Competitive Hebbian learning through spike timing -dependent plasticity (STDP)

Recent experimental findings indicate that the difference in pre- and postsynaptic spike time can dictate both the direction and magnitude of changes in synaptic strength. This thesis explores the consequences of spike-timing-dependent plasticity (STDP) on single synapse, single neuron and network levels. I incorporate experimental data into a mathematical model and study the consequences of this plasticity rule. First, I find that, STDP gives rise to a stable distribution of synaptic strengths without the requirement of additional constraints. Furthermore, it normalizes the postsynaptic firing rate and coefficient of variation for inputs of varying rates. This is explained by studying the correlation between the input and output spike trains. STDP automatically places the postsynaptic neuron into a balanced regime in which it is highly sensitive to the timing of presynaptic spikes. STDP is sensitive to correlations in the inputs, especially synchronous spiking events. I developed a firing-rate based mathematical model which can predict the changes in synaptic strengths for various kinds of correlations in the inputs with good accuracy. I also find that changing the basic STDP rule by incorporating a delay in expression time, synaptic redistribution mechanism, or variations in maximal synaptic strength confers additional interesting properties, while retaining basic features of STDP. In a network, STDP can lead to the formation of columns and maps without the use of additional constraints. The process of formation of columns and maps can be explained by a novel process of transfer of connectivity patterns that involves sensitivity to spike timing. I demonstrate the formation of maps with both plastic feedforward and recurrent connections when global inhibition is imposed. STDP can also explain experimental findings concerning adult plasticity in cortical maps after lesions. I also find that making the STDP rule dynamic and dependent on the postsynaptic firing rate normalizes against changing input synaptic correlations and makes the rule more stable.