Seung Kee Han

Attack vulnerability of complex networks

We study the response of complex networks subject to attacks on vertices and edges. Several existing complex network models as well as real-world networks of scientific collaborations and Internet traffic are numerically investigated, and the network performance is quantitatively measured by the average inverse geodesic length and the size of the largest connected subgraph. For each case of attacks on vertices and edges, four different attacking strategies are used: removals by the descending order of the degree and the betweenness centrality, calculated for either the initial network or the current network during the removal procedure. It is found that the removals by the recalculated degrees and betweenness centralities are often more harmful than the attack strategies based on the initial network, suggesting that the network structure changes as important vertices or edges are removed. Furthermore, the correlation between the betweenness centrality and the degree in complex networks is studied.

Path finding strategies in scale-free networks

We numerically investigate the scale-free network model of Barabási and Albert [A. L. Barabási and R. Albert, Science 286, 509 (1999)] through the use of various path finding strategies. In real networks, global network information is not accessible to each vertex, and the actual path connecting two vertices can sometimes be much longer than the shortest one. A generalized diameter depending on the actual path finding strategy is introduced, and a simple strategy, which utilizes only local information on the connectivity, is suggested and shown to yield small-world behavior: the diameter D of the network increases logarithmically with the network size N, the same as is found with global strategy. If paths are sought at random, D is equivalent to N(0.5) is found.

Independent Component Analysis of Ictal EEG in Medial Temporal Lobe Epilepsy

Summary:  Purpose: Application of independent component analysis (ICA) to interictal EEGs and to event‐related potentials has helped noise reduction and source localization. However, ICA has not been used for the analysis of ictal EEGs in partial seizures. In this study, we applied ICA to the ictal EEGs of patients with medial temporal lobe epilepsy (TLE) and investigated whether ictal components can be separated and whether they indicate correct lateralization.

Spontaneous Oscillatory Rhythm in Retinal Activities of Two Retinal Degeneration (rd1 and rd10) Mice

Previously, we reported that besides retinal ganglion cell (RGC) spike, there is ~ 10 Hz oscillatory rhythmic activity in local field potential (LFP) in retinal degeneration model, rd1 mice. The more recently identified rd10 mice have a later onset and slower rate of photoreceptor degeneration than the rd1 mice, providing more therapeutic potential. In this study, before adapting rd10 mice as a new animal model for our electrical stimulation study, we investigated electrical characteristics of rd10 mice. From the raw waveform of recording using 8×8 microelectrode array (MEA) from in vitro-whole mount retina, RGC spikes and LFP were isolated by using different filter setting. Fourier transform was performed for detection of frequency of bursting RGC spikes and oscillatory field potential (OFP). In rd1 mice, ~10 Hz rhythmic burst of spontaneous RGC spikes is always phase-locked with the OFP and this phase-locking property is preserved regardless of postnatal ages. However, in rd10 mice, there is a strong phase-locking tendency between the spectral peak of bursting RGC spikes (~5 Hz) and the first peak of OFP (~5 Hz) across different age groups. But this phase-locking property is not robust as in rd1 retina, but maintains for a few seconds. Since rd1 and rd10 retina show phase-locking property at different frequency (~10 Hz vs. ~5 Hz), we expect different response patterns to electrical stimulus between rd1 and rd10 retina. Therefore, to extract optimal stimulation parameters in rd10 retina, first we might define selection criteria for responding rd10 ganglion cells to electrical stimulus.

Computerized recognition of Alzheimer disease-EEG using genetic algorithms and neural network

We propose an automatic recognition method of Alzheimers disease (AD) with single channel EEG recording using combined the genetic algorithms (GA) and the artificial neural network (ANN). Five min of the resting spontaneous EEG and the ERP in an auditory oddball task were recorded at P4 site in 16 early AD patients and 16 age-matched normal subjects. EEG and ERP were analyzed to compute their 28 statistical and 2 nonlinear features as well as 88 spectral features and 10 ERP features, to make a feature pool for each 30-s segment of the recording data. The combined GA/ANN was applied to find the minimal set of the dominant features from the feature pool that are most efficient to classify into two groups automatically. The effective 35 features were found and used as inputs of the artificial neural network. The recognition rate of ANN fed by these input was 81.9% for untrained data set. These results suggest that the combined GA/ANN approach may be useful for early detection of AD and that single channel EEG data might be enough to recognize AD. This approach could be extended to a reliable classification system using EEG recording that can discriminate between groups.

Chaotic bursting as chaotic itinerancy in coupled neural oscillators

We show that chaotic bursting activity observed in coupled neural oscillators is a kind of chaotic itinerancy. In neuronal systems with phase deformation along the trajectory, diffusive coupling induces a dephasing effect. Because of this effect, an antiphase synchronized solution is stable for weak coupling, while an in-phase solution is stable for very strong coupling. For intermediate coupling, a chaotic bursting activity is generated. It is a mixture of three different states: an antiphase firing state, an in-phase firing state, and a nonfiring resting state. As we construct numerically the deformed torus manifold underlying the chaotic bursting state, it is shown that the three unstable states are connected to give rise to a global chaotic itinerancy structure. Thus we claim that chaotic itinerancy provides an alternative route to chaos via torus breakdown.

STOCHASTIC SYNCHRONIZATION OF COUPLED COHERENCE RESONANCE OSCILLATORS

The effect of coherence resonance can change the firing process in noise-driven excitable systems towards rather regular dynamics. This effect provides a mechanism of the generation of stochastic oscillations whose characteristics are controlled by noise intensity. Following this, a noisy excitable system can be considered as a corehence resonance oscillator. For such functional units, we investigate the mutual and forced synchronization in terms of locking of the peak frequencies in the power spectrum and also in terms of phase locking. The connection of synchronization phenomenon of noise-induced oscillations and coherence resonance effect is discussed. The examples, studied numerically and experimentally, include Morris–Lecar neuron model and a monovibrator electronic circuit, respectively.

Chaotic Bursting Behavior of Coupled Neural Oscillators

Recently it was shown that dephasing of diffusively coupled neural oscillators leads to a new class of bursting phenomena, where neural oscillators switch between high and low oscillation amplitudes. To analyze this behavior we study a system of three-coupled neurons, which is the most simple one that shows chaotic bursting behavior. For the intermediate values of coupling constant kc, the chaotic bursting behavior occurs. For a quantitative analysis of chaotic bursting, we introduce three mean activities of oscillators. From the Poincare sections we find a period-doubling route to chaos. We illustrate the busting behavior in terms of competition of the single oscillator behavior with the collective one arising from the diffusive coupling of oscillators.

SYNCHRONIZATION AND DECODING INTERSPIKE INTERVALS

Decoding of a sequence of interspike intervals (ISIs) of a neuron model driven by a chaotic stimulus is performed based on the attractor reconstruction method. As stimulus strength increases, both the stimulus estimation error and the prediction error in predicting stimulus crosswise by exploiting ISIs information tend to decrease with transitional drops at certain parameter values. It is analyzed that such behaviors are well explained in the context of synchronization between two chaotic patterns of stimulus and ISI sequence. The result implies that a new scheme of temporal coding at low firing rate regime can be achieved which exploits the preservation of nonlinear deterministic structures in stimulus.

Estimating network link weights from inverse phase synchronization indices

We investigated the possibility of estimating network link weights from the multivariate time series of phase oscillators on a complex network. The inverse phase synchronization index of the coupled oscillator network is found to grow in proportion to the corresponding link weight, as network synchronization occurs for a strong coupling strength. This implies that the network link weights can be estimated from the measurement of the inverse phase synchronization indices. By adopting this estimation method, we successfully reconstructed the minimal spanning tree of the original network from the inverse phase synchronization indices. Even for the weak coupling case, the estimation of the network link weights could be improved significantly by taking the average of a sufficiently large number of configurations.