A Critical Analysis of an Attractor Network Model of Schizophrenia
Hoffman and Dobscha (1989) (H+D hereafter) used an attractor (Hopfield) neural network to model the effects of over-pruning in the human cerebral cortex in an attempt to demonstrate an underlying brain mechanism for schizophrenia. One of their main findings is the emergence of autonomous regions of activity in the network unrelated to input—they call these ‘parasitic foci’. I have looked at the analytical work on attractor networks and show that parasitic foci are overlaps in spin glass states which automatically exist in the network, and that pruning is equivalent to raising the ‘temperature’ in the stochastic state update equation. In addition I show that overloading rather than pruning the network will give the same qualitative results. This analysis is supported by various computer simulations. The conclusion is that while parasitic foci may form a very weak analogy with schizophrenic symptoms, their pathogenesis cannot be attributed to one particular process in the model. Even if the analogy holds the model has still not helped to elucidate the brain mechanism which underlies schizophrenia. I also discuss the inadequacy of the Hopfield network as a model of biological processes and schizophrenia.
KeywordsHuman Cerebral Cortex Pruning Rule Hopfield Network Spin Glass State Attractor Network
Unable to display preview. Download preview PDF.
- Hertz, J; Krogh, A; Palmer, RG (1991): Introduction to the theory of neural computation. Pubs Addison Wesley.Google Scholar