Hiroyuki Mino

Stochastic Resonance Can Enhance Information Transmission in Neural Networks

Stochastic resonance (SR) is a noise-induced phenomenon whereby signal detection can be improved by the addition of background noise in nonlinear systems. SR can also improve the transmission of information within single neurons. Since information processing in the brain is carried out by neural networks and noise is present throughout the brain, the hypothesis that noise and coupling play an important role in the control of information processing within a population of neurons to control was tested. Using computer simulations, we investigate the effect of noise on the transmission of information in an array of neurons, known as array-enhanced SR (AESR) in an interconnected population of hippocampal neurons. A subthreshold synaptic current (signal) modeled by a filtered homogeneous Poisson process was applied to a distal position in each of the apical dendrites, while background synaptic signals (uncorrelated noise) were presented to the midpoint in the basal dendrite. The transmembrane potentials were recorded in each cell of an array of CA1 neuron models, in order to determine spike firing times and to estimate the total and noise entropies from the spike firing times. The results show that the mutual information is maximized for a specific amplitude of uncorrelated noise, implying the presence of AESR. The results also show that the maximum mutual information increases with increased numbers of neurons and the strength of connections. Moreover, the relative levels of excitation and inhibition modulate the mutual information transfer. It is concluded that uncorrelated noise can enhance information transmission of subthreshold synaptic input currents in a population of hippocampal CA1 neuron models. Therefore, endogenous neural noise could play an important role in neural tissue by modulating the transfer of information across the network.

Response: Implementation Issues in Approximate Methods for Stochastic Hodgkin-Huxley models

JOHN A. WHITE , JAY T. RUBINSTEIN, and HIROYUKI MINO Department of Biomedical Engineering, Boston University, Boston, MA 02215, USA; Department of Otolaryngology/Head & Neck Surgery, University of Washington, Seattle, WA 98195, USA; Department of Bioengineering, University of Washington, Seattle, WA 98195, USA; and Department of Electrical and Computer Engineering, Kanto Gakuin University, Yokohama, 236-8501, Japan

Enhancement of information transmission of sub-threshold signals applied to distal positions of dendritic trees in hippocampal CA1 neuron models with stochastic resonance

Stochastic resonance (SR) has been shown to enhance the signal-to-noise ratio and detection of low level signals in neurons. It is not yet clear how this effect of SR plays an important role in the information processing of neural networks. The objective of this article is to test the hypothesis that information transmission can be enhanced with SR when sub-threshold signals are applied to distal positions of the dendrites of hippocampal CA1 neuron models. In the computer simulation, random sub-threshold signals were presented repeatedly to a distal position of the main apical branch, while the homogeneous Poisson shot noise was applied as a background noise to the mid-point of a basal dendrite in the CA1 neuron model consisting of the soma with one sodium, one calcium, and five potassium channels. From spike firing times recorded at the soma, the mutual information and information rate of the spike trains were estimated. The simulation results obtained showed a typical resonance curve of SR, and that as the activity (intensity) of sub-threshold signals increased, the maximum value of the information rate tended to increased and eventually SR disappeared. It is concluded that SR can play a key role in enhancing the information transmission of sub-threshold stimuli applied to distal positions on the dendritic trees.

Estimation of the density of the filtered Poisson impulse process: a parametric approach

The Poisson driven pth order autoregressive (PDAR(p)) process is defined as the output of a continuous-time autoregressive system, driven by a stationary Poisson impulse process. An explicit formula for estimating the density of the Poisson impulse process is derived by combining the second- and third-order cumulants of the discretized PDAR(p) process. The validity of the proposed method is assessed through Monte Carlo simulations in some specific examples. >

Reverse stochastic resonance in a hippocampal CA1 neuron model

Stochastic resonance (SR) is a ubiquitous and counter- intuitive phenomenon whereby the addition of noise to a non-linear system can improve the detection of sub-threshold signals. The “signal” is normally periodic or deterministic whereas the “noise” is normally stochastic. However, in neural systems, signals are often stochastic. Moreover, periodic signals are applied near neurons to control neural excitability (i.e. deep brain stimulation). We therefore tested the hypothesis that a quasi-periodic signal applied to a neural network could enhance the detection of a stochastic neural signal (reverse stochastic resonance). Using computational methods, a CA1 hippocampal neuron was simulated and a Poisson distributed subthreshold synaptic input (“signal”) was applied to the synaptic terminals. A periodic or quasi periodic pulse train at various frequencies (“noise”) was applied to an extracellular electrode located near the neuron. The mutual information and information transfer rate between the output and input of the neuron were calculated. The results display the signature of stochastic resonance with information transfer reaching a maximum value for increasing power (or frequency) of the “noise”. This result shows that periodic signals applied extracellularly can improve the detection of subthreshold stochastic neural signals. The optimum frequency (110Hz) is similar to that used in patients with Parkinsons suggesting that this phenomenon could play a role in the therapeutic effect of high frequency stimulation.

Effects of rates of spontaneous synaptic vesicle secretions in inner hair cells on information transmission in an auditory nerve fiber model

In this article, we investigate how the rates of spontaneous synaptic vesicle secretions affect information transmission of the spike trains in response to the inner hair cell (IHC) synaptic currents in an auditory nerve fiber (ANF) model through computer simulations. The IHC synaptic currents were modeled by a filtered inhomogeneous Poisson process modulated with sinusoidal functions, while the stochastic ion channel model was incorporated into each node of Ranvier in the ANF model with spiral ganglion. The information rates were estimated from the entropies of the inter-spike intervals of the spike trains to evaluate information transmission in the ANF model. The results show that the information rates increased, reached a maximum, and then decreased as the rate of spontaneous secretion increased, implying a resonance phenomenon dependent on the rate of spontaneous IHC synaptic secretions. In conclusion, this phenomenon similar to the regular stochastic resonance may be observed due to that spontaneous IHC synaptic secretions may act as an origin of fluctuation or noise, and these findings may play a key role in the design of better auditory prostheses.

An influence of spontaneous spike rates on information transmission in a spherical bushy neuron model with stochastic ion channels

This article presents an effect of spontaneous spike firing rates on information transmission of the spike trains in a spherical bushy neuron model of antero-ventral cochlear nuclei. In computer simulations, the synaptic current stimuli ascending from auditory nerve fibers (ANFs) were modeled by a filtered inhomogeneous Poisson process modulated with sinusoidal functions, while the stochastic sodium and stochastic high- and low-threshold potassium channels were incorporated into a single compartment model of the soma in spherical bushy neurons. The information rates were estimated from the entropies of the inter-spike intervals of the spike trains to quantitatively evaluate information transmission in the spherical busy neuron model. The results show that the information rates increased, reached a maximum, and then decreased as the rate of spontaneous spikes from the ANFs increased, implying a resonance phenomenon dependent on the rate of spontaneous spikes from ANFs. In conclusion, this phenomenon similar to the stochastic resonance would be observed due to that spontaneous random spike firings coming from auditory nerves may act as an origin of fluctuation or noise, and these findings may play a key role in the design of better auditory prostheses.

Noise induced oscillations in recurrent neural networks

It has been shown that oscillations can be generated by additive Gaussian white noise in a recurrent Hodgkin-Huxley neuron model. Type 1 oscillation was induced with Stochastic Resonance (SR) by additive Gaussian noise at lower amplitudes, while Type 2 oscillation was observed at higher amplitudes. However, the mechanism of Type 2 oscillation is not clear. In this article, we test the hypothesis through computer simulations that the period of the Type 2 oscillation can be affected by temperature in a recurrent neural network in which the recurrent model is constructed by four Hodgkin-Huxley (HH) neuron models. Each HH neuron model is driven by Gaussian noise and sub-threshold excitatory synaptic currents with an alpha function from another HH neuron model, and the action potentials (spike firings) of each HH neuron model are transferred to the other HH neuron model via sub-threshold synaptic currents. From spike firing times recorded, the inter spike interval (ISI) histogram was generated, and the periodicity of spike firings was detected from the ISI histogram at each HH neuron model. The results show that the probability of spike firings in the Type1 oscillation is maximized at a specific standard deviation (S.D.) of the Gaussian white noise with SR at 6.3, 15.0 and 25.0 oC, while the period of the Type 2 oscillation depends on temperature. It is concluded that the Type1 oscillation can be induced by additive Gaussian white noise on the basis of a synaptic delay in the recurrent HH neuron model, whereas ISIs of the Type 2 oscillation may be determined by refractory periods of HH neuron models.

Stochastic resonance can induce oscillation in a recurrent Hodgkin-Huxley neuron model with added Gaussian noise

Stochastic resonance (SR) has been shown to enhance the signal to noise ratio or detection of signals in neurons. It has not been clear yet if SR can play an important role in information processing in neural networks. In this paper, we test the hypothesis through computer simulations that SR can induce an oscillation phenomenon in a recurrent neural network with added Gaussian noise in which the recurrent model is constructed by four Hodgkin-Huxley (HH) neuron models. Each HH neuron model is driven by Gaussian noise and sub-threshold excitatory synaptic currents with an alpha function from another HH neuron model, and the action potentials (spike firings) of each HH neuron model are transferred to the other HH neuron model via sub-threshold synaptic currents. From spike firing times recorded, the inter spike interval (ISI) histogram was generated, and the periodicity of spike firings was detected from the ISI histogram at each HH neuron model. The results show that the probability of spike firings in the oscillation period (about 50 ms or 20 Hz) increases as the standard deviation (S.D.) of the Gaussian white noise increases, and reach a maximum value at a specific S.D. of the Gaussian white noise, implying that SR can improve sub-threshold synaptic transmission in the recurrent HH neuron model. It is concluded that an oscillation (20 Hz) can be induced by adding Gaussian white noise at lower amplitudes with intrinsic characteristics in the recurrent HH neuron model, while another oscillation (100 Hz) can be generated by the noise at greater amplitudes with extrinsic characteristics.

Information Transmission in Hippocampal CA1 Neuron Models in the Presence of Poisson Shot Noise: the Case of Periodic Sub-threshold Spike Trains

This article presents an analysis of the information transmission of periodic sub-threshold spike trains in a hippocampal CA1 neuron model in the presence of a homogeneous Poisson shot noise. In the computer simulation, periodic sub-threshold spike trains were presented repeatedly to the midpoint of the main apical branch, while the homogeneous Poisson shot noise was applied to the mid-point of a basal dendrite in the CA1 neuron model consisting of the soma with one sodium, one calcium, and five potassium channels. From spike firing times recorded at the soma, the inter spike intervals were generated and then the probability, p(T), of the inter-spike interval histogram corresponding to the spike interval, r, of the periodic input spike trains was estimated to obtain an index of information transmission. In the present article, it is shown that at a specific amplitude of the homogeneous Poisson shot noise, p(T) was found to be maximized, as well as the possibility to encode the periodic sub-threshold spike trains became greater. It was implied that setting the amplitude of the homogeneous Poisson shot noise to the specific values which maximize the information transmission might contribute to efficiently encoding the periodic sub-threshold spike trains by utilizing the stochastic resonance