Duality of rate coding and temporal coding in multilayered feedforward networks

A minimum-error, energy-constrained neural code is an instantaneous-rate code

Sensory neurons code information about stimuli in their sequence of action potentials (spikes). Intuitively, the spikes should represent stimuli with high fidelity. However, generating and propagating spikes is a metabolically expensive process. It is therefore likely that neural codes have been selected to balance energy expenditure against encoding error. Our recently proposed optimal, energy-constrained neural coder (Jones et al. Frontiers in Computational Neuroscience, 9, 61 2015) postulates that neurons time spikes to minimize the trade-off between stimulus reconstruction error and expended energy by adjusting the spike threshold using a simple dynamic threshold. Here, we show that this proposed coding scheme is related to existing coding schemes, such as rate and temporal codes. We derive an instantaneous rate coder and show that the spike-rate depends on the signal and its derivative. In the limit of high spike rates the spike train maximizes fidelity given an energy constraint (average spike-rate), and the predicted interspike intervals are identical to those generated by our existing optimal coding neuron. The instantaneous rate coder is shown to closely match the spike-rates recorded from P-type primary afferents in weakly electric fish. In particular, the coder is a predictor of the peristimulus time histogram (PSTH). When tested against in vitro cortical pyramidal neuron recordings, the instantaneous spike-rate approximates DC step inputs, matching both the average spike-rate and the time-to-first-spike (a simple temporal code). Overall, the instantaneous rate coder relates optimal, energy-constrained encoding to the concepts of rate-coding and temporal-coding, suggesting a possible unifying principle of neural encoding of sensory signals.

Intrinsic excitability state of local neuronal population modulates signal propagation in feed-forward neural networks

Reliable signal propagation across distributed brain areas is an essential requirement for cognitive function, and it has been investigated extensively in computational studies where feed-forward network (FFN) is taken as a generic model. But it is still unclear how distinct local network states, which are intrinsically generated by synaptic interactions within each layer, would affect the ability of FFN to transmit information. Here we investigate the impact of such network states on propagating transient synchrony (synfire) and firing rate by a combination of numerical simulations and analytical approach. Specifically, local network dynamics is attributed to the competition between excitatory and inhibitory neurons within each layer. Our results show that concomitant with different local network states, the performance of signal propagation differs dramatically. For both synfire propagation and firing rate propagation, there exists an optimal local excitability state, respectively, that optimizes the performance of signal propagation. Furthermore, we find that long-range connections strongly change the dependence of spiking activity propagation on local network state and propose that these two factors work jointly to determine information transmission across distributed networks. Finally, a simple mean field approach that bridges response properties of long-range connectivity and local subnetworks is utilized to reveal the underlying mechanism.

A mixed-signal implementation of a polychronous spiking neural network with delay adaptation

We present a mixed-signal implementation of a re-configurable polychronous spiking neural network capable of storing and recalling spatio-temporal patterns. The proposed neural network contains one neuron array and one axon array. Spike Timing Dependent Delay Plasticity is used to fine-tune delays and add dynamics to the network. In our mixed-signal implementation, the neurons and axons have been implemented as both analog and digital circuits. The system thus consists of one FPGA, containing the digital neuron array and the digital axon array, and one analog IC containing the analog neuron array and the analog axon array. The system can be easily configured to use different combinations of each. We present and discuss the experimental results of all combinations of the analog and digital axon arrays and the analog and digital neuron arrays. The test results show that the proposed neural network is capable of successfully recalling more than 85% of stored patterns using both analog and digital circuits.

Synthesis of neural networks for spatio-temporal spike pattern recognition and processing

The advent of large scale neural computational platforms has highlighted the lack of algorithms for synthesis of neural structures to perform predefined cognitive tasks. The Neural Engineering Framework (NEF) offers one such synthesis, but it is most effective for a spike rate representation of neural information, and it requires a large number of neurons to implement simple functions. We describe a neural network synthesis method that generates synaptic connectivity for neurons which process time-encoded neural signals, and which makes very sparse use of neurons. The method allows the user to specify—arbitrarily—neuronal characteristics such as axonal and dendritic delays, and synaptic transfer functions, and then solves for the optimal input-output relationship using computed dendritic weights. The method may be used for batch or online learning and has an extremely fast optimization process. We demonstrate its use in generating a network to recognize speech which is sparsely encoded as spike times.

Cortical gamma oscillations: the functional key is activation, not cognition

Cortical oscillatory synchrony in the gamma range has been attracting increasing attention in cognitive neuroscience ever since being proposed as a solution to the so-called binding problem. This growing literature is critically reviewed in both its basic neuroscience and cognitive aspects. A physiological "default assumption" regarding these oscillations is introduced, according to which they signal a state of physiological activation of cortical tissue, and the associated need to balance excitation with inhibition in particular. As such these oscillations would belong among a variety of generic neural control operations that enable neural tissue to perform its systems level functions, without implementing those functions themselves. Regional control of cerebral blood flow provides an analogy in this regard, and gamma oscillations are tightly correlated with this even more elementary control operation. As correlates of neural activation they will also covary with cognitive activity, and this typically suffices to account for the covariation between gamma activity and cognitive task variables. A number of specific cases of gamma synchrony are examined in this light, including the original impetus for attributing cognitive significance to gamma activity, namely the experiments interpreted as evidence for "binding by synchrony". This examination finds no compelling reasons to assign functional roles to oscillatory synchrony in the gamma range beyond its generic functions at the level of infrastructural neural control.

Adaptive coupling of inferior olive neurons in cerebellar learning

In the cerebellar learning hypothesis, inferior olive neurons are presumed to transmit high fidelity error signals, despite their low firing rates. The idea of chaotic resonance has been proposed to realize efficient error transmission by desynchronized spiking activities induced by moderate electrical coupling between inferior olive neurons. A recent study suggests that the coupling strength between inferior olive neurons can be adaptive and may decrease during the learning process. We show that such a decrease in coupling strength can be beneficial for motor learning, since efficient coupling strength depends upon the magnitude of the error signals. We introduce a scheme of adaptive coupling that enhances the learning of a neural controller for fast arm movements. Our numerical study supports the view that the controlling strategy of the coupling strength provides an additional degree of freedom to optimize the actual learning in the cerebellum.

Population rate coding in recurrent neuronal networks with unreliable synapses

Neuron transmits spikes to postsynaptic neurons through synapses. Experimental observations indicated that the communication between neurons is unreliable. However most modelling and computational studies considered deterministic synaptic interaction model. In this paper, we investigate the population rate coding in an all-to-all coupled recurrent neuronal network consisting of both excitatory and inhibitory neurons connected with unreliable synapses. We use a stochastic on-off process to model the unreliable synaptic transmission. We find that synapses with suitable successful transmission probability can enhance the encoding performance in the case of weak noise; while in the case of strong noise, the synaptic interactions reduce the encoding performance. We also show that several important synaptic parameters, such as the excitatory synaptic strength, the relative strength of inhibitory and excitatory synapses, as well as the synaptic time constant, have significant effects on the performance of the population rate coding. Further simulations indicate that the encoding dynamics of our considered network cannot be simply determined by the average amount of received neurotransmitter for each neuron in a time instant. Moreover, we compare our results with those obtained in the corresponding random neuronal networks. Our numerical results demonstrate that the network randomness has the similar qualitative effect as the synaptic unreliability but not completely equivalent in quantity.

Dual roles for spike signaling in cortical neural populations

A prominent feature of signaling in cortical neurons is that of randomness in the action potential. The output of a typical pyramidal cell can be well fit with a Poisson model, and variations in the Poisson rate repeatedly have been shown to be correlated with stimuli. However while the rate provides a very useful characterization of neural spike data, it may not be the most fundamental description of the signaling code. Recent data showing γ frequency range multi-cell action potential correlations, together with spike timing dependent plasticity, are spurring a re-examination of the classical model, since precise timing codes imply that the generation of spikes is essentially deterministic. Could the observed Poisson randomness and timing determinism reflect two separate modes of communication, or do they somehow derive from a single process? We investigate in a timing-based model whether the apparent incompatibility between these probabilistic and deterministic observations may be resolved by examining how spikes could be used in the underlying neural circuits. The crucial component of this model draws on dual roles for spike signaling. In learning receptive fields from ensembles of inputs, spikes need to behave probabilistically, whereas for fast signaling of individual stimuli, the spikes need to behave deterministically. Our simulations show that this combination is possible if deterministic signals using γ latency coding are probabilistically routed through different members of a cortical cell population at different times. This model exhibits standard features characteristic of Poisson models such as orientation tuning and exponential interval histograms. In addition, it makes testable predictions that follow from the γ latency coding.

Representing spike trains using constant sampling intervals

Sensory neurons encode external information by a series of times of action potentials, which is called a spike train. However, since it is a point process, it is hard to analyze. Here we propose a method for converting a spike train into a real-valued time series with a fixed sampling interval under the assumption of temporal codes. The proposed method yields time series that represent encoded signals. Especially when the method is applied to spike trains generated using integrate-and-fire models, the yielded time series look very similar to those of encoded information. The method works robustly even when a spike train is contaminated with noise. Since unlike filters it does not use its original signals for the conversion, the proposed method can be widely used for investigating spike train data in the real world.

Self-organization of feed-forward structure and entrainment in excitatory neural networks with spike-timing-dependent plasticity

Spike-timing dependent plasticity (STDP) is an organizing principle of biological neural networks. While synchronous firing of neurons is considered to be an important functional block in the brain, how STDP shapes neural networks possibly toward synchrony is not entirely clear. We examine relations between STDP and synchronous firing in spontaneously firing neural populations. Using coupled heterogeneous phase oscillators placed on initial networks, we show numerically that STDP prunes some synapses and promotes formation of a feedforward network. Eventually a pacemaker, which is the neuron with the fastest inherent frequency in our numerical simulations, emerges at the root of the feedforward network. In each oscillatory cycle, a packet of neural activity is propagated from the pacemaker to downstream neurons along layers of the feedforward network. This event occurs above a clear-cut threshold value of the initial synaptic weight. Below the threshold, neurons are self-organized into separate clusters each of which is a feedforward network.

Data-driven significance estimation for precise spike correlation

The mechanisms underlying neuronal coding and, in particular, the role of temporal spike coordination are hotly debated. However, this debate is often confounded by an implicit discussion about the use of appropriate analysis methods. To avoid incorrect interpretation of data, the analysis of simultaneous spike trains for precise spike correlation needs to be properly adjusted to the features of the experimental spike trains. In particular, nonstationarity of the firing of individual neurons in time or across trials, a spike train structure deviating from Poisson, or a co-occurrence of such features in parallel spike trains are potent generators of false positives. Problems can be avoided by including these features in the null hypothesis of the significance test. In this context, the use of surrogate data becomes increasingly important, because the complexity of the data typically prevents analytical solutions. This review provides an overview of the potential obstacles in the correlation analysis of parallel spike data and possible routes to overcome them. The discussion is illustrated at every stage of the argument by referring to a specific analysis tool (the Unitary Events method). The conclusions, however, are of a general nature and hold for other analysis techniques. Thorough testing and calibration of analysis tools and the impact of potentially erroneous preprocessing stages are emphasized.

Can spike coordination be differentiated from rate covariation?

There has been a long and lively debate on whether rate covariance and temporal coordination of spikes, regarded as potential origins for correlations in cortical spike signals, fulfill different roles in the cortical code. In this context, studies that report spike coordination have often been criticized for ignoring fast nonstationarities, which would result in wrongly assigned spike coordination. The underlying hypothesis of this critique is that spike coordination is essentially identical to rate covariation, only on a shorter timescale. This study investigates the validity of this critique. We provide a decomposition for the cross-correlation function of doubly stochastic point processes, where each of the components corresponds precisely to the concepts of dependence under investigation. This allows us to correct the correlation function for rate effects, which implies that spike coordination and rate covariation are statistically separable concepts of dependence. Furthermore, we present direct and intuitive model implementations of the discussed concepts and illustrate that their difference is not a matter of timescale. Analysis of data generated by our models and analytical description of the relevant estimators reveals, however, that spike coordination dramatically influences the accuracy of rate covariance estimation. As a consequence, extreme parameter combinations can lead to situations where the concept of dependence cannot be identified empirically. However, for a wide range of parameters, the concept of dependence underlying a given data set can be identified regardless of its timescale.

Testing a neural coding hypothesis using surrogate data

Determining how a particular neuron, or population of neurons, encodes information in their spike trains is not a trivial problem, because multiple coding schemes exist and are not necessarily mutually exclusive. Coding schemes generally fall into one of two broad categories, which we refer to as rate and temporal coding. In rate coding schemes, information is encoded in the variations of the average firing rate of the spike train. In contrast, in temporal coding schemes, information is encoded in the specific timing of the individual spikes that comprise the train. Here, we describe a method for testing the presence of temporal encoding of information. Suppose that a set of original spike trains is given. First, surrogate spike trains are generated by randomizing each of the original spike trains subject to the following constraints: the local average firing rate is approximately preserved, while the overall average firing rate and the distribution of primary interspike intervals are perfectly preserved. These constraints ensure that any rate coding of information present in the original spike trains is preserved in the members of the surrogate population. The null-hypothesis is rejected when additional information is found to be present in the original spike trains, implying that temporal coding is present. The method is validated using artificial data, and then demonstrated using real neuronal data.

Implications of noise and neural heterogeneity for vestibulo-ocular reflex fidelity

The vestibulo-ocular reflex (VOR) is characterized by a short-latency, high-fidelity eye movement response to head rotations at frequencies up to 20 Hz. Electrophysiological studies of medial vestibular nucleus (MVN) neurons, however, show that their response to sinusoidal currents above 10 to 12 Hz is highly nonlinear and distorted by aliasing for all but very small current amplitudes. How can this system function in vivo when single cell response cannot explain its operation? Here we show that the necessary wide VOR frequency response may be achieved not by firing rate encoding of head velocity in single neurons, but in the integrated population response of asynchronously firing, intrinsically active neurons. Diffusive synaptic noise and the pacemaker-driven, intrinsic firing of MVN cells synergistically maintain asynchronous, spontaneous spiking in a population of model MVN neurons over a wide range of input signal amplitudes and frequencies. Response fidelity is further improved by a reciprocal inhibitory link between two MVN populations, mimicking the vestibular commissural system in vivo, but only if asynchrony is maintained by noise and pacemaker inputs. These results provide a previously missing explanation for the full range of VOR function and a novel account of the role of the intrinsic pacemaker conductances in MVN cells. The values of diffusive noise and pacemaker currents that give optimal response fidelity yield firing statistics similar to those in vivo, suggesting that the in vivo network is tuned to optimal performance. While theoretical studies have argued that noise and population heterogeneity can improve coding, to our knowledge this is the first evidence indicating that these parameters are indeed tuned to optimize coding fidelity in a neural control system in vivo.

Discharge synchrony during the transition of behavioral goal representations encoded by discharge rates of prefrontal neurons

To investigate the temporal relationship between synchrony in the discharge of neuron pairs and modulation of the discharge rate, we recorded the neuronal activity of the lateral prefrontal cortex of monkeys performing a behavioral task that required them to plan an immediate goal of action to attain a final goal. Information about the final goal was retrieved via visual instruction signals, whereas information about the immediate goal was generated internally. The synchrony of neuron pair discharges was analyzed separately from changes in the firing rate of individual neurons during a preparatory period. We focused on neuron pairs that exhibited a representation of the final goal followed by a representation of the immediate goal at a later stage. We found that changes in synchrony and discharge rates appeared to be complementary at different phases of the behavioral task. Synchrony was maximized during a specific phase in the preparatory period corresponding to a transitional stage when the neuronal activity representing the final goal was replaced with that representing the immediate goal. We hypothesize that the transient increase in discharge synchrony is an indication of a process that facilitates dynamic changes in the prefrontal neural circuits in order to undergo profound state changes.

Computational properties of networks of synchronous groups of spiking neurons

We demonstrate a model in which synchronously firing ensembles of neurons are networked to produce computational results. Each ensemble is a group of biological integrate-and-fire spiking neurons, with probabilistic interconnections between groups. An analogy is drawn in which each individual processing unit of an artificial neural network corresponds to a neuronal group in a biological model. The activation value of a unit in the artificial neural network corresponds to the fraction of active neurons, synchronously firing, in a biological neuronal group. Weights of the artificial neural network correspond to the product of the interconnection density between groups, the group size of the presynaptic group, and the postsynaptic potential heights in the synchronous group model. All three of these parameters can modulate connection strengths between neuronal groups in the synchronous group models. We give an example of nonlinear classification (XOR) and a function approximation example in which the capability of the artificial neural network can be captured by a neural network model with biological integrate-and-fire neurons configured as a network of synchronously firing ensembles of such neurons. We point out that the general function approximation capability proven for feedforward artificial neural networks appears to be approximated by networks of neuronal groups that fire in synchrony, where the groups comprise integrate-and-fire neurons. We discuss the advantages of this type of model for biological systems, its possible learning mechanisms, and the associated timing relationships.

Dynamic switching of neural codes in networks with gap junctions

Population rate coding and temporal coding are common neural codes. Recent studies suggest that these two codes may be alternatively used in one neural system. Based on the fact that there are massive gap junctions in the brain, we explore how this switching behavior may be related to neural codes in networks of neurons connected by gap junctions. First, we show that under time-varying inputs, such neural networks show switching between synchronous and asynchronous states. Then, we quantify network dynamics by three mutual information measures to show that population rate coding carries more information in asynchronous states and temporal coding does so in synchronous states.

Dual coding hypotheses for neural information representation

Information is represented and processed in neural systems in various ways. The rate coding, population coding, and temporal coding are typical examples of representation. It is a hot issue in neuroscience what kinds of coding is used in real neural systems. Different regions of the brain may resort to different coding strategies. Moreover, recent studies suggest the possibility of dual or multiple codes, in which different modes of information are embedded in one neural system. The present paper reviews various possibilities of neural codes focusing on dual codes.

Simultaneous rate-synchrony codes in populations of spiking neurons

Firing rates and synchronous firing are often simultaneously relevant signals, and they independently or cooperatively represent external sensory inputs, cognitive events, and environmental situations such as body position. However, how rates and synchrony comodulate and which aspects of inputs are effectively encoded, particularly in the presence of dynamical inputs, are unanswered questions. We examine theoretically how mixed information in dynamic mean input and noise input is represented by dynamic population firing rates and synchrony. In a subthreshold regime, amplitudes of spatially uncorrelated noise are encoded up to a fairly high input frequency, but this requires both rate and synchrony output channels. In a suprathreshold regime, means and common noise amplitudes can be simultaneously and separately encoded by rates and synchrony, respectively, but the input frequency for which this is possible has a lower limit.

Relation between single neuron and population spiking statistics and effects on network activity

To simplify theoretical analyses of neural networks, individual neurons are often modeled as Poisson processes. An implicit assumption is that even if the spiking activity of each neuron is non-Poissonian, the composite activity obtained by summing many spike trains limits to a Poisson process. Here, we show analytically and through simulations that this assumption is invalid. Moreover, we show with Fokker-Planck equations that the behavior of feedforward networks is reproduced accurately only if the tendency of neurons to fire periodically is incorporated by using colored noise whose autocorrelation has a negative component.

Bidirectional modulation of neuronal responses by depolarizing GABAergic inputs

The reversal potential of GABAA receptor channels is known to be less negative than the resting membrane potential under some cases. Recent electrophysiological experiments revealed that a GABAergic unitary conductance with such a depolarized reversal potential could not only prevent but also facilitate action potential generation depending on the timing of its application relative to the excitatory unitary conductance. Using a two-dimensional point neuron model, we simulate the experiments regarding the integration of unitary conductances, and execute bifurcation analysis. Then we extend our analysis to the case in which the neuron receives two kinds of periodic input trains-an excitatory one and a GABAergic one. We show that the periodic depolarizing GABAergic input train can modulate the output time-averaged firing rate bidirectionally, namely as an increase or a decrease, in a devil's-staircase-like manner depending on the phase difference with the excitatory input train. Bifurcation analysis reveals the existence of a wide variety of phase-locked solutions underlying such a graded response of the neuron. We examine how the input time-width and the value of the GABAA reversal potential affect the response. Moreover, considering a neuronal population, we show that depolarizing GABAergic inputs bidirectionally modulate the amplitude of the oscillatory population activity.

Coding of temporally varying signals in networks of spiking neurons with global delayed feedback

Oscillatory and synchronized neural activities are commonly found in the brain, and evidence suggests that many of them are caused by global feedback. Their mechanisms and roles in information processing have been discussed often using purely feedforward networks or recurrent networks with constant inputs. On the other hand, real recurrent neural networks are abundant and continually receive information-rich inputs from the outside environment or other parts of the brain. We examine how feedforward networks of spiking neurons with delayed global feedback process information about temporally changing inputs. We show that the network behavior is more synchronous as well as more correlated with and phase-locked to the stimulus when the stimulus frequency is resonant with the inherent frequency of the neuron or that of the network oscillation generated by the feedback architecture. The two eigenmodes have distinct dynamical characteristics, which are supported by numerical simulations and by analytical arguments based on frequency response and bifurcation theory. This distinction is similar to the class I versus class II classification of single neurons according to the bifurcation from quiescence to periodic firing, and the two modes depend differently on system parameters. These two mechanisms may be associated with different types of information processing.

Self-Organizing Dual Coding Based on Spike-Time-Dependent Plasticity

It has been a matter of debate how firing rates or spatiotemporal spike patterns carry information in the brain. Recent experimental and theoretical work in part showed that these codes, especially a population rate code and a synchronous code, can be dually used in a single architecture. However, we are not yet able to relate the role of firing rates and synchrony to the spatiotemporal structure of inputs and the architecture of neural networks. In this article, we examine how feedforward neural networks encode multiple input sources in the firing patterns. We apply spike-time-dependent plasticity as a fundamental mechanism to yield synaptic competition and the associated input filtering. We use the Fokker-Planck formalism to analyze the mechanism for synaptic competition in the case of multiple inputs, which underlies the formation of functional clusters in downstream layers in a self-organizing manner. Depending on the types of feedback coupling and shared connectivity, clusters are independently engaged in population rate coding or synchronous coding, or they interact to serve as input filters. Classes of dual codings and functional roles of spike-time-dependent plasticity are also discussed.

Ergodicity of spike trains: when does trial averaging make sense?

Neuronal information processing is often studied on the basis of spiking patterns. The relevant statistics such as firing rates calculated with the peri-stimulus time histogram are obtained by averaging spiking patterns over many experimental runs. However, animals should respond to one experimental stimulation in real situations, and what is available to the brain is not the trial statistics but the population statistics. Consequently, physiological ergodicity, namely, the consistency between trial averaging and population averaging, is implicitly assumed in the data analyses, although it does not trivially hold true. In this letter, we investigate how characteristics of noisy neural network models, such as single neuron properties, external stimuli, and synaptic inputs, affect the statistics of firing patterns. In particular, we show that how high membrane potential sensitivity to input fluctuations, inability of neurons to remember past inputs, external stimuli with large variability and temporally separated peaks, and relatively few contributions of synaptic inputs result in spike trains that are reproducible over many trials. The reproducibility of spike trains and synchronous firing are contrasted and related to the ergodicity issue. Several numerical calculations with neural network examples are carried out to support the theoretical results.