A hierarchical neural model sensitive to interaural time differences.

Abstract

Event Abstract Back to Event A hierarchical neural model sensitive to interaural time differences. Viacheslav Vasilkov1*, Irina Ischenko1 and Ruben Tikidji - Hamburyan1 1 Southern Federal University, Research Institute for Neurocybernetics, Russia The exquisite localization of sound sources in the horizontal plane is achieved due to an exceptional sensitivity of the auditory system to interaural time differences (ITDs) [1-3]. The temporal precision with which the binaural auditory system can detect these short ITDs is significantly greater than sensitivity to ITDs of single auditory neurons and can be in the range of a few microseconds [1]. The classical idea of binaural hearing, explaining the ITD processing with such high temporal accuracy, is represented by the Jeffress coincidence detection model [2]. Nevertheless, several recent physiological evidences can not be interpreted in the framework of the coincidence model, and specific neural mechanisms underlying ITD detection acuity are still open issues. In this paper, we consider another biologically inspired ITD detection mechanism based on impulse activity comparison of two symmetric EI neuron [3] populations, and present a hierarchical neural network model which is highly sensitive to microsecond ITDs. Suggested neural network comprises two layers and mimics two processing stages of binaural auditory signal. Both layers contain couple of homogeneous populations each of which consists of no less than one hundred single neuron models. Every single neuron model of first layer receives excitatory synaptic projection from ipsilateral input element. Neuron model of second layer simulates the behavior of auditory EI neuron and has excitatory and inhibitory synapse with single neurons of ipsilateral and contralateral populations of first layer, respectively. The parameters of single neuron model are similar through all populations, whereas the conductivity of synaptic transmission has a normal distribution. We use a non-linear model of single neuron [4] and reverse model of synapse [5] to obtain more biologically plausible results. In order to examine the influence of noise in the neuron activity on the ITDs detection by proposed neural network model, the generator modules of bipolar white noise (injected into single neuron model) were incorporated. Our present study provides computational evidence that the proposed hierarchical neural model detects microsecond ITDs with high resolution even though the membrane time constant and synaptic time constant are in the range of one millisecond. Considered network model provides high-precision ITDs detection only in the case of bursting neuron activity at first layer, which is well correlated with physiological data. We have also assessed the robustness of ITD sensitivity of the hierarchical model to amplitude noise in the neuron of first layer, and shown that such noise can cause the phase noise, which reflects time travel differences of ipsi- and contralateral spikes. Refinement of the ITD detection in the case of random distribution of synaptic weight has been also illustrated.