Communications of the ACM
Did These Researchers Just Create an Autistic Computer Program?
Two researchers say they have modeled a theorized cause of autism in an artificial neural network.
Credit: Extreme Tech
Baylor College of Medicine researchers Ari Rosenberg and Jaclyn Sky Patterson claim to have modeled a theorized cause of autism--divisive normalization (DN)--in an artificial neural network (ANN), with the effect of that simulation exhibiting recognizably autistic behavior.
The DN theory suggests autistic brains are so active that the activity of any single neuron is inhibited by the activity of the general population of surrounding neurons. ANNs are organized along the principle that one minuscule adjustment to the behavior of all neurons carries a major cumulative effect on the final outcome of the data being processed.
Rosenberg and Patterson's experiment used an ANN to simulate the primary visual cortex, and they began tinkering with its parameters to see if the DN theory could be used to span the gap between autism's subjective effects and the ANN's numerical operations. Comparing two ANNs--one representing a normal person's visual processing and the other an autistic subject's--showed the former consistently bested the latter, mirroring the same trends as human test subjects.
Additional tests with "tunnel vision" and the known linkage between statistical inference and autism demonstrated correlation with autistic subjects as well.
From Extreme Tech
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found