanding could leadto important insights concerning neurological and psychiatricdisorders. Motivated by this consideration, we present a simpleanalytical framework for estimating the functional damage resultingfrom focal structural lesions to a neural network model. Theeffects of focal lesions of varying area, shape, and number on theretrieval capacities of a spatially organized associative memoryare quantified, leading to specific scaling laws that may befurther examined experimentally. It is predicted that multiplefocal lesions will impair performance more than a single lesion ofthe same size, that slit like lesions are more damaging thanrounder lesions, and that the same fraction of damage (relative tothe total network size) will result in significantly lessperformance decrease in larger networks. Our study is clinicallymotivated by the observation that in multi-infarct dementia, thesize of metabolically impaired tissue correlates with the level ofcognitive impairment more than the size of structural damage. Ourresults account for the detrimental effect of the number ofinfarcts rather than their overall size of structural damage, andfor the "multiplicative" interaction between Alzheimer's diseaseand multi-infarct dementia.