Y. G. Drevs, A. S.Migalev
This review is devoted to investigation, modeling and implementation of spike-timing-dependent plasticity. Spike-timing dependent plasticity (STDP) is a type of synaptic plastisticity where change in synaptic strength depends on time order of pre- and postsynaptic spike. It is a subject of many neurophysiological investigations. The properties of STDP depend on type of synaptic transmission, synapse location and type of the targeting cell. Works on experimental research of this type of plasticity on excitatory and inhibitory synapses, using complex stimuli, are reviewed. Theoretical studies investigate its role in working neural network, properties that STDP add to a model of neural network, compute optimal function of STDP using various hypotheses on tasks of this type of synaptical plasticity in neural networks. Some investigations is devoted to reward modulated STDP and how it is related to theory of reinforcement learning. They offer a connection of cell mechanisms of synaptic plasticity with animal learning.
Learning algorithms of neural networks models on basis of STDP are reviewed, algorithms that take into account complex spike trains on neuron input. An important property of this learning algorithms is that they use information about activity of two cells for synaptic strength change. This property allows to implement learning algorithm in neural network model with large amount of neurons. Some studies are devoted to methods for modeling neural networks, using STDP on basis of analog and digital computing systems. Implementation of neuron models with STDP on analog VLSI is made using various circuits on chips or using external units for synaptic strength change. STDP is used as an learning algorithm in robots navigating systems, for synchrony search in spike trains and in many others technical applications