Compositionality of arm movements can be realized by propagating synchrony
- 715 Downloads
- 2 Citations
Abstract
We present a biologically plausible spiking neuronal network model of free monkey scribbling that reproduces experimental findings on cortical activity and the properties of the scribbling trajectory. The model is based on the idea that synfire chains can encode movement primitives. Here, we map the propagation of activity in a chain to a linearly evolving preferred velocity, which results in parabolic segments that fulfill the two-thirds power law. Connections between chains that match the final velocity of one encoded primitive to the initial velocity of the next allow the composition of random sequences of primitives with smooth transitions. The model provides an explanation for the segmentation of the trajectory and the experimentally observed deviations of the trajectory from the parabolic shape at primitive transition sites. Furthermore, the model predicts low frequency oscillations (<10 Hz) of the motor cortex local field potential during ongoing movements and increasing firing rates of non-specific motor cortex neurons before movement onset.
Keywords
Motor cortex Compositionality Synfire chains LFP Spike synchrony Motor control1 Introduction
Motor learning can be tedious. The optimisation of the interconnected processes involves the adaptation of internal models, the prediction of external rewards and the exploration of synergistic patterns of muscle activations. Often it occurs on time scales comparable to the life span and can be achieved only on the cost of the plasticity of the available motor actions. In order to reconcile the complexities of behavioural learning with the necessary flexibility, smoothness and efficiency of the movements, evolution has brought about a compositional approach that allows behavior to be generated from a set of possibly highly optimized elements. All movements that are possible for the organism can brought about by a combination of elementary movements, or primitives, however movement efficiency is improved by concatenating well-matched primitives in a purposeful manner.
The existence of behavioural primitives is supported by a number of experimental studies. For example, force field primitives that can be combined vectorially have been identified by stimulating the frog’s spinal cord (Bizzi et al. 1991, 2008; Mussa-Ivaldi et al. 1994; Mussa-Ivaldi and Bizzi 2000). Hart and Giszter (2010) revealed the neuronal basis of these motor primitives in the spinal cord. Recently, movement primitives have been identified in a free monkey scribbling task on a two dimensional plane (Polyakov et al. 2009a, b). Segments of constant acceleration constitute the parabolic movement primitives which satisfy the two-thirds power law of the end-effector dynamics observed in experiments (Lacquaniti et al. 1983; Viviani and Flash 1995).
An obvious question arising within this approach is how primitives are represented in a dynamic network of spiking neurons. One hypothesis is based on feed-forward sub-networks known as synfire chains (Abeles 1991). These structures were originally postulated to explain the occurrence of precise spike timing in cortical neurons (e.g. Eckhorn et al. 1988; Abeles et al. 1993; Prut et al. 1998). The general concept of feed-forward networks has recently been reviewed by Kumar et al. (2010) with respect to the transmission of information. Although there is substantial indirect evidence for synfire chains, a direct proof of their existence is has yet to be established due to the difficulties of determining such fine grained connectivity in neural tissue. However, methods to detect their activity continue to improve (Abeles and Gat 2001; Schrader et al. 2008; Berger et al. 2010) and theoretical and computational studies have shown that such structures would support the stable propagation of activity under quite general conditions (Herrmann et al. 1995; Diesmann et al. 1999; Gewaltig et al. 2001). Bienenstock (1995) proposed that each primitive is represented by an individual synfire chain and that the synchronization of activity between different chains binds the associated primitives into a composed object. A further study (Bienenstock 1996) elaborated on the relevance of the concept of compositionality for brain processing. Since then, it has been shown that synfire chains can be synchronized (Arnoldi and Brauer 1996) and that this mechanism can indeed realize compositionality (Hayon et al. 2005; Abeles et al. 2004; Schrader et al. 2010).
In the present study, we focus on compositionality in the sense of how behavioral primitives can be concatenated into sequences. It has been shown that networks of synfire chains can generate sequences in the context of abstract primitives (Schrader et al. 2010) or syllables in bird song (Li and Greenside 2006; Jin et al. 2007; Glaze and Troyer 2008; Jin 2009; Hanuschkin et al. 2010b). We present a functional model that is simultaneously capable of reproducing several experimental findings on cortical activity and generating trajectories which exhibit key features of free monkey scribbling. Our model consists of a topologically organized network of synfire chains. Neurons in the same pool of a chain encode the same preferred velocity vector, thus realizing a population coding for movement (Georgopoulos et al. 1982, 1986b). The neural activity is characterized by asynchronous and irregular dynamics (Burns and Webb 1976; Softky and Koch 1993; van Vreeswijk and Sompolinsky 1996; Ponce-Alvarez et al. 2010), but due to the synfire activity, precise interspike timing and patterns can also be observed during ongoing motion and locked to relevant task features, as have been found experimentally in the motor cortex (Riehle et al. 1997, 2000; Shmiel et al. 2005, 2006; Ghosh et al. 2009; Putrino et al. 2010). The topological arrangement of our network is supported by the finding that correlation strength between neurons decreases as the distance between them increases (Murthy and Fetz 1996; Dombeck et al. 2009), and that nearby neurons tend to prefer similar motion parameters (Georgopoulos et al. 2007; Stark et al. 2009). The trajectories generated by our model consist of a series of parabolic segments similar to those identified experimentally (Polyakov et al. 2009a, b) which fulfill the well established two-thirds power law relationship of velocity and curvature (Lacquaniti et al. 1983; Viviani and Flash 1995). It has previously been shown that population vectors of neural activity also fulfill this relationship (Schwartz 1994).
The paper is organized as follows. In Section 2, we summarize how stable activity propagation in a feed-forward network of spiking neurons can be mapped to parabolic movement primitives in position space. In Section 3, we present our model architecture and determine optimal parameters to enable long sequences of primitives. We then derive experimental predictions for the signature of synfire chain based composition of trajectories in collective signals of neural activity such as the local field potential (LFP). We analyze the characteristics of an example model trajectory and demonstrate that it shares key properties with trajectories generated by monkeys in a scribbling task. Finally, in Section 4 we discuss the model findings, limitations and predictions.
Preliminary results of this model study on free monkey scribbling have been published in abstract form (Hanuschkin et al. 2009a, b).
2 Materials and methods
2.1 Mapping uniform motion to parabolic motion
2.2 Propagation of activity in cell assemblies
Summary of model structure after Nordlie et al. (2009)
Model summary | |
---|---|
Populations | One network of interconnected synfire chains (SFCN), one backward-and-forward connected chain (BFCN) |
Connectivity | SFCN: excitatory feed-forward (FF) connections within each chain and between final and initial groups of selected pairs of chains, cross-inhibition between chains, global random inhibition. BFCN: FF and feed-back (FB) connections. SFCN to BFCN: random inhibition. BFCN to SFCN: FF connections from final group of BFCN to initial group of chain 1 in SFCN. All connections realized using random divergent (RD) or random convergent (RC) wiring. |
Neuron model | Leaky integrate-and-fire (IaF), fixed voltage threshold, fixed absolute refractory time |
Synapse model | α-current inputs |
Input | Independent fixed-rate Poisson spike trains to all neurons |
Measurements | Spike activity, membrane potential |
Populations |
Name | Elements | Size |
---|---|---|
SFCN | Synfire chain SFC_{j} | 10 |
BFCN | Backward-and-forward connected chain BFC | 1 |
SFC_{j} | SFC groups G_{i,j} | 50 |
BFC | BFC groups \(G_{k}^{*}\) | 50 |
G_{i,j} | Neuron populations E_{i,j}, I_{i,j} | 2 populations per group G_{i,j}: E_{i,j} and I_{i,j} |
\(G_{k}^{*}\) | Neuron populations \({\rm E}_{k}^{*}\), \({\rm I}_{k}^{*}\) | 2 population per group \(G_{k}^{*}\): \({\rm E}_{k}^{*}\) and \({\rm I}_{k}^{*}\) |
E_{i,j} , \({\rm E}_{k}^{*}\) | Excitatory IaF neuron | 100 |
\(\rm I_{\it i,j}\) , \(\rm I_{\it k}^{*}\) | Inhibitory IaF neuron | 25 |
Connectivity |
Name | Source | Target | Pattern |
---|---|---|---|
FF | E_{i,j} | E_{i + 1,j} + I_{i + 1,j} | RD, 1→C_{Ex}, weight J_{E}, delay d |
Ch_{conn} | E_{50,j} | E_{1,j′} + I_{1,j′}, with j,j′ as Fig. 4 | RC, C_{Ex}→1, weight J_{E}, delay d |
I_{global} | I_{i,j} | SFC_{j} | RD, 1→k_{g}, weight J_{I}, delay d |
I_{cross} | I_{i,j} | \({\rm SFC}_{j^{'}}\), with j,j′ as Fig. 4 | RD, 1→k_{c}, weight J_{I}, delay d |
I_{BFC} | I_{i,j} | BFC | RD, 1→k_{B}, weight J_{I}, delay d |
FF^{*} | \({\rm E}_{k}^{*}\) | \({\rm E}_{k+1}^{*}+{\rm I}_{k+1}^{*}\) | RD, 1→C_{Ex}, weight J_{E}, delay d |
FB^{*} | \({\rm E}_{k}^{*}\) | \({\rm E}_{k-1}^{*}+{\rm I}_{k-1}^{*}\) | RD, 1→C_{Ex}, weight J_{E}, delay d |
\(I_{\mathrm{global}}^{*}\) | \({\rm I}_{k}^{*}\) | BFC | RD, \(1\rightarrow{k_{\mathrm{g}}^{*}}\), weight J_{I}, delay d |
\(E_{\mathrm{SFC}}^{*}\) | \({\rm E}_{50}^{*}\) | SFC_{1,1} | RC, C_{Ex}→1, weight J_{E}, delay d |
2.3 Mapping activity in cell assemblies to parabolic motion
2.4 Sequences of primitives
2.5 Analysis of the trajectories
2.6 Numerical simulations
Summary of model dynamics after Nordlie et al. (2009)
Neuron models | |
---|---|
Name | IaF neuron |
Type | Leaky integrate-and-fire, α-current input |
Subthreshold dynamics | \( \tau_{m} \dot{V}(t) = -V(t) + R I(t) \quad \text{if} \quad t > t^{*}+\tau_{\text{ref}} \) |
\( V(t) = V_{0} {\kern34pt}\qquad \text{else}\) | |
\( I(t) = I_{0} + \hat{i}\dfrac{e}{\tau_{\alpha}}\displaystyle\sum\nolimits_{\tilde{t}} e^{\frac{t-\tilde{t}}{\tau_\alpha}} \delta(t-\tilde{t}) \) | |
Spiking | If V(t − ) < V_{th} ∧ V(t + ) ≥ V_{th} |
1. calculate retrospective threshold crossing with bisectioning method (Hanuschkin et al. 2010c) | |
2. set \(t^{*}=t+\Delta_{\mathrm{of\/fset}}\) | |
3. emit spike with time stamp t^{*} | |
Input |
Type | Target | Description |
---|---|---|
Poisson generator | SFCN, BFCN | Independent for all targets, rate ν_{x}, weight J_{E} |
Measurements | ||
Spike activity of all neurons, membrane potential of neurons of SFC_{j} with j ∈ (1,2,7) |
Specification of model parameters
Name | Value | Description |
---|---|---|
Connectivity | ||
C_{Ex} | 93 | Number of feed-forward connections from each excitatory neuron |
k_{g} | 7 | Number of outgoing global connections from each inhibitory neuron |
k_{c} | 19 | Number of cross connections from inhibitory neuron to each competing SFC_{j} |
k_{B} | 6 | Number of connections from each inhibitory neuron to the BFCN |
\(k_{\mathrm g}^{*}\) | 0 | Number of connections from each inhibitory neuron in the BFCN |
J_{E} | 20.68 pA | Amplitude of excitatory connection, \(\Rightarrow\)0.1 mV EPSP amplitude |
J_{I} | − 124.68 pA | Amplitude of inhibitory connection, \(\Rightarrow\)− 0.6 mV EPSP amplitude |
d | 1.5 ms | Synaptic transmission delay |
Neuron model | ||
τ_{m} | 20 ms | Membrane time constant |
C_{m} | 250 pF | Membrane capacitance |
V_{th} | 20 mV | Fixed firing threshold |
V_{0} | 0 mV | Resting potential |
V_{reset} | 0 mV | Reset potential |
τ_{ref} | 2 ms | Absolute refractory period |
τ_{α} | 0.5 ms | Rise time of post-synaptic current |
Input | ||
ν_{x} | 7.7 kHz | External Poisson rate |
Simulations were performed with NEST revision 8257 (see www.nest-initiative.org and Gewaltig and Diesmann 2007) using a computational step size of 1 ms on a standard workstation running Linux. To avoid synchrony artefacts (Hansel et al. 1998), we employ precise simulation techniques in a globally time driven framework (Morrison et al. 2007b; Hanuschkin et al. 2010c). To allow other researchers to perform their own experiments, at the time of publication we are making a module available for download at www.nest-initiative.org containing all relevant scripts.
3 Results
3.1 Network model architecture
In strongly recurrently connected networks of spiking neurons synfire activity can be ignited spontaneously (Tetzlaff et al. 2002). Although this is an unwanted effect in studies investigating embedded synfire chains in balanced recurrent networks as it limits the density of local feed forward structures, we exploit this property in our model. We create a synfire chain with feed-backward as well as feed-forward connections, both with a dilution factor of p. This backward-and-forward connected chain, or BFC, constitutes the second network (BFCN). One end of the BFC makes excitatory feed-forward connections with dilution factor p to the initial pool of chain 1 in SFCN. Each inhibitory neuron in SFCN makes k_{B} connections to neurons randomly selected from BFCN, thus inhibiting its activity when synfire activity is present. If synfire activity is extinguished, the drop in inhibition causes a self-ignition in the unstable BFC, which in turn triggers a fresh wave of activity in chain 1. Thus the recurrent connections between SFCN and the BFCN ensure sustained activity. The dynamics of the BFCN and of the interaction with SFCN are investigated in Section 3.3. The arbitrary choice of chain 1 to re-ignite the SFCN is made here for the sake of simplicity; a more complex interaction could be assumed which would allow re-ignition of the SFCN at the beginning of any chain.
The scaling of inhibitory synapses with respect to excitatory synapses and the rate of the external excitatory Poisson input to each neuron in SFCN are chosen such that in the absence of synfire activity, the network spikes in the asynchronous irregular (AI) regime (Brunel 2000). A tabular description of our model is given in Tables 1 and 2; unless otherwise stated, model parameters are as given in Table 3.
3.2 Competition between synfire chains
Each of the chains in SFCN represents a parabolic movement primitive. To produce a series of primitives, it is necessary that activity reliably propagates from one chain to exactly one of multiple (here two) potential successor chains at the vertices of the network graph (see Fig. 4). In our model, cross-inhibition realizes this switching between two simultaneously activated and competing chains. We investigate two approaches to achieve reliable switching. In Section 3.2.1, cross-inhibition is structured such that synchronous activity in each pool directly inhibits the activity in the next pool of the competitor chain. This approach is motivated by the idea of synfire binding (Abeles et al. 2004; Hayon et al. 2005; Schrader et al. 2010), in which two simultaneously active chains can bind a third chain due to structured excitation. In an alternative approach in Section 3.2.2, the cross-inhibition is unstructured. Synfire chain competition relying solely on global inhibition has recently been proposed by Chang and Jin (2009). However, in their study the synfire chain activity is ‘driven’: a suprathreshold driving input is combined with dominant global inhibition. In contrast, our model exhibits activity in the asynchronous irregular regime due to balanced global inhibition (van Vreeswijk and Sompolinsky 1996) and only exhibits synfire activity if the initial pool of a chain receives additional stimulation. Due to our different activity regime, additional assumptions on the inhibition between chains need to be made to realize reliable switching.
3.2.1 Competition by structured cross-inhibition
To demonstrate the switching and its effects on collective signals such as average firing rate or local field potential (LFP), we change the inter-chain connectivity such that only chains 1, 2 and 7 can be activated. A reduced network which still exhibits synfire chain competition is provided by the following connectivity: the final pool of chain 1 is connected to the initial pools of chains 2 and 7, the final pools of these chains are connected back to the initial pool of chain 1. Chains 2 and 7 are mutually cross inhibited as shown in Fig. 6. A reduced network with no synfire chain competition is realized by connecting chains 1, 2 and 7 in a cyclical fashion but retaining the structured cross-inhibition between chains 2 and 7. Both reduced networks are activated by an external Gaussian pulse packet to the initial group of chain 1. We calculate the average firing rate (Gaussian smoothing kernel with σ = 1 ms) and two approximations of the local field potential (LFP) of the network. A first approximation of the LFP is given by the average membrane potential of 10% of the neurons in the network (Ursino and Cara 2006). A more sophisticated approximation of the LFP (in arbitrary units) is given by calculating the sum of the absolute values of the excitatory and inhibitory postsynaptic currents (PSCs) in 1 ms steps and smoothing the result with a Gaussian filter with σ = 1 ms (Mazzoni et al. 2008).
Figure 8(c) shows the probability of activating both successor chains and Fig. 8(d) illustrates the probability of neither of the successor chains as functions of k_{g} and k_{c}. As the cross-inhibition parameter k_{c} and the global inhibition parameter k_{g} increase, the probability of activating both successor chains decreases whilst the probability that neither chain is activated increases. The cross-inhibition parameter k_{c} directly influences the competition of the chains and so has a stronger effect on the switching behavior than the global inhibition k_{g} parameter, which regulates the overall activity in the network. A good choice for the number of inhibitory connections is k_{c} = 7 and k_{g} = 7: for this configuration, the probability of activating both successor chains is \(p_{\mathrm{2}}=5.44\mbox{\%}\pm2.26\), whereas the probability that neither chain is activated is p_{0} = \(4.98\mbox{\%}\pm1.85\). The means and standard deviations for these two values are calculated by performing the 100 switching trials for 100 different network realizations. This configuration exhibits a reasonable compromise between reliable switching and stability. The probability of activating two chains can be reduced to zero by increasing k_{c}, but at the cost of increasing the probability of activating neither chain. Likewise, the probability of activating neither chain reaches zero for small k_{c}, but the probability of activating both chains increases. For a given realization of the network a bias towards selecting chain 2 or chain 7 can be observed, however on average there is no bias due to the symmetry of the randomly chosen connections (data not shown).
3.2.2 Competition by unstructured cross-inhibition
As in Fig. 7(b), the average firing rate in Fig. 10(b) increases whenever two chains are competing, returning to its initial value after the activity of the losing chain has died away. The time of cell assembly competition is also clearly visible in both LFP approximations as indicated by the red and blue arrows. Figure 10(c) shows the spiking activity for the corresponding network without synfire chain competition. As for the structured cross-inhibition case, Fig. 10(d) illustrates that the effects marked by arrows in Fig. 10(b) do not occur in the absence of competition between the synfire chains. In this case, a step-like modulation is visible for the average membrane potential as well as the approximated LFP. These steps are not observed during simulations of the full network model, as each chain competes with some other chain and thus all chains generate the same excitatory and inhibitory activity. However, they might be observable in a less symmetric model where chains may have differing numbers of potential successor chains.
3.2.3 Competition between multiple chains
However, in most natural movement scenarios the choice of the next action is unlikely to be truly random, but influenced by the existence of imprinted repetitive movement sequences or by priming from motor planning and control areas. In order to illustrate the effect of a very simple priming mechanism on the reliability of switching, we apply an additional Poisson input of 1 kHz with variable synaptic weight J_{prim} to all neurons of the first 10 pools of one successor (chain 2) in a network with four potential successor chains. The inhibition parameters are chosen as k_{c} = 25 and k_{g} = 7 (see red cross in Fig. 12(d)). The conditional probability of switching to chain 2, given a correct switch to exactly one of the successor chains, is shown in Fig. 12(e) as a function of J_{prim}. The conditional probability of switching to chain 2 increases with increasing excitatory priming and decreases with increasing inhibitory priming. It can be fitted by a sigmoidal function (\(f(x)=100/\left(1+e^{-a(x+b)}\right)\) where a = 1.8 and b = 0.4).
The total error probability decreases with increasing excitatory or inhibitory priming and is maximal when the network is operating at chance level, i.e. the probability of switching to any of the four successor chains is 25%. The chance level shown in Fig. 12(e) is estimated from the fitted function and is slightly shifted from the unprimed case towards inhibitory priming with J_{prim} = − 0.98 pA. This is due to the random network connectivity which results in minor asymmetries in the switching probabilities. Specifically, the probability that no successor chain is chosen is maximal at chance level as the symmetry between the activity in all potential successor chains increases the likelihood that no winner emerges. Applying inhibitory or excitatory priming introduces a bias into the competition, thus making it more likely that activity will propagate in at least one chain. Conversely, the probability that multiple chains are activated is approximately constant for excitatory priming and increases with increasing inhibitory priming. This effect is due to the general reduction of activity in chain 2, which results in less global inhibition in the network and thus a greater probability of activity propagating in more than one chain (see also Fig. 11(a)). At around J_{prim} > 4 pA the general increase of activity in chain 2 enables the spontaneous triggering of synfire activity (Tetzlaff et al. 2002). In this network regime, reliable switching is no longer possible and so the total error probability increases. Fig. 12(f) shows the case of four chain switching with excitatory priming to chain 2 (J_{prim} = 3 pA). The introduction of priming results in a large increase of the working regime (compare Fig. 12(d)). For the rest of the manuscript we restrict our analysis to the case of two successor chains with no priming, as this is sufficient to model the key features of the behavior of a highly trained and motivated monkey in the free scribbling task.
3.3 Activity state transition in the BFC
As can be seen in Fig. 14(b), synfire volleys caused by spontaneous self-ignition tend to be of limited duration. Activity volleys traveling in different directions cancel each other when they meet and volleys reaching either end of the BFCN are not reflected. Furthermore, the high activity in the BFCN during synfire activity results in strong global inhibition and a reset of nearly all neurons, which in turn decreases the probability of self-ignition until activity has built up again. However, if the pool size is chosen sufficiently large (e.g. 175 neurons for \({k_{\mathrm{g}}^{*}}=0\) and f_{ex} = 7.7 kHz), a single self-ignition results in ongoing pathological high firing rates.
3.4 Sustained activity
3.5 Generation of scribbling trajectories
4 Discussion
In this study we have demonstrated that activity in synfire chains can realize the generation of experimentally observed parabolic segments that fulfill the two-thirds power law (Viviani and Flash 1995; Polyakov et al. 2009b). The key insights are that uniform linear motion in velocity space maps to parabolic motion in position space and that the propagation of a wave of activity in a synfire chain is an ideal neural mechanism for a process characterized by uniform linear motion. We further show that a network of synfire chains can produce on-going trajectories consisting of series of parabolic segments with smooth transitions among them. Necessary assumptions for this to happen are that the terminal pools of the synfire chains have a similar preferred velocity as the initial pools of the chains to which they connect to, and that the switching mechanism that selects an appropriate successor chain is reliable. A final aspect of the model is that the extinction of synfire activity triggers a mechanism by which it is re-ignited. In summary, by postulating an appropriate structure in a simulated network of biologically realistic neurons we have developed a model that reliably reproduces macroscopic properties of monkey scribbling. In the following sections we discuss the assumptions, limits and predictions of our model in greater detail.
4.1 Synfire chains
The choice of synfire chains as the chief computational element of the model results from a possible interpretation of the two-thirds power law as a linear propagation in velocity space. While other interpretations of the two-thirds power law are conceivable, the simplicity of the present assumption is not only theoretically appealing, but is also a good match to the known properties of synfire chains in maintaining an accurate representation of a constant progression (Haβ et al. 2008). The assumption of synfire chains as constituents of the model is also in good agreement with Fitts’ law, since the speed of the synfire activity can be reliably controlled, for example by the variation of global noise or by the modulation of neural threshold values (Wennekers and Palm 1996). The example studied here considers only five basic speeds and their one-parametric linear combinations. The model is obviously generalizable to a larger number of accelerating forces to the arm provided that the matching conditions between terminal speed of one chain and the initial speed of another one are observed. Even a continuous arrangement of synfire structures in form of a two-dimensional neural field (Amari 1977) is imaginable, but it should be noted that the data (Polyakov et al. 2009a, b) are largely representable by as little as three unidirectional chains. This impoverishment of the movement manifold might be due to continued optimization under constant experimental conditions and can be assumed to reflect a compromise between the goal of the task and the energy-efficient control of the arm. Thus, we might speculate that even in a more general framework, certain movements would be preferred to others. Consequently, a discrete representation, save for some extraneously encoded invariances, might be sufficient to account for planned optimized movements of the extremities.
4.2 Competition between synfire chains
We investigated two alternate switching mechanisms on the basis of mutual cross-inhibition (see Section 3.2). We have shown that both structured and unstructured cross-inhibition realize reliable switching, but that unstructured cross-inhibition is more effective at eliminating the occurrence of two successor chains being activated without substantially increasing the risk of activity dying away altogether. Additionally, the working regime in which reliable switching can be achieved is greater for unstructured than for structured cross-inhibition. Note that in this study, each inhibitory neuron has the same number of global and the same number of local outgoing connections. A finer tuning could be achieved by assuming that the numbers of outgoing inhibitory connections are drawn from distributions or not homogeneous along the length of a chain. The question of whether the assumption of structured or unstructured cross-inhibition is more realistic in the motor cortex cannot be answered with the experimental data currently available. In Section 3.2.3 we investigated the scaling properties of the switching mechanism by unstructured mutual inhibition in networks with greater numbers of potential successors. The working regime shrinks with increasing number of successor chains, however, even a very simple priming mechanism restores the working regime.
The two architectures investigated here are not the only possible candidates for a reliable switching mechanism. For example, an alternative architecture realizing synfire chain competition on the basis of a dominant global inhibition was recently proposed by Chang and Jin (2009). Their simplified model, with pulse coupling (i.e. zero delays) and constant superthreshold drive to the individual neurons ensures that only one neuron is active at a time (Jin 2002). In this case the robustness of the switching can be shown analytically. In a subsequent study, Jin (2009) showed that their proposed mechanism also enables robust synfire chain switching in a biologically realistic model of songbird HVC (high vocal center) nuclei. However, syllable-coding excitatory neurons in HVC exhibit sparse bursting behavior (Hahnloser et al. 2002) while the inhibitory interneurons fire with high frequencies (Dutar et al. 1998). This activity regime differs markedly from that of the motor cortex investigated in our model, which is characterized by on-going asynchronous irregular activity.
4.3 Ignition of synfire activity
Our model makes use of a backward-and-forward connected chain to ignite synfire chain activity in the SFCN at the beginning of the simulation and after switching failures cause the synfire activity to die out. We have characterized the dynamical properties of the BFCN, such as periodic ignition and self-extinction of synfire activity. The choice of the BFCN is not crucial for our model; clearly, a variety of network architectures could fulfil its role. The critical mechanism is the existence of an inhibitory/excitatory reciprocal connection between the networks, such that synfire activity in the SFCN inhibits activity in the other network, which in turn excites the SFCN when the inhibition is reduced. The advantage of the BFCN is that it creates a pulse packet of exactly the right width and amplitude to guarantee that synfire activity is triggered in the SFCN. We therefore include it to demonstrate that changes in the standard synfire chain architecture can lead to interesting functional properties that have not previously been investigated. For the sake of simplicity we restricted the model to a single BFC which can only ignite the first chain; in this study we are investigating a possible neuronal architecture that could underlie the observed segmentation of 2D movement trajectories, rather than the characteristics of movement initialization after a still period. However, the model could be extended to include several BFCs projecting to different first pools in order to investigate this aspect of behavior. The interconnection between BFCs and directional synfire chains can be learned by a realistic Hebbian rule as described in Haβ et al. (2008) for a somewhat simpler model.
4.4 Learning and synfire chains
Our model is predicated on the existence of synfire chains. Many studies have reported precise spike timing, for example in mammalian cortex (Eckhorn et al. 1988; Gray and Singer 1989; Prut et al. 1998; Abeles et al. 1993; Ikegaya et al. 2004; Pulvermüeller and Shtyrov 2009) or songbird HVC (Hahnloser et al. 2002; Kozhevnikov and Fee 2007), which suggests an underlying feed-forward connectivity. Moreover, a local convergent-divergent connectivity profile has been experimentally observed in HVC (Mooney and Prather 2005). It has also been shown experimentally that the learning of new motor tasks leads to changes in synaptic strength and the rapid formation of new synaptic connections (Rioult-Pedotti et al. 1998; Xu et al. 2009), however there is as yet no consensus on the mechanisms required for the brain to develop such structures. Although some modeling studies have reported the development of feed-forward sub-networks (Izhikevich et al. 2004; Buonomano 2005; Doursat and Bienenstock 2006; Jun and Jin 2007; Fiete et al. 2010) on the basis of Hebbian synaptic plasticity, studies of large-scale networks with biologically realistic numbers of synapses per neuron have not reproduced these findings (Morrison et al. 2007a; Kunkel et al. 2010). This highlights the need for further studies to determine the necessary and sufficient conditions for the divergent-convergent connectivity of synfire chains to develop.
The introduction of synaptic plasticity could also address another limitation of our model, namely that the transition probabilities between one primitive and the next are constant. Hebbian plasticity at the transitions between one chain and the next would imprint trajectories that are carried out more often (Hanuschkin et al. 2010a) whereas reward-modulated plasticity would imprint trajectories that are more likely to generate a reward (e.g. Izhikevich 2007; Farries and Fairhall 2007; Baras and Meir 2007; Legenstein et al. 2008; Potjans et al. 2009, 2010). As shown in Section 3.2.3 the probability of a switching failure increase with the number of successors. A reward can be externally applied, but it is also conceivable that the increase in the network rate caused when more than one potential successor chain is activated, or the rate decrease when no successor chain is activated, could be interpreted by the network as a negative reward signal. The effect of the reward-modulated plasticity in this case would be to decrease the variability (number of possible successors) in generated motor sequences.
The decision where to continue after reaching the end of a synfire chain or, considered here as equivalent, a movement primitive, might be biased by a number of factors such as movement intentions, visual feedback or biomechanical constraints. An alternative approach to generate specific rather than random trajectories is therefore to influence the switching mechanism by priming one potential successor chain, as investigated in Section 3.2.3. The priming signals can be generated by an additional network (Hanuschkin et al. 2010b) or even within the same network if neural populations with different time constants are assumed (Yamashita and Tani 2008). Experiments suggest that sequential information and directional coding are indeed coded within the same neuronal population (Carpenter et al. 1999; Ben-Shaul et al. 2004). A repeated application of priming to generate a specific movement sequence could induce the imprinting of the sequence through the Hebbian mechanisms discussed above, rendering future priming unnecessary. On a behavioral level this can be interpreted as the convergence of a motor pattern from a consciously controlled variable to a fully automatic execution of a stereotyped movement after intensive training (Sforza et al. 2000).
4.5 Predictions of the model
In our model study of free monkey scribbling we show that cell assembly competition generates low frequency oscillations in collective signals that approximate the LFP. In our simulations, all synfire chains are the same length and generate primitives with durations of about 140 ms, leading to competition-driven oscillations in the LFP at around 7 Hz. In experiments the median drawing time of a parabolic movement is 250 ms; our model therefore predicts competition-driven oscillations in the LFP around 4 Hz. This component is unlikely to be as well-defined as in our simplified model, as feed-forward structures embedded in real cortical tissue are likely to be overlapping and of varying lengths, masking the effect of low LFP oscillations. Additionally, when the switching is biased due to priming or learned asymmetries in the connections to potential successor chains, the degree of competition is lowered. This suggests that the low frequency component of the LFP will decrease as training progresses, as stereotypical movements are imprinted or priming is activated. Experimental studies have shown that the low frequency LFP (<13 Hz) is suitable for extracting motion direction of stereotyped movements by investigating the movement evoked potential (mEP) ((Rickert et al. 2005); O’Leary and Hatsopoulos 2006), but so far studies investigating oscillations during continuous movements have focussed on the beta to gamma band (∼15 − 90 Hz) rather than the low frequency band (Donoghue et al. 1998). In our model the approximated LFP signal contains no information about the direction of movement, as the neurons are not spatially organized and each neuron contributes equally to the signal. However, in future studies this simplification could be relaxed in order to take the spatial extent of an LFP signal into consideration, as could the high degree of symmetry which is unlikely to occur in a biological system.
Along with the effect of introducing low frequency components to the LFP, competition between cell assemblies also produces deviations from the pure parabolic trajectories at the transition sites between movement primitives. This effect can indeed also be seen in the analysis of the experimental movement trajectories: in Fig. 4(a) of Polyakov et al. (2009b), deviations from the κ = 0 line can be observed that may indicate transition points between movement primitives. Our model predicts that these deflections will decrease with training, as training reduces the competition by priming or imprinting the sequence of cell assemblies. As our model does not include any priming or learning during the generation of movement sequences, the length of the trajectories follows the geometric distribution \(P(n)=p_{0}\left(1-p_{0}\right)^{n}\), where p_{0} is the constant probability that no successor chain is activated at a switching point (Section 3.5). This distribution is unlikely to occur in real experimental data, as a real monkey will exhibit varying levels of motivation during a trial. However, our experiments on priming suggest that the mean length of trajectories will increase with training, as the transitions from one primitive to the next become more reliable due to priming of the successor chain or strengthening of the transition connections (imprinting). To verify these predictions, further experimental studies on the development of the equi-affine properties and characteristic lengths of on-going trajectories with training could be carried out.
A final prediction of the model is the existence of directionally un-tuned neurons in the motor cortex that increase their firing rate immediately before movement onset (see Fig. 15). A possible signature of this population activity is the peak in the mEP observed just before movement onset (Rickert et al. 2005). Additional evidence to support this interpretation is that the peak amplitude decreases when the monkey can anticipate the movement to execute (Roux et al. 2006). In our model, this would correspond to inducing a new movement primitive through a priming signal rather than through ignition. To investigate these issues further, more sophisticated theoretical models of the generation of the LFP in a simulated network are required (Lindén et al. 2009a, b). Instead of high frequency oscillations serving to bind distributed cortical representations as proposed by Singer and Gray (1995), in our study composition is expressed in low frequency oscillations resulting from cell assembly interaction. We therefore suggest greater attention should be paid to low frequency components in experimental data and that data sets recorded from partially trained animals may be particularly helpful in revealing the mechanisms of motor compositionality.
Notes
Acknowledgements
Partially funded by DIP F1.2, BMBF Grants 01GQ0420/01GQ0432 to BCCNs Freiburg and Göttingen, EU Grant 15879 (FACETS), Helmholtz Alliance on Systems Biology (Germany), Next-Generation Supercomputer Project of MEXT (Japan) and the Junior Professor Program of Baden-Württemberg. The authors would like to cordially thank the members of the DIP F1.2 collaboration for their always stimulating and constructive discussions of this work, in particular T. Flash, F. Polyakov, M. Abeles, M. Teicher and T. Geisel. Finally, we would like to thank S. Schrader for his support at the outset of this project and M. Denker for useful comments on an earlier version of the manuscript.
Open Access
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
References
- Abbott, L. F. (1999). Lapicque’s introduction of the integrate-and-fire model neuron (1907). Brain Research Bulletin, 50(5–6), 303–304.PubMedCrossRefGoogle Scholar
- Abeles, M. (1991). Corticonics: Neural circuits of the cerebral cortex (1st ed.). Cambridge: Cambridge University Press.CrossRefGoogle Scholar
- Abeles, M., Bergman, H., Margalit, E., & Vaadia, E. (1993). Spatiotemporal firing patterns in the frontal cortex of behaving monkeys. Journal of Neurophysiology, 70(4), 1629–1638.PubMedGoogle Scholar
- Abeles, M., & Gat, I. (2001). Detecting precise firing sequences in experimental data. Journal of Neuroscience Methods, 107(1–2), 141–154.PubMedCrossRefGoogle Scholar
- Abeles, M., Hayon, G., & Lehmann, D. (2004). Modeling compositionality by dynamic binding of synfire chains. Journal of Computational Neuroscience, 17(2), 179–201.PubMedCrossRefGoogle Scholar
- Amari, S.-i. (1977). Dynamics of pattern formation in lateral-inhibition type neural fields. Biological Cybernetics, 27, 77–87.PubMedCrossRefGoogle Scholar
- Arnoldi, H. M., & Brauer, W. (1996). Synchronization without oscillatory neurons. Biological Cybernetics, 74(3), 209–223.PubMedCrossRefGoogle Scholar
- Baras, D., & Meir, R. (2007). Reinforcement learning, spike-time-dependent plasticity, and the BCM rule. Neural Computation, 19, 2245–2279.PubMedCrossRefGoogle Scholar
- Ben-Shaul, Y., Drori, R., Asher, I., Stark, E., Nadasdy, Z., & Abeles, M. (2004). Neuronal activity in motor cortical areas reflects the sequential context of movement. Journal of Neurophysiology, 91(4), 1748–1762 (comparative study).PubMedCrossRefGoogle Scholar
- Berger, D., Borgelt, C., Louis, S., Morrison, A., & Grün, S. (2010). Efficient identification of assembly neurons within massively parallel spike trains. Computational Intelligence and Neuroscience, 2010, 439648.Google Scholar
- Bienenstock, E. (1995). A model of neocortex. Network: Computational Neural Systems, 6, 179–224.CrossRefGoogle Scholar
- Bienenstock, E. (1996). Composition. In A. Aertsen & V. Braitenberg (Eds.), Brain theory—Biological basis and computational principles (pp. 269–300). Amsterdam: Elsevier.Google Scholar
- Bizzi, E., Cheung, V., d’Avella, A., Saltiel, P., & Tresch, M. (2008). Combining modules for movement. Brain Research Reviews, 57(1), 125–133. Networks in Motion.PubMedCrossRefGoogle Scholar
- Bizzi, E., Mussa-Ivaldi, F., & Giszter, S. (1991). Computations underlying the execution of movement: A biological perspective. Science, 253(5017), 287–291.PubMedCrossRefGoogle Scholar
- Brunel, N. (2000). Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons. Journal of Computational Neuroscience, 8(3), 183–208.PubMedCrossRefGoogle Scholar
- Buonomano, D. V. (2005). A learning rule for the emergence of stable dynamics and timing in recurrent networks. Journal of Neurophysiology, 94, 2275–2283.PubMedCrossRefGoogle Scholar
- Burns, B. D., & Webb, A. C. (1976). The spontaneous activity of neurones in the cat’s visual cortex. Proceedings of the Royal Society of London B, 194, 211–223.CrossRefGoogle Scholar
- Calabi, E., Olver, P. J., Shakiban, C., Tannenbaum, A., & Haker, S. (1998). Differential and numerically invariant signature curves applied to object recognition. International Journal of Computer Vision, 26(2), 107–135.CrossRefGoogle Scholar
- Calabi, E., Olver, P. J., & Tannenbaum, A. (1996). Affine geometry, curve flows, and invariant numerical approximations. Advances in Mathematics, 124(1), 154–196.CrossRefGoogle Scholar
- Carpenter, A. F., Georgopoulos, A. P., & Pellizzer, G. (1999). Motor cortical encoding of serial order in a context-recall task. Science, 283(5408), 1752–1757.PubMedCrossRefGoogle Scholar
- Chang, W., & Jin, D. Z. (2009). Spike propagation in driven chain networks with dominant global inhibition. Physical Review E, 79(5), 051917.CrossRefGoogle Scholar
- Diesmann, M., Gewaltig, M.-O., & Aertsen, A. (1999). Stable propagation of synchronous spiking in cortical neural networks. Nature, 402(6761), 529–533.PubMedCrossRefGoogle Scholar
- Dombeck, D. A., Graziano, M. S., & Tank, D. W. (2009). Functional clustering of neurons in motor cortex determined by cellular resolution imaging in awake behaving mice. Journal of Neuroscience, 29(44), 13751–13760.PubMedCrossRefGoogle Scholar
- Donoghue, J. P., Sanes, J. N., Hatsopoulos, N. G., & Gaal, G. (1998). Neural discharge and local field potential oscillations in primate motor cortex during voluntary movements. Journal of Neurophysiology, 79, 159–173.PubMedGoogle Scholar
- Doursat, R., & Bienenstock, E. (2006). The self-organized growth of synfire patterns. In 10th international conference on cognitive and neural systems (ICCNS). Massachusetts: Boston University.Google Scholar
- Dutar, P., Vu, H. M., & Perkel, D. J. (1998). Multiple cell types distinguished by physiological, pharmacological, and anatomic properties in nucleus HVC of the adult zebra finch. Journal of Neurophysiology, 80(4), 1828–1838.PubMedGoogle Scholar
- Eckhorn, R., Bauer, R., Jordan, W., Brosch, M., Kruse, W., Munk, M., et al. (1988). Coherent oscillations: A mechanism of feature linking in the visual cortex? Biological Cybernetics, 60, 121–130.PubMedCrossRefGoogle Scholar
- Farries, M. A., & Fairhall, A. L. (2007). Reinforcement learning with modulated spike timing-dependent synaptic plasticity. Journal of Neurophysiology, 98, 3648–3665.PubMedCrossRefGoogle Scholar
- Fiete, I. R., Senn, W., Wang, C. Z. H., & Hahnloser, R. H. R. (2010). Spike-time-dependent plasticity and heterosynaptic competition organize networks to produce long scale-free sequences of neural activity. Neuron, 65, 563–576.PubMedCrossRefGoogle Scholar
- Georgopoulos, A., Kalaska, J., Caminiti, R., & Massey, J.T. (1982). On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. Journal of Neuroscience, 11(2), 1527–1537.Google Scholar
- Georgopoulos, A., Schwartz, A., & Kettner, R. (1986a). Neuronal population coding of movement direction. Science, 4771(233), 1416–1419.CrossRefGoogle Scholar
- Georgopoulos, A. P., Kettner, R. E., & Schwartz, A. B. (1988). Primate motor cortex and free arm movements to visual targets in three-dimensional space. II. Coding of the direction of movement by a neuronal population. Journal of Neuroscience, 8(8), 2928–2937.PubMedGoogle Scholar
- Georgopoulos, A. P., Merchant, H., Naselaris, T., & Amirikian, B. (2007). Mapping of the preferred direction in the motor cortex. PNAS, 104(26), 11068–11072.PubMedCrossRefGoogle Scholar
- Georgopoulos, A. P., Schwartz, A. B., & Kettner, R. E. (1986b). Neuronal population coding of movement direction. Science, 233, 1416–1419.PubMedCrossRefGoogle Scholar
- Gewaltig, M.-O., & Diesmann, M. (2007). NEST (neural simulation tool). Scholarpedia, 2(4), 1430.CrossRefGoogle Scholar
- Gewaltig, M.-O., Diesmann, M., & Aertsen, A. (2001). Propagation of cortical synfire activity: Survival probability in single trials and stability in the mean. Neural Networks, (14), 657–673.Google Scholar
- Ghosh, S., Putrino, D., Burro, B., & Ring, A. (2009). Patterns of spatio-temporal correlations in the neural activity of the cat motor cortex during trained forelimb movements. Somatosensory and Motor Research, 26(2–3), 31–49.PubMedCrossRefGoogle Scholar
- Glaze, C., & Troyer, T. (2008). Temporal variability in a synfire chain model of birdsong. BMC Neuroscience, 9(Suppl 1), 28.CrossRefGoogle Scholar
- Gray, C. M., & Singer, W. (1989). Stimulus-specific neuronal oscillations in orientation columns of cat visual cortex. Proceedings of the National Academy of Sciences of the United States of America, 86, 1698–1702.PubMedCrossRefGoogle Scholar
- Hahnloser, R. H., Kozhevnikov, A. A., & Fee, M. S. (2002). An ultra-sparse code underlies the generation of neural sequences in a songbird. Nature, 419(6902), 65–70.PubMedCrossRefGoogle Scholar
- Hansel, D., Mato, G., Meunier, C., & Neltner, L. (1998). On numerical simulations of integrate-and-fire neural networks. Neural Computation, 10(2), 467–483.PubMedCrossRefGoogle Scholar
- Hanuschkin, A., Diesmann, M., & Morrison, A. (2010a). Functional compositionality realized in biological realistic spiking neural networks by synfire chain competition. In Proceedings of the 40th Annual Meeting of the Society for Neuroscience.Google Scholar
- Hanuschkin, A., Diesmann, M., & Morrison, A. (2010b). A reafferent model of song syntax generation in the Bengalese finch. BMC Neuroscience, 11(Suppl 1), 33.CrossRefGoogle Scholar
- Hanuschkin, A., Herrmann, J. M., Morrison, A., & Diesmann, M. (2009a). A model of free monkey scribbling based on the propagation of cell assembly activity. BMC Neuroscience, 10(Suppl 1), 300.CrossRefGoogle Scholar
- Hanuschkin, A., Herrmann, J. M., Morrison, A., & Diesmann, M. (2009b). Modeling free monkey scribbling by the propagation of synchronous activity. In Proceedings of the 8th Göttingen Meeting of the German Neuroscience Society.Google Scholar
- Hanuschkin, A., Kunkel, S., Helias, M., Morrison, A., & Diesmann, M. (2010). A general and efficient method for incorporating precise spike times in globally time-driven simulations. Frontiers in Neuroinformatics (4)113. doi:10.3389/fninf.2010.00113.PubMedGoogle Scholar
- Hart, C. B., & Giszter, S. F. (2010). A neural basis for motor primitives in the spinal cord. Journal of Neuroscience, 30(4), 1322–1336.PubMedCrossRefGoogle Scholar
- Hayon, G., Abeles, M., & Lehmann, D. (2005). A model for representing the dynamics of a system of synfire chains. Journal of Computational Neuroscience, 18, 41–53.PubMedCrossRefGoogle Scholar
- Haβ, J., Blaschke, S., Rammsayer, T., & Herrmann, J. M. (2008). A neurocomputational model for optimal temporal processing. Journal of Computational Neuroscience, 25(3), 449–464.CrossRefGoogle Scholar
- Herrmann, M., Hertz, J. A., & Prügel-Bennett, A. (1995). Analysis of synfire chains. Network, 6, 403–414.CrossRefGoogle Scholar
- Ikegaya, Y., Aaron, G., Cossart, R., Aronov, D., Lampl, I., Ferster, D., et al. (2004). Synfire chains and cortical songs: Temporal modules of cortical activity. Science, 5670(304), 559–564.CrossRefGoogle Scholar
- Izhikevich, E. M. (2007). Solving the distal reward problem through linkage of STDP and dopamine signaling. Cerebral Cortex, 17(10), 2443–2452.PubMedCrossRefGoogle Scholar
- Izhikevich, E. M., Gally, J. A., & Edelman, G. M. (2004). Spike-timing dynamics of neuronal groups. Cerebral Cortex, 14, 933–944.PubMedCrossRefGoogle Scholar
- Jin, D. Z. (2002). Fast convergence of spike sequences to periodic patterns in recurrent networks. Physical Review Letters, 89(20), 208102.PubMedCrossRefGoogle Scholar
- Jin, D. Z. (2009). Generating variable birdsong syllable sequences with branching chain networks in avian premotor nucleus HVC. Physical Review E, 80(5), 051902.CrossRefGoogle Scholar
- Jin, D. Z., Ramazanoglu, F. M., & Seung, H. S. (2007). Intrinsic bursting enhances the robustness of a neural network model of sequence generation by avian brain area HVC. Journal of Computational Neuroscience, 23(3), 283–299.PubMedCrossRefGoogle Scholar
- Jun, J. K., & Jin, D. Z. (2007). Development of neural circuitry for precise temporal sequences through spontaneous activity, axon remodeling, and synaptic plasticity. PLoS ONE, 2(8), e723.CrossRefGoogle Scholar
- Kozhevnikov, A., & Fee, M. S. (2007). Singing-related activity of identified HVC neurons in the zebra finch. Journal of Neurophysiology, 97, 4271–4283.PubMedCrossRefGoogle Scholar
- Kumar, A., Rotter, S., & Aertsen, A. (2010). Spiking activity propagation in neuronal networks: Reconciling different perspectives on neural coding. Nature Reviews Neuroscience, 11, 615–627.PubMedCrossRefGoogle Scholar
- Kunkel, S., Diesmann, M., & Morrison, A. (2010). Limits to the development of feed-forward structures in large recurrent neuronal networks (submitted).Google Scholar
- Lacquaniti, F., Terzuolo, C., & Viviani, P. (1983). The law relating the kinematic and figural aspects of drawing movements. Acta Psychologica (Amst), 54(1–3), 115–130.CrossRefGoogle Scholar
- Lapicque, L. (1907). Recherches quantitatives sur l’excitation electrique des nerfs traitee comme une polarization. Journal de Physiologie et de Pathologie Générale, 9, 620–635.Google Scholar
- Legenstein, R., Pecevski, D., & Maass, W. (2008). A learning theory for reward-modulated spike-timing-dependent plasticity with application to biofeedback. PLoS Computational Biology, 4(10), e1000180.CrossRefGoogle Scholar
- Li, M., & Greenside, H. (2006). Stable propagation of a burst through a one-dimensional homogeneous excitatory chain model of songbird nucleus HVC. Physical Review E, 74(1), 011918.CrossRefGoogle Scholar
- Lindén, H., Pettersen, K. H., Tetzlaff, T., Potjans, T. C., Denker, M., Diesmann, M., et al. (2009a). Estimating the spatial range of local field potentials in a cortical population model. BMC Neuroscience, 10(Suppl I), 224.CrossRefGoogle Scholar
- Lindén, H., Potjans, T. C., Einevoll, G. T., Grün, S., & Diesmann, M. (2009b). Modeling the local field potential by a large-scale layered cortical network model. In Frontiers in Neuroinformatics. Conference abstract: 2nd INCF congress of Neuroinformatics. doi:10.3389/conf.neuro.11.2009.08.046.
- Mazzoni, A., Panzeri, S., Logothetis, N. K., & Brunel, N. (2008). Encoding of naturalistic stimuli by local field potential spectra in networks of excitatory and inhibitory neurons. PLoS Computational Biology, 4(12), e1000239.CrossRefGoogle Scholar
- Mooney, R., & Prather, J. F. (2005). The HVC microcircuit: The synaptic basis for interactions between song motor and vocal plasticity pathways. Journal of Neuroscience, 25(8), 1952–1964.PubMedCrossRefGoogle Scholar
- Morrison, A., Aertsen, A., & Diesmann, M. (2007a). Spike-timing dependent plasticity in balanced random networks. Neural Computation, 19, 1437–1467.PubMedCrossRefGoogle Scholar
- Morrison, A., Straube, S., Plesser, H. E., & Diesmann, M. (2007b). Exact subthreshold integration with continuous spike times in discrete time neural network simulations. Neural Computation, 19(1), 47–79.PubMedCrossRefGoogle Scholar
- Murthy, V. N., & Fetz, E. (1996). Oscillatory activity in sensorimotor cortex of awake monkeys: Synchronization of local field potentials and relation to behavior. Journal of Neurophysiology, 76, 3949–3967.PubMedGoogle Scholar
- Mussa-Ivaldi, F. A., & Bizzi, E. (2000). Motor learning through the combination of primitives. Philosophical Transactions of the Royal Society London, Series B, 355(1404), 1755–1769.CrossRefGoogle Scholar
- Mussa-Ivaldi, F. A., Giszter, S., & Bizzi, E. (1994). Linear combinations of primitives in vertebrate motor control. Proceedings of the National Academy of Sciences of the United States of America, 91, 7534–7538.PubMedCrossRefGoogle Scholar
- Nordlie, E., Gewaltig, M.-O., & Plesser, H. E. (2009). Towards reproducible descriptions of neuronal network models. PLoS Computational Biology, 5(8), e1000456.CrossRefGoogle Scholar
- O’Leary, J. G., & Hatsopoulos, N. G. (2006). Early visuomotor representations revealed from evoked local field potentials in motor and premotor cortical areas. Journal of Neurophysiology, 96(3), 1492–1506.PubMedCrossRefGoogle Scholar
- Plesser, H. E., & Diesmann, M. (2009). Simplicity and efficiency of integrate-and-fire neuron models. Neural Computation, 21, 353–359.PubMedCrossRefGoogle Scholar
- Polyakov, F., Drori, R., Ben-Shaul, Y., Abeles, M., & Flash, T. (2009a). A compact representation of drawing movements with sequences of parabolic primitives. PLoS Computational Biology, 5(7), e1000427.CrossRefGoogle Scholar
- Polyakov, F., Stark, E., Drori, R., Abeles, M., & Flash, T. (2009b). Parabolic movement primitives and cortical states: Merging optimality with geometric invariance. Biological Cybernetics, 100(2), 159–184.PubMedCrossRefGoogle Scholar
- Ponce-Alvarez, A., Kilavik, B. E., & Riehle, A. (2010). Comparison of local measures of spike time irregularity and relating variability to firing rate in motor cortical neurons. Journal of Computational Neuroscience, 29(1–2), 351–365.PubMedCrossRefGoogle Scholar
- Potjans, W., Morrison, A., & Diesmann, M. (2009). A spiking neural network model of an actor-critic learning agent. Neural Computation, 21, 301–339.PubMedCrossRefGoogle Scholar
- Potjans, W., Diesmann, M, & Morrison, A. (2010). An imperfect dopaminergic signal can drive temporal-difference learning (submitted).Google Scholar
- Prut, Y., Vaadia, E., Bergman, H., Haalman, I., Hamutal, S., & Abeles, M. (1998). Spatiotemporal structure of cortical activity: Properties and behavioral relevance. Journal of Neurophysiology, 79(6), 2857–2874.PubMedGoogle Scholar
- Pulvermüller, F., & Shtyrov, Y. (2009). Spatiotemporal signatures of large-scale synfire chains for speech processing as revealed by MEG. Cerebral Cortex, 19, 79–88.PubMedCrossRefGoogle Scholar
- Putrino, D., Brown, E. N., Mastaglia, F. L., & Ghosh, S. (2010). Differential involvement of excitatory and inhibitory neurons of cat motor cortex in coincident spike activity related to behavioral context. Journal of Neuroscience, 30(23), 8048–8056.PubMedCrossRefGoogle Scholar
- Rickert, J., Cardoso de Oliveira, S., Vaadia, E., Aertsen, A., Rotter, S., & Mehring, C. (2005). Encoding of movement direction in different frequency ranges of motor cortical local field potentials. Journal of Neuroscience, 25(39), 8815–8824.PubMedCrossRefGoogle Scholar
- Riehle, A., Grammont, F., Diesmann, M., & Grün, S. (2000). Dynamical changes and temporal precision of synchronized spiking activity in monkey motor cortex during movement preparation. Journal of Physiology (Paris), 94(5–6), 569–582. Erratum in “Journal of Physiology (Paris)”, 95(1–6), 499 (2001).Google Scholar
- Riehle, A., Grün, S., Diesmann, M., & Aertsen, A. (1997). Spike synchronization and rate modulation differentially involved in motor cortical function. Science, 278(5345), 1950–1953.PubMedCrossRefGoogle Scholar
- Rioult-Pedotti, M. S., Friedman, D., Hess, G., & Donoghue, J. P. (1998). Strengthening of horizontal cortical connections following skill learning. Nature Neuroscience, 1, 230–234.PubMedCrossRefGoogle Scholar
- Rotter, S., & Diesmann, M. (1999). Exact digital simulation of time-invariant linear systems with applications to neuronal modeling. Biological Cybernetics, 81(5–6), 381–402.PubMedCrossRefGoogle Scholar
- Roux, S., Mackay, W. A., & Riehle, A. (2006). The pre-movement component of motor cortical local field potentials reflects the level of expectancy. Behavioural Brain Research, 169(2), 335–351.PubMedCrossRefGoogle Scholar
- Schrader, S., Diesmann, M., & Morrison, A. (2010). A compositionality machine realized by a hierarchic architecture of synfire chains (submitted).Google Scholar
- Schrader, S., Grün, S., Diesmann, M., & Gerstein, G. (2008). Detecting synfire chain activity using massively parallel spike train recording. Journal of Neurophysiology, 100, 2165–2176.PubMedCrossRefGoogle Scholar
- Schwartz, A. B. (1994). Direct cortical representation of drawing. Science, 265(5171), 540–542.PubMedCrossRefGoogle Scholar
- Sforza, C., Turci, M., Grassi, G., Fragnito, N., Pizzini, G., & Ferrario, V. (2000). The repeatability of choku-tsuki and oi-tsuki in traditional shotokan karate: A morphological three-dimensional analysis. Perceptual and Motor Skills, 90, 947–960.PubMedCrossRefGoogle Scholar
- Shmiel, T., Drori, R., Shmiel, O., Ben-Shaul, Y., Nadasdy, Z., Shemesh, M., et al. (2005). Neurons of the cerebral cortex exhibit precise interspike timing in correspondence to behavior. PNAS, 102(51), 18655–18657.PubMedCrossRefGoogle Scholar
- Shmiel, T., Drori, R., Shmiel, O., Ben-Shaul, Y., Nadasdy, Z., Shemesh, M., et al. (2006). Temporally precise cortical firing patterns are associated with distinct action segments. Journal of Neurophysiology, 96(5), 2645–2652.PubMedCrossRefGoogle Scholar
- Singer, W., & Gray, C. (1995). Visual feature integration and the temporal correlation hypothesis. Annuual Review of Neuroscience, 18, 555–586.CrossRefGoogle Scholar
- Softky, W. R., & Koch, C. (1993). The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs. Journal of Neuroscience, 13(1), 334–350.PubMedGoogle Scholar
- Stark, E., Drori, R., & Abeles, M. (2009). Motor cortical activity related to movement kinematics exhibits local spatial organization. Cortex, 45, 418–431.PubMedCrossRefGoogle Scholar
- Tetzlaff, T., Geisel, T., & Diesmann, M. (2002). The ground state of cortical feed-forward networks. Neurocomputing, 44–46, 673–678.CrossRefGoogle Scholar
- Ursino, M., & Cara, G.-E. L. (2006). Travelling waves and EEG patterns during epileptic seizure: Analysis with an integrate-and-fire neural network. Journal of Theoretical Biology, 242(1), 171–187.PubMedCrossRefGoogle Scholar
- van Vreeswijk, C., & Sompolinsky, H. (1996). Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science, 274, 1724–1726.PubMedCrossRefGoogle Scholar
- Viviani, P., & Flash, T. (1995). Minimum-jerk, two-thirds power law, and isochrony: Converging approaches to movement planning. Journal of Experimental Psychology: Human Perception and Performance, 21(1), 32–53.PubMedCrossRefGoogle Scholar
- Wennekers, T., & Palm, G. (1996). Controlling the speed of synfire chains. In C. von der Malsburg, W. von Seelen, J. C. Vorbrüggen, & B. Sendhoff (Eds.), Artificial neural networks—ICANN 96 (pp. 451–456). Berlin: Springer-Verlag.Google Scholar
- Xu, T., Yu, X., Perlik, A. J., Tobin, W. F., Zweig, J. A., Tennant, K., et al. (2009). Rapid formation and selective stabilization of synapses for enduring motor memories. Nature, 462, 915–919.PubMedCrossRefGoogle Scholar
- Yamashita, Y., & Tani, J. (2008). Emergence of functional hierarchy in a multiple timescale neural network model: A humanoid robot experiment. PLoS Computational Biology, 4(11), e1000220.CrossRefGoogle Scholar