Mechanisms Underlying The Robust Synchronization In Excitatory-inhibitory Network And Its Application

Given the fact that the communication and processing of neural information are implemented through the firing patterns of the neuronal population, specifically, the spiking sequence of neurons, the interaction between spiking activities and the learning procedures is a promising topic in the recent researches of computational neuroscience. Whether and how the two activities can be integrated into one single model, therefore, are the main issues we have studied in this thesis.This thesis firstly investigated the general principles of the rhythmic activity of neural population, from the perspective of reductionism. And among the diverse phenomena of periodic dynamics, this thesis clarified the behaviors and the general mechanisms of neural oscillation at first. Both analytical methods and numerical simulations are hired to testify our theories. Several model properties, such as the phase response curve (PRC) of model neurons, parameters setting of synaptic dynamics, and the architecture of the networks, are considered for their contributions to the synchronizing effect.Secondly, this thesis focused on the details of the mechanism of robust synchrony that occurs in the E/I networks (networks of excitatory and inhibitory neurons). The stability of synchronization in E/I networks has been explored in many studies. Neurons in real neural circuits receive highly heterogeneous input. However, how the robust synchronization occurs in these heterogeneous network has not been studied sufficiently. The thesis considered the robustness of the synchronization in E/I network, and found that both the frequency and the synchrony of the network are highly dependent on the parameter settings. A short but indispensable inhibition delay, sufficiently fast decay of inhibitory synapses, and strong connections are critical to turn the E/I networks into robust gamma frequency oscillators. The networks under these conditions can achieve a level of robustness against heterogeneity as high as 30%, which is comparable to the level of heterogeneity observed in real neural systems.Finally, we surveyed the information-processing significance of the robust synchronization in E/I net, by building a feed-forward network with two layers, each of which is a recurrent network based on E/I network. We introduced the coincidence detection as the coding strategy, and took the classic spiking time dependent plasticity (STDP) rules for synaptic learning. By feeding the network with natural images, we can obtain the Gabor basis-like receptive fields form this algorithm, which is an sign of sparse representation.