Output Spike Trains Research Articles

Temporal pattern separation in hippocampal neurons through multiplexed neural codes.

Pattern separation is a central concept in current theories of episodic memory: this computation is thought to support our ability to avoid confusion between similar memories by transforming similar cortical input patterns of neural activity into dissimilar output patterns before their long-term storage in the hippocampus. Because there are many ways one can define patterns of neuronal activity and the similarity between them, pattern separation could in theory be achieved through multiple coding strategies. Using our recently developed assay that evaluates pattern separation in isolated tissue by controlling and recording the input and output spike trains of single hippocampal neurons, we explored neural codes through which pattern separation is performed by systematic testing of different similarity metrics and various time resolutions. We discovered that granule cells, the projection neurons of the dentate gyrus, can exhibit both pattern separation and its opposite computation, pattern convergence, depending on the neural code considered and the statistical structure of the input patterns. Pattern separation is favored when inputs are highly similar, and is achieved through spike time reorganization at short time scales (< 100 ms) as well as through variations in firing rate and burstiness at longer time scales. These multiplexed forms of pattern separation are network phenomena, notably controlled by GABAergic inhibition, that involve many celltypes with input-output transformations that participate in pattern separation to different extents and with complementary neural codes: a rate code for dentate fast-spiking interneurons, a burstiness code for hilar mossy cells and a synchrony code at long time scales for CA3 pyramidal cells. Therefore, the isolated hippocampal circuit itself is capable of performing temporal pattern separation using multiplexed coding strategies that might be essential to optimally disambiguate multimodal mnemonic representations.

Pattern separation of spiketrains in hippocampal neurons

Pattern separation is a process that minimizes overlap between patterns of neuronal activity representing similar experiences. Theoretical work suggests that the dentate gyrus (DG) performs this role for memory processing but a direct demonstration is lacking. One limitation is the difficulty to measure DG inputs and outputs simultaneously. To rigorously assess pattern separation by DG circuitry, we used mouse brain slices to stimulate DG afferents and simultaneously record DG granule cells (GCs) and interneurons. Output spiketrains of GCs are more dissimilar than their input spiketrains, demonstrating for the first time temporal pattern separation at the level of single neurons in the DG. Pattern separation is larger in GCs than in fast-spiking interneurons and hilar mossy cells, and is amplified in CA3 pyramidal cells. Analysis of the neural noise and computational modelling suggest that this form of pattern separation is not explained by simple randomness and arises from specific presynaptic dynamics. Overall, by reframing the concept of pattern separation in dynamic terms and by connecting it to the physiology of different types of neurons, our study offers a new window of understanding in how hippocampal networks might support episodic memory.

Efficient Multispike Learning for Spiking Neural Networks Using Probability-Modulated Timing Method.

Error functions are normally based on the distance between output spikes and target spikes in supervised learning algorithms for spiking neural networks (SNNs). Due to the discontinuous nature of the internal state of spiking neuron, it is challenging to ensure that the number of output spikes and target spikes kept identical in multispike learning. This problem is conventionally dealt with by using the smaller of the number of desired spikes and that of actual output spikes in learning. However, if this approach is used, information is lost as some spikes are neglected. In this paper, a probability-modulated timing mechanism is built on the stochastic neurons, where the discontinuous spike patterns are converted to the likelihood of generating the desired output spike trains. By applying this mechanism to a probability-modulated spiking classifier, a probability-modulated SNN (PMSNN) is constructed. In its multilayer and multispike learning structure, more inputs are incorporated and mapped to the target spike trains. A clustering rule connection mechanism is also applied to a reservoir to improve the efficiency of information transmission among synapses, which can map the highly correlated inputs to the adjacent neurons. Results of comparisons between the proposed method and popular the SNN algorithms showed that the PMSNN yields higher efficiency and requires fewer parameters.

(Invited) Multi-Terminal Oxide-based Electric-Double-Layer Thin-Film Transistors for Neuromorphic Systems

Our brain is a highly parallel, energy efficient and event-driven information processing system, which is fundamentally different from traditional von Neumann computers. Neuron is the most important building block, and information processing in individual neuron involves the transformation of input synaptic spike train into an appropriate output spike train. Inspired by biological neural computing, neuromorphic systems may open up new paradigms to deal with complicated problems such as pattern recognition, classification and decision making. The idea of building brain-inspired adaptive artificial devices has been around for decades, and hardware implementation of neuron by individual ionic/electronic hybrid device is of great significance for enhancing our understanding of the brain and solving sensory processing and complex recognition tasks. Although two-terminal memristors can performe some basic synaptic and neural functions, our human brain contains many more synapses than neurons. This fact suggests that multi-terminal devices are more favorable for complex neural nwtwork emulation. In recent years, multi-terminal electric-double-layer (EDL) Thin-film transistors (TFTs) based on interfacial ion-modulation have attracted significant attention in mimicking synaptic dynamic plasticity and neural functions. Here, we provide the proof-of-principle artificial synapses/neurons based on solid electrolytes coupled oxide-based EDL TFTs with multiple driving and modulatory inputs terminals. Paired-pulse facilitation, dendritic integration and orientation tuning were successfully emulated. Additionally, neuronal gain control (arithmetic) in the scheme of rate coding is also experimentally demonstrated. Our results provide a new-concept approach for building brain-like cognitive systems.

A Retinotopic Spiking Neural Network System for Accurate Recognition of Moving Objects Using NeuCube and Dynamic Vision Sensors.

This paper introduces a new system for dynamic visual recognition that combines bio-inspired hardware with a brain-like spiking neural network. The system is designed to take data from a dynamic vision sensor (DVS) that simulates the functioning of the human retina by producing an address event output (spike trains) based on the movement of objects. The system then convolutes the spike trains and feeds them into a brain-like spiking neural network, called NeuCube, which is organized in a three-dimensional manner, representing the organization of the primary visual cortex. Spatio-temporal patterns of the data are learned during a deep unsupervised learning stage, using spike-timing-dependent plasticity. In a second stage, supervised learning is performed to train the network for classification tasks. The convolution algorithm and the mapping into the network mimic the function of retinal ganglion cells and the retinotopic organization of the visual cortex. The NeuCube architecture can be used to visualize the deep connectivity inside the network before, during, and after training and thereby allows for a better understanding of the learning processes. The method was tested on the benchmark MNIST-DVS dataset and achieved a classification accuracy of 92.90%. The paper discusses advantages and limitations of the new method and concludes that it is worth exploring further on different datasets, aiming for advances in dynamic computer vision and multimodal systems that integrate visual, aural, tactile, and other kinds of information in a biologically plausible way.

A Highly Effective and Robust Membrane Potential-Driven Supervised Learning Method for Spiking Neurons.

Spiking neurons are becoming increasingly popular owing to their biological plausibility and promising computational properties. Unlike traditional rate-based neural models, spiking neurons encode information in the temporal patterns of the transmitted spike trains, which makes them more suitable for processing spatiotemporal information. One of the fundamental computations of spiking neurons is to transform streams of input spike trains into precisely timed firing activity. However, the existing learning methods, used to realize such computation, often result in relatively low accuracy performance and poor robustness to noise. In order to address these limitations, we propose a novel highly effective and robust membrane potential-driven supervised learning (MemPo-Learn) method, which enables the trained neurons to generate desired spike trains with higher precision, higher efficiency, and better noise robustness than the current state-of-the-art spiking neuron learning methods. While the traditional spike-driven learning methods use an error function based on the difference between the actual and desired output spike trains, the proposed MemPo-Learn method employs an error function based on the difference between the output neuron membrane potential and its firing threshold. The efficiency of the proposed learning method is further improved through the introduction of an adaptive strategy, called skip scan training strategy, that selectively identifies the time steps when to apply weight adjustment. The proposed strategy enables the MemPo-Learn method to effectively and efficiently learn the desired output spike train even when much smaller time steps are used. In addition, the learning rule of MemPo-Learn is improved further to help mitigate the impact of the input noise on the timing accuracy and reliability of the neuron firing dynamics. The proposed learning method is thoroughly evaluated on synthetic data and is further demonstrated on real-world classification tasks. Experimental results show that the proposed method can achieve high learning accuracy with a significant improvement in learning time and better robustness to different types of noise.

Two Methods to Improve the Efficacy of ReSuMe

Neuron learning is the basis of the more complex learning of neuron network. Resume is the one of the most popular used supervised learning algorithm for spiking neurons. It corresponds to the Widrow-Hoff rule and its weights adjustment is derived from the spike-based Hebbian processes. Although it makes quite success the learning accuracy decreases quickly when the desired output spike train gets longer. This paper analyzes two important factors related to the learning convergence of Resume. And we proposed two methods based on them to improve the efficacy of Resume. The experimental results show that the two improved algorithms both can achieve better performance.

Spike and burst coding in thalamocortical relay cells.

Mammalian thalamocortical relay (TCR) neurons switch their firing activity between a tonic spiking and a bursting regime. In a combined experimental and computational study, we investigated the features in the input signal that single spikes and bursts in the output spike train represent and how this code is influenced by the membrane voltage state of the neuron. Identical frozen Gaussian noise current traces were injected into TCR neurons in rat brain slices as well as in a validated three-compartment TCR model cell. The resulting membrane voltage traces and spike trains were analyzed by calculating the coherence and impedance. Reverse correlation techniques gave the Event-Triggered Average (ETA) and the Event-Triggered Covariance (ETC). This demonstrated that the feature selectivity started relatively long before the events (up to 300 ms) and showed a clear distinction between spikes (selective for fluctuations) and bursts (selective for integration). The model cell was fine-tuned to mimic the frozen noise initiated spike and burst responses to within experimental accuracy, especially for the mixed mode regimes. The information content carried by the various types of events in the signal as well as by the whole signal was calculated. Bursts phase-lock to and transfer information at lower frequencies than single spikes. On depolarization the neuron transits smoothly from the predominantly bursting regime to a spiking regime, in which it is more sensitive to high-frequency fluctuations. The model was then used to elucidate properties that could not be assessed experimentally, in particular the role of two important subthreshold voltage-dependent currents: the low threshold activated calcium current (IT) and the cyclic nucleotide modulated h current (Ih). The ETAs of those currents and their underlying activation/inactivation states not only explained the state dependence of the firing regime but also the long-lasting concerted dynamic action of the two currents. Finally, the model was used to investigate the more realistic “high-conductance state”, where fluctuations are caused by (synaptic) conductance changes instead of current injection. Under “standard” conditions bursts are difficult to initiate, given the high degree of inactivation of the T-type calcium current. Strong and/or precisely timed inhibitory currents were able to remove this inactivation.

Identification of Linear and Nonlinear Sensory Processing Circuits from Spiking Neuron Data.

Inferring mathematical models of sensory processing systems directly from input-output observations, while making the fewest assumptions about the model equations and the types of measurements available, is still a major issue in computational neuroscience. This letter introduces two new approaches for identifying sensory circuit models consisting of linear and nonlinear filters in series with spiking neuron models, based only on the sampled analog input to the filter and the recorded spike train output of the spiking neuron. For an ideal integrate-and-fire neuron model, the first algorithm can identify the spiking neuron parameters as well as the structure and parameters of an arbitrary nonlinear filter connected to it. The second algorithm can identify the parameters of the more general leaky integrate-and-fire spiking neuron model, as well as the parameters of an arbitrary linear filter connected to it. Numerical studies involving simulated and real experimental recordings are used to demonstrate the applicability and evaluate the performance of the proposed algorithms.

A Spiking Neural Network Model of the Lateral Geniculate Nucleus on the SpiNNaker Machine.

We present a spiking neural network model of the thalamic Lateral Geniculate Nucleus (LGN) developed on SpiNNaker, which is a state-of-the-art digital neuromorphic hardware built with very-low-power ARM processors. The parallel, event-based data processing in SpiNNaker makes it viable for building massively parallel neuro-computational frameworks. The LGN model has 140 neurons representing a “basic building block” for larger modular architectures. The motivation of this work is to simulate biologically plausible LGN dynamics on SpiNNaker. Synaptic layout of the model is consistent with biology. The model response is validated with existing literature reporting entrainment in steady state visually evoked potentials (SSVEP)—brain oscillations corresponding to periodic visual stimuli recorded via electroencephalography (EEG). Periodic stimulus to the model is provided by: a synthetic spike-train with inter-spike-intervals in the range 10–50 Hz at a resolution of 1 Hz; and spike-train output from a state-of-the-art electronic retina subjected to a light emitting diode flashing at 10, 20, and 40 Hz, simulating real-world visual stimulus to the model. The resolution of simulation is 0.1 ms to ensure solution accuracy for the underlying differential equations defining Izhikevichs neuron model. Under this constraint, 1 s of model simulation time is executed in 10 s real time on SpiNNaker; this is because simulations on SpiNNaker work in real time for time-steps dt ⩾ 1 ms. The model output shows entrainment with both sets of input and contains harmonic components of the fundamental frequency. However, suppressing the feed-forward inhibition in the circuit produces subharmonics within the gamma band (>30 Hz) implying a reduced information transmission fidelity. These model predictions agree with recent lumped-parameter computational model-based predictions, using conventional computers. Scalability of the framework is demonstrated by a multi-node architecture consisting of three “nodes,” where each node is the “basic building block” LGN model. This 420 neuron model is tested with synthetic periodic stimulus at 10 Hz to all the nodes. The model output is the average of the outputs from all nodes, and conforms to the above-mentioned predictions of each node. Power consumption for model simulation on SpiNNaker is ≪1 W.

Estimating the Information Extracted by a Single Spiking Neuron from a Continuous Input Time Series.

Understanding the relation between (sensory) stimuli and the activity of neurons (i.e., “the neural code”) lies at heart of understanding the computational properties of the brain. However, quantifying the information between a stimulus and a spike train has proven to be challenging. We propose a new (in vitro) method to measure how much information a single neuron transfers from the input it receives to its output spike train. The input is generated by an artificial neural network that responds to a randomly appearing and disappearing “sensory stimulus”: the hidden state. The sum of this network activity is injected as current input into the neuron under investigation. The mutual information between the hidden state on the one hand and spike trains of the artificial network or the recorded spike train on the other hand can easily be estimated due to the binary shape of the hidden state. The characteristics of the input current, such as the time constant as a result of the (dis)appearance rate of the hidden state or the amplitude of the input current (the firing frequency of the neurons in the artificial network), can independently be varied. As an example, we apply this method to pyramidal neurons in the CA1 of mouse hippocampi and compare the recorded spike trains to the optimal response of the “Bayesian neuron” (BN). We conclude that like in the BN, information transfer in hippocampal pyramidal cells is non-linear and amplifying: the information loss between the artificial input and the output spike train is high if the input to the neuron (the firing of the artificial network) is not very informative about the hidden state. If the input to the neuron does contain a lot of information about the hidden state, the information loss is low. Moreover, neurons increase their firing rates in case the (dis)appearance rate is high, so that the (relative) amount of transferred information stays constant.

Solving the two-dimensional Fokker-Planck equation for strongly correlated neurons.

Pairs of neurons in brain networks often share much of the input they receive from other neurons. Due to essential nonlinearities of the neuronal dynamics, the consequences for the correlation of the output spike trains are generally not well understood. Here we analyze the case of two leaky integrate-and-fire neurons using an approach which is nonperturbative with respect to the degree of input correlation. Our treatment covers both weakly and strongly correlated dynamics, generalizing previous results based on linear response theory.

Correlation between discharge timings of pairs of motor units reveals the presence but not the proportion of common synaptic input to motor neurons.

We investigated whether correlation measures derived from pairs of motor unit (MU) spike trains are reliable indicators of the degree of common synaptic input to motor neurons. Several 50-s isometric contractions of the biceps brachii muscle were performed at different target forces ranging from 10 to 30% of the maximal voluntary contraction relying on force feedback. Forty-eight pairs of MUs were examined at various force levels. Motor unit synchrony was assessed by cross-correlation analysis using three indexes: the output correlation as the peak of the cross-histogram (ρ) and the number of synchronous spikes per second (CIS) and per trigger (E). Individual analysis of MU pairs revealed that ρ, CIS, and E were most often positively associated with discharge rate (87, 85, and 76% of the MU pairs, respectively) and negatively with interspike interval variability (69, 65, and 62% of the MU pairs, respectively). Moreover, the behavior of synchronization indexes with discharge rate (and interspike interval variability) varied greatly among the MU pairs. These results were consistent with theoretical predictions, which showed that the output correlation between pairs of spike trains depends on the statistics of the input current and motor neuron intrinsic properties that differ for different motor neuron pairs. In conclusion, the synchronization between MU firing trains is necessarily caused by the (functional) common input to motor neurons, but it is not possible to infer the degree of shared common input to a pair of motor neurons on the basis of correlation measures of their output spike trains.NEW & NOTEWORTHY The strength of correlation between output spike trains is only poorly associated with the degree of common input to the population of motor neurons. The synchronization between motor unit firing trains is necessarily caused by the (functional) common input to motor neurons, but it is not possible to infer the degree of shared common input to a pair of motor neurons on the basis of correlation measures of their output spike trains.

Directed Information Between Connected Leaky Integrate-and-Fire Neurons

The connectivity structure between neurons is useful for determining how groups of neurons perform tasks. Directed information is a measure that can be used to infer connectivity between neurons using their recorded time series. In this paper, we develop a method of calculating the directed information rate from one neuron to another neuron it is connected to, given a particular neuronal topology. We assume a leaky integrate-and-fire (LIF) neuron model with independent and identically distributed random spike train inputs, which governs how the membrane potential of the output neuron evolves. We use this neuron model to find the dynamics of the resulting output spike train from its membrane potential dynamics, both for when the past of the input neuron is observed and when it is not. We show that an action potential in the LIF model causes a conditional independence of the activity before and after it, and we capture this conditional independence via a Markov model. We use these spike train dynamics to then calculate the directed information between the spike train of the input neuron to the spike train generated by the LIF model. In addition, we show how changing the refractory period of the LIF model affects the directed information, and also how the spike train dynamics are affected by memory constraints, which are commonly imposed in estimators of directed information.

Analysis of Encoding Degradation in Spiking Sensors Due to Spike Delay Variation

Spiking sensors such as the silicon retina and cochlea encode analog signals into massively parallel asynchronous spike train output where the information is contained in the precise spike timing. The variation of the spike timing that arises from spike transmission degrades signal encoding quality. Using the signal-to-distortion ratio (SDR) metric with nonlinear spike train decoding based on frame theory, two particular sources of delay variation including comparison delay $T_{\mathbf {DC}}$ and queueing delay $T_{\mathbf {DQ}}$ are evaluated on two encoding mechanisms which have been used for implementations of silicon array spiking sensors: asynchronous delta modulation and self-timed reset. As specific examples, $T_{\mathbf {DC}}$ is obtained from a 2T current-mode comparator, and $T_{\mathbf {DQ}}$ is obtained from an M/D/1 queue for 1-D sensors like the silicon cochlea and an $\text {M}^{\mathrm {\mathbf {X}}}$ /D/1 queue for 2-D sensors like the silicon retina. Quantitative relations between the SDR and the circuit and system parameters of spiking sensors are established. The analysis method presented in this work will be useful for future specifications-guided designs of spiking sensors.

On the applicability of STDP-based learning mechanisms to spiking neuron network models

The ways to creating practically effective method for spiking neuron networks learning, that would be appropriate for implementing in neuromorphic hardware and at the same time based on the biologically plausible plasticity rules, namely, on STDP, are discussed. The influence of the amount of correlation between input and output spike trains on the learnability by different STDP rules is evaluated. A usability of alternative combined learning schemes, involving artificial and spiking neuron models is demonstrated on the iris benchmark task and on the practical task of gender recognition.

Inverse Stochastic Resonance in Cerebellar Purkinje Cells.

Purkinje neurons play an important role in cerebellar computation since their axons are the only projection from the cerebellar cortex to deeper cerebellar structures. They have complex internal dynamics, which allow them to fire spontaneously, display bistability, and also to be involved in network phenomena such as high frequency oscillations and travelling waves. Purkinje cells exhibit type II excitability, which can be revealed by a discontinuity in their f-I curves. We show that this excitability mechanism allows Purkinje cells to be efficiently inhibited by noise of a particular variance, a phenomenon known as inverse stochastic resonance (ISR). While ISR has been described in theoretical models of single neurons, here we provide the first experimental evidence for this effect. We find that an adaptive exponential integrate-and-fire model fitted to the basic Purkinje cell characteristics using a modified dynamic IV method displays ISR and bistability between the resting state and a repetitive activity limit cycle. ISR allows the Purkinje cell to operate in different functional regimes: the all-or-none toggle or the linear filter mode, depending on the variance of the synaptic input. We propose that synaptic noise allows Purkinje cells to quickly switch between these functional regimes. Using mutual information analysis, we demonstrate that ISR can lead to a locally optimal information transfer between the input and output spike train of the Purkinje cell. These results provide the first experimental evidence for ISR and suggest a functional role for ISR in cerebellar information processing.

The human motor neuron pools receive a dominant slow-varying common synaptic input.

Motor neurons in a pool receive both common and independent synaptic inputs, although the proportion and role of their common synaptic input is debated. Classic correlation techniques between motor unit spike trains do not measure the absolute proportion of common input and have limitations as a result of the non-linearity of motor neurons. We propose a method that for the first time allows an accurate quantification of the absolute proportion of low frequency common synaptic input (<5Hz) to motor neurons in humans. We applied the proposed method to three human muscles and determined experimentally that they receive a similar large amount (>60%) of common input, irrespective of their different functional and control properties. These results increase our knowledge about the role of common and independent input to motor neurons in force control. Motor neurons receive both common and independent synaptic inputs. This observation is classically based on the presence of a significant correlation between pairs of motor unit spike trains. The functional significance of different relative proportions of common input across muscles, individuals and conditions is still debated. One of the limitations in our understanding of correlated input to motor neurons is that it has not been possible so far to quantify the absolute proportion of common input with respect to the total synaptic input received by the motor neurons. Indeed, correlation measures of pairs of output spike trains only allow for relative comparisons. In the present study, we report for the first time an approach for measuring the proportion of common input in the low frequency bandwidth (<5Hz) to a motor neuron pool in humans. This estimate is based on a phenomenological model and the theoretical fitting of the experimental values of coherence between the permutations of groups of motor unit spike trains. We demonstrate the validity of this theoretical estimate with several simulations. Moreover, we applied this method to three human muscles: the abductor digiti minimi, tibialis anterior and vastus medialis. Despite these muscles having different functional roles and control properties, as confirmed by the results of the present study, we estimate that their motor pools receive a similar and large (>60%) proportion of common low frequency oscillations with respect to their total synaptic input. These results suggest that the central nervous system provides a large amount of common input to motor neuron pools, in a similar way to that for muscles with different functional and control properties.

Reconstruction of neuronal input through modeling single-neuron dynamics and computations.

Mathematical models provide a mathematical description of neuron activity, which can better understand and quantify neural computations and corresponding biophysical mechanisms evoked by stimulus. In this paper, based on the output spike train evoked by the acupuncture mechanical stimulus, we present two different levels of models to describe the input-output system to achieve the reconstruction of neuronal input. The reconstruction process is divided into two steps: First, considering the neuronal spiking event as a Gamma stochastic process. The scale parameter and the shape parameter of Gamma process are, respectively, defined as two spiking characteristics, which are estimated by a state-space method. Then, leaky integrate-and-fire (LIF) model is used to mimic the response system and the estimated spiking characteristics are transformed into two temporal input parameters of LIF model, through two conversion formulas. We test this reconstruction method by three different groups of simulation data. All three groups of estimates reconstruct input parameters with fairly high accuracy. We then use this reconstruction method to estimate the non-measurable acupuncture input parameters. Results show that under three different frequencies of acupuncture stimulus conditions, estimated input parameters have an obvious difference. The higher the frequency of the acupuncture stimulus is, the higher the accuracy of reconstruction is.

Enhanced Sensitivity to Rapid Input Fluctuations by Nonlinear Threshold Dynamics in Neocortical Pyramidal Neurons.

The way in which single neurons transform input into output spike trains has fundamental consequences for network coding. Theories and modeling studies based on standard Integrate-and-Fire models implicitly assume that, in response to increasingly strong inputs, neurons modify their coding strategy by progressively reducing their selective sensitivity to rapid input fluctuations. Combining mathematical modeling with in vitro experiments, we demonstrate that, in L5 pyramidal neurons, the firing threshold dynamics adaptively adjust the effective timescale of somatic integration in order to preserve sensitivity to rapid signals over a broad range of input statistics. For that, a new Generalized Integrate-and-Fire model featuring nonlinear firing threshold dynamics and conductance-based adaptation is introduced that outperforms state-of-the-art neuron models in predicting the spiking activity of neurons responding to a variety of in vivo-like fluctuating currents. Our model allows for efficient parameter extraction and can be analytically mapped to a Generalized Linear Model in which both the input filter—describing somatic integration—and the spike-history filter—accounting for spike-frequency adaptation—dynamically adapt to the input statistics, as experimentally observed. Overall, our results provide new insights on the computational role of different biophysical processes known to underlie adaptive coding in single neurons and support previous theoretical findings indicating that the nonlinear dynamics of the firing threshold due to Na+-channel inactivation regulate the sensitivity to rapid input fluctuations.