STDP, rate-coded Hebbian learning and auto-associative network models of the hippocampus

Auto-associative network models have proven extremely useful in modelling the hypothesised function of the CA3 region of the hippocampus in declarative memory. To date, the majority of these models have made use of Hebbian plasticity rules mediated by correlations between mean firing rates. However, recent neurobiological evidence suggests that synaptic plasticity in the hippocampus, and many other cortical regions, also depends explicitly on the temporal relationship between afferent action potentials and efferent spiking – a phenomena known as spike-timing dependent plasticity (STDP). Few attempts have been made to reconcile previous rate-coded Hebbian learning rules or autoassociative network function with this novel plasticity formulation. Further complications arise from the fact that there are many computational interpretations of the empirical data regarding STDP, each of which can have a unique effect on emergent network properties. The aims of this research are therefore three-fold: firstly, to provide a comprehensive description of the emergent dynamics generated by a wide range of STDP implementations within a biologically inspired spiking recurrent neural network; secondly, to reconcile these emergent dynamics with those generated by previous ratecoded Hebbian learning rules, as characterised by the BCM formulation; and finally, to employ these STDP implementations within a simple rate-coded auto-associative network model. The results presented demonstrate that several incarnations of the STDP rule can mediate rate-coded Hebbian learning and replicate numerous other features of synaptic and neural dynamics that are realistic of the CA3 region. These forms of STDP are consequently demonstrated to mediate efficient and robust autoassociative network function, although several issues which might affect the long-term stability and performance of the plasticity rule are identified and discussed. The results should therefore allow the successes of previous auto-associative network models of the hippocampus to be replicated, whilst increasing their versatility and computational power and providing them with a firmer basis in modern neurobiology. Submitted for the degree of DPhil University of Sussex September 2008