Computational Intelligence and Neuroscience

Computational Intelligence and Neuroscience / 2016 / Article

Research Article | Open Access

Volume 2016 |Article ID 8939218 | https://doi.org/10.1155/2016/8939218

L. L. Neves, L. H. A. Monteiro, "A Linear Analysis of Coupled Wilson-Cowan Neuronal Populations", Computational Intelligence and Neuroscience, vol. 2016, Article ID 8939218, 6 pages, 2016. https://doi.org/10.1155/2016/8939218

A Linear Analysis of Coupled Wilson-Cowan Neuronal Populations

Academic Editor: Saeid Sanei
Received23 Feb 2016
Accepted04 Aug 2016
Published20 Sep 2016

Abstract

Let a neuronal population be composed of an excitatory group interconnected to an inhibitory group. In the Wilson-Cowan model, the activity of each group of neurons is described by a first-order nonlinear differential equation. The source of the nonlinearity is the interaction between these two groups, which is represented by a sigmoidal function. Such a nonlinearity makes difficult theoretical works. Here, we analytically investigate the dynamics of a pair of coupled populations described by the Wilson-Cowan model by using a linear approximation. The analytical results are compared to numerical simulations, which show that the trajectories of this fourth-order dynamical system can converge to an equilibrium point, a limit cycle, a two-dimensional torus, or a chaotic attractor. The relevance of this study is discussed from a biological perspective.

1. Introduction

In the last decades, the features of action potentials propagating along axonal membranes have been related to the oscillations found in electroencephalogram (EEG) records [13]. From a theoretical perspective, analogies between EEG and nonlinear dynamics [4] were formulated prior to knowledge of the seminal numerical simulation by Lorenz, published in 1963, about a chaotic system [5]. All these studies aim to disclose the neural code; that is, to understand how information is represented, carried, and transformed by neurons [6]. The ambition is to explain how cognitive functions emerge from neuron spikes. In this context, here we analytically investigate the activity of a simple neuronal assembly, in order to examine how its parameter values influence its dynamical behavior.

Consider a population formed by two groups of interacting neurons, so that the synapses of the first group are excitatory and the synapses of the second group are inhibitory. Consider also that these two groups are connected. In 1972, Wilson and Cowan proposed a mathematical model to describe the time evolution of the activity of this neuronal population [7]. In this model, the activity of each group obeys a nonlinear first-order ordinary differential equation with the following form: in which denotes the time, is a positive constant related to the natural decay of , is a positive constant proportional to the refractory period, and represents the input to this group, written as a linear combination of the activities of both groups plus the influence of external stimuli. Resting activity corresponds to ; thus, a negative value of expresses a depression from this background level. The function must be sigmoidal. However, as stated by Wilson and Cowan [7], “no particular significance is to be attached to the choice of” . Usual choices are [8], [9], and [10]. In several studies, the properties of a single population were explored via numerical simulations [8, 9, 11], since these usual sigmoidal functions pose difficulties to analytical works. The dynamics of the two-population model was also already numerically investigated [10, 1215].

Monteiro et al. suggested to use and some analytical results were derived for the Wilson-Cowan model [16]. Here, we also take this particular form of to analytically investigate the dynamics of two coupled Wilson-Cowan populations (composed of four groups). We also assume that (1) the isolated populations are identical (in the sense that they are characterized by the same parameter values, as considered in other studies [10, 1215]); (2) the populations are coupled by links starting from their excitatory groups (because most synapses are excitatory; for instance, they are in the cat cortex [17]); (3) the refractory period is negligible (which corresponds to taking in (1), as supposed in several works [9, 10, 12, 14, 1820]).

Some authors took as a piecewise linear function [14, 1921]. In our analyses, we assume that the parameter values related to this neuronal assembly allow us to take ; thus, the model becomes linear. From this approximation, the occurrence of a Hopf bifurcation in the nonlinear model is analytically inferred. Other bifurcations are numerically found. Recall that bifurcation corresponds to a qualitative change in the dynamical behavior of a system caused by variation of parameter value(s).

This manuscript about the dynamics of a pair of coupled Wilson-Cowan neuronal populations is structured as follows. In Section 2, the model is presented and analyzed by taking into account the linear approximation just described. In Section 3, the analytical results are compared to numerical simulations performed by considering the nonlinear version of the model. We found that, as the time passes, the variables of the model can converge to an equilibrium point, a limit cycle, a two-dimensional torus, or a chaotic attractor. In Section 4, the results of this work are discussed from a biological point of view.

2. Analytical Results

Let two identical populations be coupled by links that can be symmetrical or asymmetrical. Assume that the variables and denote the activity of the excitatory group and the activity of the inhibitory group, respectively, of th population, with . According to Wilson and Cowan, the dynamics of these coupled populations can be described by The parameters and represent the natural (exponential) decay. The parameters , , , and are the strengths of the connections between the excitatory and inhibitory groups of an isolated population, as shown in Figure 1: is the inhibitory connection from to , is the excitatory connection from to , is the self-excitatory connection (from to ), and is the self-inhibitory connection (from to ). The strengths of the connections between the populations are denoted by the parameters and : is the connection from to , is the connection from to , is the connection from to , and is the connection from to . All these ten parameters are positive constants. The constants , , , and stand for stimuli from external sources reaching , , , and , respectively. Thus, the dimension of the parameter space is 14.

A steady state is a stationary solution, corresponding to an equilibrium point in the four-dimensional state space . The constants , , , and are obtained from , , , and . By taking into consideration the linear approximation proposed in Section 1, there is only one equilibrium point given by with and . Note that if the populations are isolated, that is, if , then and , for . Obviously, if and , then and .

The local stability of an equilibrium point can be determined from the eigenvalues of the Jacobian matrix obtained from the system of (2) linearized around this point [2224]. By the Hartman-Grobman theorem, such a point is locally asymptotically stable if all eigenvalues have negative real parts [2224]. For this fourth-order system, the eigenvalues are the roots of the polynomial: with Routh-Hurwitz criterion states that all eigenvalues have negative real parts if , , , , , and [25]. For instance, in the case of isolated populations, local stability of the steady-state solution is assured if and .

Observe that the stability conditions do not depend on the values of and . Thus, constant external stimuli do not modify the stability of the stationary solution (in the linear approximation). Observe also that a necessary condition for stability is ; that is, or . Therefore, the equilibrium point is locally asymptotically stable only if the strength of the self-excitatory connection is below the critical value and/or if the strength of the self-inhibitory connection is above the critical value .

If , then diminishes as and/or increase; if , then , , and diminish as and/or increase. Thus, both kinds of connections between the populations, characterized by and , can singly destabilize the equilibrium point, if their strengths exceed critical numbers.

If the parameter values are varied so that , then a Hopf bifurcation [2224] can occur (because two roots of (4) are purely imaginary complex-conjugate numbers and the other two roots have negative real parts; i.e., (4) can be written as , with ). As a consequence, a limit cycle with oscillation period appears. Recall that a limit cycle is an isolated closed trajectory in the state space, corresponding to a periodic solution [2224].

If and , the condition is equivalent to . In this particular case, there is the birth of a limit cycle for if . The period of the newborn cycle is approximately given by .

As shown in the next section, other bifurcations can take place by varying the parameter values of this dynamical system.

3. Numerical Simulations

In the simulations presented in this section, the values of , , , , , , , , and are kept fixed and the values of , , , , and are varied. Thus, we investigate how the dynamics is influenced by the strengths of the self-excitatory loops and of the links between the populations.

Here, we take , , , , , , and . In Figures 2(a)2(i), only is plotted in function of . In each case, the behaviors of the other three variables are qualitatively similar; hence, their plots were omitted. For instance, if converges to a periodic solution, then , , and also tend to a periodic oscillation as the time passes. In (a), , ; in (b), , ; in (c), , , ; in (d), , , and ; in (e), , , and , ; in (f), , , and ; in (g), , , and ; in (h), , , and ; and in (i), , , and . In these simulations, (2) were numerically solved by employing the fourth-order Runge-Kutta integration method with integration step of 0.01. In addition, in all simulations, the initial condition is .

In (a) and (b), the populations are isolated. Therefore, the attractor can be either an equilibrium point or a limit cycle, because (2) split into two decoupled second-order systems. In fact, these are the attractors that can exist in the state space of second-order autonomous nonlinear systems [2224].

In (a), the conditions for the asymptotical stability of the steady-state solution are satisfied (because and ). In this case, converges to , which is the same value obtained from the analytical expression . Thus, the linear approximation of the sigmoidal function can be considered as valid to determine this equilibrium point and its stability. For , the system experiences a Hopf bifurcation. Hence, in (b), as , tends to a periodic solution.

In cases (c)–(i), the populations are coupled. Now, the nature of the attractor is determined from its Lyapunov exponents , , , and , which were numerically computed by using the algorithm proposed by Wolf et al. [26]. When , the attractor is an equilibrium point; when and , the attractor is a limit cycle; when and , the attractor is a two-dimensional torus; when , , and , the attractor is chaotic [2224].

In (c), the Routh-Hurwitz criterion is satisfied (because , , , , , and ) and converges to , which matches the number calculated from (3). In this case, and .

As stated in Section 2, by varying the parameter values, if becomes equal to zero, then a Hopf bifurcation can take place. For instance, for and , then if ; thus, for , the attractor is a limit cycle, as shown in (d). In this case, , , , and . Recall that is equivalent to when and ; thus, corresponds to , which is equal to the value numerically found. For , the oscillation period is about . In (d), for , the period is still .

Even when and , a limit cycle arises if . Hence, in (e), with , the asymptotical behavior is a regular oscillation. It corresponds to a limit cycle, because , , and .

It is easy to check that the steady state loses its stability when , that is, when . In (f), with , , and , the trajectories in the state space converge to a two-dimensional torus. In fact, the Lyapunov exponents are and . Therefore, by increasing , the system experiences a bifurcation concerning the birth of a toroidal attractor. In (g), by taking , the trajectories tend to a limit cycle, with , , and . Thus, by increasing and , the transition from torus towards cycle occurs.

In (h), for , , and , the attractor is chaotic, because , , , and . Observe that, when compared to (d), by increasing , a transition from regular oscillation towards chaotic solution can occur. For these values of and , there is chaos if . In (i), by taking , the attractor comes back to be a limit cycle, because , , and . Thus, by increasing the values of and , a bifurcation related to the transition from chaos to periodic behavior takes place.

Starting from a stationary solution, regular oscillations can be created by increasing (please see (a) → (b) and (c) → (f)), or (see (c) → (d)), or (see (c) → (e)). From a periodic behavior, chaos can appear by a similar way (see (d) → (h)). Thus, by enhancing the excitatory connections, the activities of both populations can regularly or irregularly oscillate. However, when the strength of any excitatory connection is “too high,” the linear approximation of can no longer be valid. In this scenario, a suitable approximation is (because when ). Therefore, from (2), and as the time passes. For instance, by simulating the system with , , and , the variables asymptotically converge to and .

It is important to stress that the attractors and bifurcations reported in this section were already found in earlier works based on computer simulations on two-population models [10, 1315]. We hope that the analytical and numerical results presented here can guide experimental researches on the detection of such attractors and bifurcations in actual neural networks, in particular the ones underlying working memory, as discussed in the next section.

4. Discussion

In this work, we explicitly derived analytical conditions concerning the stability of the steady state derived from the approximation . Now, we discuss the possible relevance of this result.

Neural codes based on chaotic activity have been proposed [2729]. We observed here that chaos can be found in a simple neuronal assembly composed of two coupled Wilson-Cowan populations subject to constant stimuli, when the strengths of the excitatory connections are above critical numbers. Since excitatory synapses are more often encountered than inhibitory ones [17], then these strengths can naturally be above such critical numbers. Thus, the irregular activity commonly found in EEG would be, at least partially, a simple consequence of the neuronal connectivity, in particular, the predominance of excitatory synapses.

The strengths of excitatory connections, however, cannot be “too high,” because in this case and the activity would tend to be stationary. Thus, the balance between excitation and inhibition must satisfy constraints assuring normal oscillatory behavior. Abnormalities in this balance in the cortical circuitry have been associated with neurological disorders, such as autism and epilepsy [30, 31].

The conditions presented in Section 2 can guide laboratory experiments, in order to verify their validity. Observe that these analytical expressions were derived by supposing that . It is far from being trivial to find the regions in the 14-dimensional parameter space where this approximation holds. However, from (1) with , note that implies . Since only if , then in the steady-state condition. Hence, in the numerical simulations, we took (obviously, in (1) is equivalent to and in (2)). Neuronal populations with “slow” exponential decay (i.e., ) can be responsible for supporting working memory [32]. Hence, the validity of our conclusions could be first tested in such populations. Perhaps, these tests can reveal the true nature of the attractors and bifurcations involved in maintaining and manipulating new information.

Competing Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

Lucas L. Neves is partially supported by CAPES and Luiz H. A. Monteiro by CNPq.

References

  1. A. K. Engel, P. König, and T. B. Schillen, “Why does the cortex oscillate?” Current Biology, vol. 2, no. 6, pp. 332–334, 1992. View at: Publisher Site | Google Scholar
  2. E. Basar, C. Basar-Eroglu, S. Karakas, and M. Schürmann, “Oscillatory brain theory: a new trend in neuroscience: the role of oscillatory processes in sensory and cognitive functions,” IEEE Engineering in Medicine and Biology Magazine, vol. 18, no. 3, pp. 56–66, 1999. View at: Publisher Site | Google Scholar
  3. W. J. Freeman, “Understanding perception through neural ‘codes’,” IEEE Transactions on Biomedical Engineering, vol. 58, no. 7, pp. 1884–1890, 2011. View at: Publisher Site | Google Scholar
  4. E. M. Dewan, “Nonlinear oscillations and electroencephalography,” Journal of Theoretical Biology, vol. 7, no. 1, pp. 141–159, 1964. View at: Publisher Site | Google Scholar
  5. E. N. Lorenz, “Deterministic nonperiodic flow,” Journal of the Atmospheric Sciences, vol. 20, no. 2, pp. 130–141, 1963. View at: Publisher Site | Google Scholar
  6. R. C. deCharms and A. Zador, “Neural representation and the cortical code,” Annual Review of Neuroscience, vol. 23, pp. 613–647, 2000. View at: Publisher Site | Google Scholar
  7. H. R. Wilson and J. D. Cowan, “Excitatory and inhibitory interactions in localized populations of model neurons,” Biophysical Journal, vol. 12, no. 1, pp. 1–24, 1972. View at: Publisher Site | Google Scholar
  8. R. M. Borisyuk and A. B. Kirillov, “Bifurcation analysis of a neural network model,” Biological Cybernetics, vol. 66, no. 4, pp. 319–325, 1992. View at: Publisher Site | Google Scholar | Zentralblatt MATH
  9. P. E. Latham, B. J. Richmond, P. G. Nelson, and S. Nirenberg, “Intrinsic dynamics in neuronal networks. I. Theory,” Journal of Neurophysiology, vol. 83, no. 2, pp. 808–827, 2000. View at: Google Scholar
  10. Y. Hayashi, “Numerical bifurcation analysis of an oscillatory neural network with synchronous/asynchronous connections,” Neural Computation, vol. 6, no. 4, pp. 658–667, 1994. View at: Publisher Site | Google Scholar
  11. D. Wang, “Emergent synchrony in locally coupled neural oscillators,” IEEE Transactions on Neural Networks, vol. 6, no. 4, pp. 941–948, 1995. View at: Publisher Site | Google Scholar
  12. G. B. Ermentrout and N. Kopell, “Oscillator death in systems of coupled neural oscillators,” SIAM Journal on Applied Mathematics, vol. 50, no. 1, pp. 125–146, 1990. View at: Publisher Site | Google Scholar | Zentralblatt MATH | MathSciNet
  13. G. N. Borisyuk, R. M. Borisyuk, A. I. Khibnik, and D. Roose, “Dynamics and bifurcations of two coupled neural oscillators with different connection types,” Bulletin of Mathematical Biology, vol. 57, no. 6, pp. 809–840, 1995. View at: Publisher Site | Google Scholar
  14. T. Ueta and G. Chen, “On synchronization and control of coupled Wilson-Cowan neural oscillators,” International Journal of Bifurcation and Chaos in Applied Sciences and Engineering, vol. 13, no. 1, pp. 163–175, 2003. View at: Publisher Site | Google Scholar | Zentralblatt MATH | MathSciNet
  15. Y. Maruyama, Y. Kakimoto, and O. Araki, “Analysis of chaotic oscillations induced in two coupled Wilson-Cowan models,” Biological Cybernetics, vol. 108, no. 3, pp. 355–363, 2014. View at: Publisher Site | Google Scholar | Zentralblatt MATH | MathSciNet
  16. L. H. A. Monteiro, M. A. Bussab, and J. G. Chaui Berlinck, “Analytical results on a Wilson-Cowan neuronal network modified model,” Journal of Theoretical Biology, vol. 219, no. 1, pp. 83–91, 2002. View at: Publisher Site | Google Scholar | MathSciNet
  17. C. Beaulieu and M. Colonnier, “A laminar analysis of the number of round-asymmetrical and flat- symmetrical synapses on spines, dendritic trunks, and cell bodies in area 17 of the cat,” Journal of Comparative Neurology, vol. 231, no. 2, pp. 180–189, 1985. View at: Publisher Site | Google Scholar
  18. H. G. Schuster and P. Wagner, “A model for neuronal oscillations in the visual cortex-1. Mean-field theory and derivation of the phase equations,” Biological Cybernetics, vol. 64, no. 1, pp. 77–82, 1990. View at: Publisher Site | Google Scholar
  19. M. V. Tsodyks, W. E. Skaggs, T. J. Sejnowski, and B. L. McNaughton, “Paradoxical effects of external modulation of inhibitory interneurons,” The Journal of Neuroscience, vol. 17, no. 11, pp. 4382–4388, 1997. View at: Google Scholar
  20. E. Ledoux and N. Brunel, “Dynamics of networks of excitatory and inhibitory neurons in response to time-dependent inputs,” Frontiers in Computational Neuroscience, vol. 5, article 25, 2011. View at: Publisher Site | Google Scholar
  21. A. Muir, “A simple case of the Wilson-Cowan equations,” Biophysical Journal, vol. 27, no. 2, pp. 267–275, 1979. View at: Publisher Site | Google Scholar
  22. J. Argyris, G. Faust, M. Haase, and R. Friedrich, An Exploration of Dynamical Systems and Chaos, Springer, Berlin, Germany, 2nd edition, 2015. View at: MathSciNet
  23. L. H. A. Monteiro, “Overview of dynamical systems and chaos,” in Chaotic Signals in Digital Communications, M. Eisencraft, R. Attux, and R. Suyama, Eds., pp. 83–109, CRC Press, Boca Raton, Fla, USA, 2014. View at: Google Scholar
  24. H. R. Wilson, Spikes, Decisions, and Actions: The Dynamical Foundations of Neuroscience, Oxford University Press, Oxford, UK, 1999.
  25. K. Ogata, Modern Control Engineering, Prentice Hall, Upper Saddle River, NJ, USA, 2009.
  26. A. Wolf, J. B. Swift, H. L. Swinney, and J. A. Vastano, “Determining Lyapunov exponents from a time series,” Physica D. Nonlinear Phenomena, vol. 16, no. 3, pp. 285–317, 1985. View at: Publisher Site | Google Scholar | MathSciNet
  27. I. Tsuda, “Toward an interpretation of dynamic neural activity in terms of chaotic dynamical systems,” Behavioral and Brain Sciences, vol. 24, no. 5, pp. 793–810, 2001. View at: Publisher Site | Google Scholar
  28. H. Korn and P. Faure, “Is there chaos in the brain? II. Experimental evidence and related models,” Comptes Rendus—Biologies, vol. 326, no. 9, pp. 787–840, 2003. View at: Publisher Site | Google Scholar
  29. M. V. Kiselev, “Homogenous chaotic network serving as a rate/population code to temporal code converter,” Computational Intelligence and Neuroscience, vol. 2014, Article ID 476580, 8 pages, 2014. View at: Publisher Site | Google Scholar
  30. O. Yizhar, L. E. Fenno, M. Prigge et al., “Neocortical excitation/inhibition balance in information processing and social dysfunction,” Nature, vol. 477, no. 7363, pp. 171–178, 2011. View at: Publisher Site | Google Scholar
  31. C. E. Stafstrom and T. A. Benke, “Autism and epilepsy: exploring the relationship using experimental models,” Epilepsy Currents, vol. 15, no. 4, pp. 206–210, 2015. View at: Publisher Site | Google Scholar
  32. Z. Tiganj, M. E. Hasselmo, and M. W. Howard, “A simple biophysically plausible model for long time constants in single neurons,” Hippocampus, vol. 25, no. 1, pp. 27–37, 2015. View at: Publisher Site | Google Scholar

Copyright © 2016 L. L. Neves and L. H. A. Monteiro. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Related articles

No related content is available yet for this article.
 PDF Download Citation Citation
 Download other formatsMore
 Order printed copiesOrder
Views2148
Downloads987
Citations

Related articles

No related content is available yet for this article.

Article of the Year Award: Outstanding research contributions of 2020, as selected by our Chief Editors. Read the winning articles.