Growing synfire chains with triphasic spike-time-dependent plasticity

How collections of neurons combine into functional networks capable of intricate and accurate information processing is one of the biggest and most interesting challenges in
neuroscience today. To approach this challenge, it is necessary to address the problem one structure at a time. In this thesis the focus is the development of synfire chains. Synfire chains are feed-forward neural structures which have long been suggested as a possible mechanism by which precisely timed sequences of neural activity could be generated. Precise spatiotemporal firing patterns are known to occur in the brains of many animals including, rats, mice, song birds, monkeys and humans. Such firing patterns have been linked with a wide range of behaviours including motor responses and sensory encoding. There have been many previous computational studies which address the development
of synfire chains. However, they have all required either initial sparse connectivity or strong topological constraints in addition to any synaptic learning rules. Here, it is shown that this necessity can be removed. In this model, development is guided by an experimentally
reported spike-timing-dependent plasticity (STDP) rule, triphasic STDP, plus activity-dependent excitability. This STDP rule, which has not been previously used in
computational studies, is shown to successfully develop a synfire chain in a network of binary neurons. The width and length of the final chain can be controlled through model
parameters. In addition, it is possible to embed multiple chains within one neural network. Next, the effect of triphasic STDP is investigated in a network of more realistic leaky integrate and fire neurons. Here, synfire chain development is shown to be robust in the presence of heterogeneous delays. Finally, the development is described as a random walk, creating a concrete relationship between the model parameters and final network structure.