The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #92130

Title:Analysis of Marr'S Model of Neocortex
Date: 1992
Abstract:This project concerns the simulation of Marr's classification theory of the neocortex. In his theory, Marr proposed self-organising net consisting of two layers, with the upper layer consisting of modifiable synapses and the lower layer unmodifiable excitatory synapses except during certain phases of sleep when these excitatory connections are formed. The central ideas of the theory are: output cells in the neocortex detect for similar events occurred in the information received, the modifiable synapses of the output cell record the conditional probabilities of pre- and post-synaptic activities and an intermediate layer, so called codon cells, is useful in producing the desired activity for successful classification. In the project, the setting of the various network parameters and the possible need for using a climbing fibre in selecting a mountain in an event space of several mountains have been determined. Most important of all, the output cell is found to be able to classify successfully with modifiable synapses that record conditional probabilities when trained with the method of synchronous update but without using climbing fibres. The codon transformation process was also successfully explored and the use of the qualities of evidence collected by the codon cells in the classification performance of the output cell identified.

[Search These Pages] [DAI Home Page] [Comment]