Download files
Abstract
This thesis reports the author’s research on the role of neural self-organisation in cognition and cognitive development, the implications of the Self-Organizing Map (SOM) simulation of brain activity at the behavioural level, the prospects of SOM modelling as an explanatory framework to brain disorders, and the cognitive modelling role of SOM properties and parameters, especially the topological neighbourhood during SOM formation. This includes the construction of a number of behavioural and cognitive SOM models, demonstrating behavioural classification, behavioural prediction, and working memory load, and putting existing neuropsychological theories of brain disorders (autism, schizophrenia) into a cognitive modelling perspective. A modified SOM type, with increased biological plausibility, incorporating a type of cortical columnar oscillation in the form of an oscillating topological neighbourhood, is introduced and evaluated alongside the standard SOM.
The artificial neural network class of self-organizing maps is of particular theoretical and engineering importance, and a principal constituent of the neurocomputational cognitive paradigm. SOM networks have a number of properties and characteristics that offer remarkable statistical and engineering computational power, and are biologically relevant to the developmental aspect of cognition as well as to structural and functional elements of the neocortex.
This research offers insights on brain functioning, cognitive development and the mechanisms of higher mental processes, a novel way of applying connectionism to computational developmental neuropsychology and to behavioural modelling, and an assessment of SOM cognitive modelling. The thesis demonstrates that the SOM modelling approach offers significant levels of behavioural classification and prediction when based on an appropriate domain encoding, and could assist in revealing the etiology and mechanisms of brain-behavioural neurodevelopmental disorders such as autism and schizophrenia. It also argues that SOM topological neighbourhood oscillation is a more biologically relevant mechanism and demonstrates its functional and computational equivalence to the standard SOM.
As a result of this work, further SOM cognitive and behavioural modelling research is encouraged, particularly on educational psychology, on brain reorganisation due to impairment, and on atypical clinical phenotypes of memory and executive function.