Use this URL to cite or link to this record in EThOS:
Title: A computational investigation of neural dynamics and network structure
Author: Connor, Dustin Thomas
ISNI:       0000 0004 2706 2586
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Access from Institution:
With the overall goal of illuminating the relationship between neural dynamics and neural network structure, this thesis presents a) a computer model of a network infrastructure capable of global broadcast and competition, and b) a study of various convergence properties of spike-timing dependent plasticity (STDP) in a recurrent neural network. The first part of the thesis explores the parameter space of a possible Global Neuronal Workspace (GNW) realised in a novel computational network model using stochastic connectivity. The structure of this model is analysed in light of the characteristic dynamics of a GNW: broadcast, reverberation, and competition. It is found even with careful consideration of the balance between excitation and inhibition, the structural choices do not allow agreement with the GNW dynamics, and the implications of this are addressed. An additional level of competition – access competition – is added, discussed, and found to be more conducive to winner-takes-all competition. The second part of the thesis investigates the formation of synaptic structure due to neural and synaptic dynamics. From previous theoretical and modelling work, it is predicted that homogeneous stimulation in a recurrent neural network with STDP will create a self-stabilising equilibrium amongst synaptic weights, while heterogeneous stimulation will induce structured synaptic changes. A new factor in modulating the synaptic weight equilibrium is suggested from the experimental evidence presented: anti-correlation due to inhibitory neurons. It is observed that the synaptic equilibrium creates competition amongst synapses, and those specifically stimulated during heterogeneous stimulation win out. Further investigation is carried out in order to assess the effect that more complex STDP rules would have on synaptic dynamics, varying parameters of a trace STDP model. There is little qualitative effect on synaptic dynamics under low frequency (< 25Hz) conditions, justifying the use of simple STDP until further experimental or theoretical evidence suggests otherwise.
Supervisor: Shanahan, Murray Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral