Abstract
We have proposed neural network mechanisms, whereby some conscious and unconscious mental activity, such as the compulsion to repeat in neurosis, is described as associative memory functioning. Freuds description of neurotic behavior helps us understand that unconscious acts are isolated from symbolic representation and association (as in reflexes), and thus contributes to current work regarding the development of machine models of consciousness. In our model, modules corresponding to sensorial and symbolic memories interact, representing unconscious and conscious processing. Memory was first modeled by a Boltzmann machine (BM), represented by a complete graph. Since it is known that brain neural topology is selectively structured, we have further developed the memory model, including known microscopic mechanisms that control synaptic properties and self-organize the complex network to a hierarchical, clustered structure. The resulting power-law and q-exponential behavior for the node degree distribution of the network’s topology suggests that memory dynamics and associativity may not be well described by Boltzmann-Gibbs statistical mechanics. We thus modeled memory access dynamics by a generalization of the BM called Generalized Simulated Annealing (GSA), derived from the nonextensive formalism. We are investigating the effects of using both the BM and GSA, in the associations resulting from memory dynamics.
When
Tuesday 21 June 2011 at noon
Where
Gabor Seminar Room 611
Electrical & Electronic Engineering Building
Imperial College London
South Kensington Campus
London
SW7 2AZ