Assessment of impaired consciousness using EEG-based connectivity features and convolutional neural networks.

Growing electroencephalogram (EEG) studies have linked the abnormities of functional brain networks with disorders of consciousness (DOC). However, due to network data's high-dimensional and non-Euclidean properties, it is difficult to exploit the brain connectivity information that can effectively detect the consciousness levels of DOC patients via deep learning. To take maximum advantage of network information in assessing impaired consciousness, we utilized the functional connectivity with convolutional neural network (CNN) and employed three rearrangement schemes to improve the evaluation performance of brain networks. In addition, the gradient-weighted class activation mapping (Grad-CAM) was adopted to visualize the classification contributions of connections among different areas. We demonstrated that the classification performance was significantly enhanced by applying network rearrangement techniques compared to those obtained by the original connectivity matrix (with an accuracy of 75.0%). The highest classification accuracy (87.2%) was achieved by rearranging the alpha network based on the anatomical regions. The inter-region connections (i.e., frontal-parietal and frontal-occipital connectivity) played dominant roles in the classification of patients with different consciousness states. The effectiveness of functional connectivity in revealing individual differences in brain activity was further validated by the correlation between behavioral performance and connections among specific regions. These findings suggest that our proposed assessment model could detect the residual consciousness of patients.

Frequency-dependent membrane polarization across neocortical cell types and subcellular elements by transcranial alternating current stimulation.

Objective.Transcranial alternating current stimulation (tACS) is a non-invasive brain stimulation technique that directly interacts with ongoing brain oscillations in a frequency-dependent manner. However, it remains largely unclear how the cellular effects of tACS vary between cell types and subcellular elements.Approach.In this study, we use a set of morphologically realistic models of neocortical neurons to simulate the cellular response to uniform oscillating electric fields (EFs). We systematically characterize the membrane polarization in the soma, axons, and dendrites with varying field directions, intensities, and frequencies.Main results.Pyramidal cells are more sensitive to axial EF that is roughly parallel to the cortical column, while interneurons are sensitive to axial EF and transverse EF that is tangent to the cortical surface. Membrane polarization in each subcellular element increases linearly with EF intensity, and its slope, i.e. polarization length, highly depends on the stimulation frequency. At each frequency, pyramidal cells are more polarized than interneurons. Axons usually experience the highest polarization, followed by the dendrites and soma. Moreover, a visible frequency resonance presents in the apical dendrites of pyramidal cells, while the other subcellular elements primarily exhibit low-pass filtering properties. In contrast, each subcellular element of interneurons exhibits complex frequency-dependent polarization. Polarization phase in each subcellular element of cortical neurons lags that of field and exhibits high-pass filtering properties. These results demonstrate that the membrane polarization is not only frequency-dependent, but also cell type- and subcellular element-specific. Through relating effective length and ion mechanism with polarization, we emphasize the crucial role of cell morphology and biophysics in determining the frequency-dependent membrane polarization.Significance.Our findings highlight the diverse polarization patterns across cell types as well as subcellular elements, which provide some insights into the tACS cellular effects and should be considered when understanding the neural spiking activity by tACS.

Electric field effects on neuronal input-output relationship by regulating NMDA spikes.

Evidence shows that the dendritic polarization induced by weak electrical field (EF) can affect the neuronal input-output function via modulating dendritic integration of AMPA synapses, indicating that the supralinear dendritic integration of NMDA synapses can also be influenced by dendritic polarization. However, it remains unknown how dendritic polarization affects NMDA-type dendritic integration, and then contributes to neuronal input-output relationship. Here, we used a computational model of pyramidal neuron with inhomogeneous extracellular potentials to characterize the relationship among EF, dendritic integration, and somatic output. Basing on singular perturbation we analyzed the subthreshold dynamics of membrane potentials in response to NMDA synapses, and found that the equilibrium mapping of a fast subsystem can characterize the asymptotic subthreshold input-output (sI/O) relationship for EF-regulated supralinear dendritic integration, allowing us to predict the tendency of EF-regulated dendritic integration by showing the variation of equilibrium mapping under EF stimulation. EF-induced depolarization at distal dendrites receiving synapses plays a crucial role in shifting the steep change of sI/O left by facilitating dendritic NMDA spike generation and in decreasing the plateau of sI/O via reducing driving force. And more effective EF modulation appears at sparsely activated NMDA receptors compared with clustered synaptic inputs. During the action potential (AP) generation, the respective contribution of EF-regulated dendritic integration and EF-induced somatic polarization was identified to show their synergetic or antagonistic effect on AP generation, depending on neuronal excitability. These results provided insight in understanding the modulation effect of EF on neuronal computation, which is important for optimizing noninvasive brain stimulation. Supplementary Information: The online version contains supplementary material available at 10.1007/s11571-022-09922-y.

Scalable Multi-Hierarchy Embedded Platform for Neural Population Simulations.

Brain-inspired structured neural circuits are the cornerstones of both computational and perceived intelligence. Real-time simulations of large-scale high-dimensional neural populations with complex nonlinearities pose a significant challenge. Taking advantage of distributed computations using embedded multi-cores, we propose an ARM-based scalable multi-hierarchy parallel computing platform (EmPaas) for neural population simulations. EmPaas is constructed using 340 ARM Cortex-M4 microprocessors to achieve high-speed and high-accuracy parallel computing. The tree-two-dimensional grid-like hybrid topology completes the overall construction, reducing communication strain and power consumption. As an instance of embedded computing, the optimized model for a biologically plausible basal ganglia-thalamus (BG-TH) network is deployed into this platform to verify the performance. At an operating frequency of 168 MHz, the BG-TH network consisting of 4000 Izhikevich neurons is simulated in the platform for 3000 ms with a power consumption of 56.565 mW per core and an actual time of 2748.57 ms, which shows the parallel computing approach significantly improves computational efficiency. EmPaas can meet the requirement of real-time performance with the maximum amount of 2000 Izhikevich neurons loaded in each Extended Community Unit (ECUnit), which provides a new practical method for research in large-scale brain network simulation and brain-inspired computing.

State-dependent modulation of low-threshold-current-regulated dendritic Ca2+ response in thalamic reticular neurons with extracellular electric fields.

Deep brain stimulation (DBS) in thalamic reticular nucleus (TRN) neuron provides a novel treatment for drug-resistant epilepsy via the induced electrical field (EFs). However, the mechanisms underlying EF effects remain unclear. This paper investigated how EFs regulate low-threshold dendritic Ca2+ (dCa) response and thus contribute to the input-output relationship of TRN cell. Our results showed that EFs modulate firing modes differently in a neuronal state-dependent manner. At the depolarized state, EFs only regulate the spike timing of a somatic stimulus-evoked single action potential (AP) with less contribution in the regulation of dCa response but could induce the transition between a dendritic stimulus-evoked single AP and a tonic burst of APs via the moderate regulation of dCa response. At the hyperpolarized state, EFs have significant effects on the dCa response, which modulate the large dCa response-dependent burst discharge and even cause a transition from this type of burst discharge to a single AP with less dCa response. Moreover, EF effects on stimulation threshold of somatic spiking prominently depend on EF-regulated dCa responses and the onset time differences between the stimulus and EF give rise to the distinct effect in the EF regulation of dCa responses. Finally, the larger neuronal axial resistance tends to result in the dendritic stimulus-evoked dCa response independent of somatic state. Interestingly, in this case, the EF application could reproduce the similar somatic state-dependent dCa response to dendritic stimulus which occurs in the case of lower axial resistance. These results suggest that the influence of EF on neuronal activities depends on neuronal intrinsic properties, which provides insight into understanding how DBS in TRN neuron modulates epilepsy from the point of view of biophysics.

BrainS: Customized multi-core embedded multiple scale neuromorphic system.

Research on modeling and mechanisms of the brain remains the most urgent and challenging task. The customized embedded neuromorphic system is one of the most effective approaches for multi-scale simulations ranging from ion channel to network. This paper proposes BrainS, a scalable multi-core embedded neuromorphic system capable of accommodating massive and large-scale simulations. It is designed with rich external extension interfaces to support various types of input/output and communication requirements. The 3D mesh-based topology with an efficient memory access mechanism makes exploring the properties of neuronal networks possible. BrainS operates at 168 MHz and contains a model database ranging from ion channel to network scale within the Fundamental Computing Unit (FCU). At the ion channel scale, the Basic Community Unit (BCU) can perform real-time simulations of a Hodgkin-Huxley (HH) neuron with 16000 ion channels, using 125.54 KB of the SRAM. When the number of ion channels is within 64000, the HH neuron is simulated in real-time by 4 BCUs. At the network scale, the basal ganglia-thalamus (BG-TH) network consisting of 3200 Izhikevich neurons, providing a vital motor regulation function, is simulated in 4 BCUs with a power consumption of 364.8 mW. Overall, BrainS has an excellent performance in real-time and flexible configurability, providing an embedded application solution for multi-scale simulation.

Aberrant temporal correlations of ongoing oscillations in disorders of consciousness on multiple time scales.

Changes in neural oscillation amplitude across states of consciousness has been widely reported, but little is known about the link between temporal dynamics of these oscillations on different time scales and consciousness levels. To address this question, we analyzed amplitude fluctuation of the oscillations extracted from spontaneous resting-state EEG recorded from the patients with disorders of consciousness (DOC) and healthy controls. Detrended fluctuation analysis (DFA) and measures of life-time and waiting-time were employed to characterize the temporal structure of EEG oscillations on long time scales (1-20 s) and short time scales (< 1 s), in groups with different consciousness states: patients in minimally conscious state (MCS), patients with unresponsive wakefulness syndrome (UWS) and healthy subjects. Results revealed increased DFA exponents that implies higher long-range temporal correlations (LRTC), especially in the central brain area in alpha and beta bands. On short time scales, declined bursts of oscillations were also observed. All the metrics exhibited lower individual variability in the UWS or MCS group, which may be attributed to the reduced spatial variability of oscillation dynamics. In addition, the temporal dynamics of EEG oscillations showed significant correlations with the behavioral responsiveness of patients. In summary, our findings shows that loss of consciousness is accompanied by alternation of temporal structure in neural oscillations on multiple time scales, and thus may help uncover the mechanism of underlying neuronal correlates of consciousness. Supplementary Information: The online version contains supplementary material available at 10.1007/s11571-022-09852-9.

Nonlinear dynamical modeling of neural activity using volterra series with GA-enhanced particle swarm optimization algorithm.

In order to improve the modeling performance of Volterra sequence for nonlinear neural activity, in this paper, a new optimization algorithm is proposed to identify Volterra sequence parameters. Algorithm combines the advantages of particle swarm optimization (PSO) and genetic algorithm (GA) improve the performance of the identification of nonlinear model parameters from rapidity and accuracy. In the modeling experiments of neural signal data generated by the neural computing model and clinical neural data set in this paper, the proposed algorithm shows its excellent potential in nonlinear neural activity modeling. Compared with PSO and GA, the algorithm can achieve less identification error, and better balance the convergence speed and identification error. Further, we explore the influence of algorithm parameters on identification efficiency, which provides possible guiding significance for parameter setting in practical application of the algorithm.

Disrupted Control Architecture of Brain Network in Disorder of Consciousness.

The human brain controls various cognitive functions via the functional coordination of multiple brain regions in an efficient and robust way. However, the relationship between consciousness state and the control mode of brain networks is poorly explored. Using multi-channel EEG, the present study aimed to characterize the abnormal control architecture of functional brain networks in the patients with disorders of consciousness (DOC). Resting state EEG data were collected from 40 DOC patients with different consciousness levels and 24 healthy subjects. Functional brain networks were constructed in five different EEG frequency bands and the broadband in the source level. Subsequently, a control architecture framework based on the minimum dominating set was applied to investigate the of control mode of functional brain networks for the subjects with different conscious states. Results showed that regardless of the consciousness levels, the functional networks of human brain operate in a distributed and overlapping control architecture different from that of random networks. Compared to the healthy controls, the patients have a higher control cost manifested by more minimum dominating nodes and increased degree of distributed control, especially in the alpha band. The ability to withstand network attack for the control architecture is positive correlated with the consciousness levels. The distributed of control increased correlation levels with Coma Recovery Scale-Revised score and improved separation between unresponsive wakefulness syndrome and minimal consciousness state. These findings may benefit our understanding of consciousness and provide potential biomarkers for the assessment of consciousness levels.

BiCoSS: Toward Large-Scale Cognition Brain With Multigranular Neuromorphic Architecture.

The further exploration of the neural mechanisms underlying the biological activities of the human brain depends on the development of large-scale spiking neural networks (SNNs) with different categories at different levels, as well as the corresponding computing platforms. Neuromorphic engineering provides approaches to high-performance biologically plausible computational paradigms inspired by neural systems. In this article, we present a biological-inspired cognitive supercomputing system (BiCoSS) that integrates multiple granules (GRs) of SNNs to realize a hybrid compatible neuromorphic platform. A scalable hierarchical heterogeneous multicore architecture is presented, and a synergistic routing scheme for hybrid neural information is proposed. The BiCoSS system can accommodate different levels of GRs and biological plausibility of SNN models in an efficient and scalable manner. Over four million neurons can be realized on BiCoSS with a power efficiency of 2.8k larger than the GPU platform, and the average latency of BiCoSS is 3.62 and 2.49 times higher than conventional architectures of digital neuromorphic systems. For the verification, BiCoSS is used to replicate various biological cognitive activities, including motor learning, action selection, context-dependent learning, and movement disorders. Comprehensively considering the programmability, biological plausibility, learning capability, computational power, and scalability, BiCoSS is shown to outperform the alternative state-of-the-art works for large-scale SNN, while its real-time computational capability enables a wide range of potential applications.

Asymptotic Input-Output Relationship Predicts Electric Field Effect on Sublinear Dendritic Integration of AMPA Synapses.

An extracellular electric field (EF) induces transmembrane polarizations on extremely inhomogeneous spaces. Evidence shows that EF-induced somatic polarization in pyramidal cells can modulate the neuronal input-output (I/O) function. However, it remains unclear whether and how dendritic polarization participates in the dendritic integration and contributes to the neuronal I/O function. To this end, we built a computational model of a simplified pyramidal cell with multi-dendritic tufts, one dendritic trunk, and one soma to describe the interactions among EF, dendritic integration, and somatic output, in which the EFs were modeled by inserting inhomogeneous extracellular potentials. We aimed to establish the underlying relationship between dendritic polarization and dendritic integration by analyzing the dynamics of subthreshold membrane potentials in response to AMPA synapses in the presence of constant EFs. The model-based singular perturbation analysis showed that the equilibrium mapping of a fast subsystem can serve as the asymptotic subthreshold I/O relationship for sublinear dendritic integration. This allows us to predict the tendency of EF-mediated dendritic integration by showing how EF changes modify equilibrium mapping. EF-induced hyperpolarization of distal dendrites receiving synapses inputs was found to play a key role in facilitating the AMPA receptor-evoked excitatory postsynaptic potential (EPSP) by enhancing the driving force of synaptic inputs. A significantly higher efficacy of EF modulation effect on global AMPA-type dendritic integration was found compared with local AMPA-type dendritic integration. During the generation of an action potential (AP), the relative contribution of EF-modulated dendritic integration and EF-induced somatic polarization was determined to show their collaboration in promoting or inhibiting the somatic excitability, depending on the EF polarity. These findings are crucial for understanding the EF modulation effect on neuronal computation, which provides insight into the modulation mechanism of noninvasive brain modulation.

A Data Driven Experimental System for Individualized Brain Stimulation Design and Validation.

Deep brain stimulation (DBS) is an effective clinical treatment for epilepsy. However, the individualized setting and adaptive adjustment of DBS parameters are still facing great challenges. This paper investigates a data-driven hardware-in-the-loop (HIL) experimental system for closed-loop brain stimulation system individualized design and validation. The unscented Kalman filter (UKF) is utilized to estimate critical parameters of neural mass model (NMM) from the electroencephalogram recordings to reconstruct individual neural activity. Based on the reconstructed NMM, we build a digital signal processor (DSP) based virtual brain platform with real time scale and biological signal level scale. Then, the corresponding hardware parts of signal amplification detection and closed-loop controller are designed to form the HIL experimental system. Based on the designed experimental system, the proportional-integral controller for different individual NMM is designed and validated, which proves the effectiveness of the experimental system. This experimental system provides a platform to explore neural activity under brain stimulation and the effects of various closed-loop stimulation paradigms.

An Embedded Multi-Core Real-Time Simulation Platform of Basal Ganglia for Deep Brain Stimulation.

Closed-loop deep brain stimulation (DBS) paradigm is gaining tremendous favor due to its potential capability of further and more efficient improvements in neurological diseases. Preclinical validation of closed-loop controller is quite necessary in order to minimize injury risks of clinical trials to patients, which can greatly benefit from real-time computational models and thus potentially reduce research and development costs and time. Here we developed an embedded multi-core real-time simulation platform (EMC-RTP) for a biological-faithful computational network model of basal ganglia (BG). The single neuron model is implemented in a highly real-time manner using a reasonable simplification. A modular mapping architecture with hierarchical routing organization was constructed to mimic the pathological neural activities of BG observed in parkinsonian conditions. A closed-loop simulation testbed for DBS validation was then set up using a host computer as the DBS controller. The availability of EMC-RTP and the testbed system was validated by comparing the performance of open-loop and proportional-integral (PI) controllers. Our experimental results showed that the proposed EMC-RTP reproduces abnormal beta bursts of BG in parkinsonian conditions while meets requirements of both real-time and computational accuracy as well. Closed-loop DBS experiments using the EMC-RTP suggested that the platform could perform reasonable output under different kinds of DBS strategies, indicating the usability of the platform.

Neural recruitment by ephaptic coupling in epilepsy.

OBJECTIVE: One of the challenges in treating patients with drug-resistant epilepsy is that the mechanisms of seizures are unknown. Most current interventions are based on the assumption that epileptic activity recruits neurons and progresses by synaptic transmission. However, several experimental studies have shown that neural activity in rodent hippocampi can propagate independently of synaptic transmission. Recent studies suggest these waves are self-propagating by electric field (ephaptic) coupling. In this study, we tested the hypothesis that neural recruitment during seizures can occur by electric field coupling. METHODS: 4-Aminopyridine was used in both in vivo and in vitro preparation to trigger seizures or epileptiform activity. A transection was made in the in vivo hippocampus and in vitro hippocampal and cortical slices to study whether the induced seizure activity can recruit neurons across the gap. A computational model was built to test whether ephaptic coupling alone can account for neural recruitment across the transection. The model prediction was further validated by in vitro experiments. RESULTS: Experimental results show that electric fields generated by seizure-like activity in the hippocampus both in vitro and in vivo can recruit neurons locally and through a transection of the tissue. The computational model suggests that the neural recruitment across the transection is mediated by electric field coupling. With in vitro experiments, we show that a dielectric material can block the recruitment of epileptiform activity across a transection, and that the electric fields measured within the gap are similar to those predicted by model simulations. Furthermore, this nonsynaptic neural recruitment is also observed in cortical slices, suggesting that this effect is robust in brain tissue. SIGNIFICANCE: These results indicate that ephaptic coupling, a nonsynaptic mechanism, can underlie neural recruitment by a small electric field generated by seizure activity and could explain the low success rate of surgical transections in epilepsy patients.

Model Predictive Control for Seizure Suppression Based on Nonlinear Auto-Regressive Moving-Average Volterra Model.

This article investigates a closed-loop brain stimulation method based on model predictive control strategy to suppress epileptic seizures. A neural mass model (NMM), exhibiting the normal and various epileptic seizures by changing physiologically meaningful parameters, is used as a black-box model of the brain. Based on system identification, an auto-regressive moving-average Volterra model is established to reveal the relationship between stimulation and neuronal responses. Then, the model predictive control strategy is implemented based the Volterra model, which can generate an optimal stimulation waveform to eliminate epileptiform waves. The computational simulation results indicate the proposed closed-loop control strategy can optimize the stimulation waveform without particular knowledge of the physiological properties of the system. The robustness of the proposed control strategy to system disturbances makes it more appropriate for future clinical application.

Altered inter-frequency dynamics of brain networks in disorder of consciousness.

OBJECTIVE: Growing evidence have linked disorders of consciousness (DOC) with the changes in frequency-specific functional networks. However, the alteration of inter-frequency dynamics in brain networks remain largely unknown. In this study, we investigated the network integration and segregation across frequency bands in a multiplex network framework. APPROACH: Resting-state EEG data were recorded and analysed from 19 patients in minimally conscious state, 35 patients in unresponsive wakefulness syndrome (UWS) and 23 healthy controls. Frequency-based multiplex (cross-frequency) networks were reconstructed by integrating the five frequency-specific networks. Multiplex graph metrics, named multiplex participation coefficient and multiplex clustering coefficient, were employed to assess the network topology of subjects with different levels of consciousness. MAIN RESULTS: Results revealed DOC networks, compared to those of healthy controls, may work at a less optimal point (closer to complete disorder) with increased integration and decreased segregation considering inter-frequency dynamics. Both metrics show increased spatial and temporal variability with the consciousness levels. Moreover, significant correlation can be found between the alteration of cross-frequency networks in DOC patients and their behavioural performance at both local and global scales. SIGNIFICANCE: These findings may contribute to the development of EEG network study and benefit our understanding of the processes of consciousness and their pathophysiology for DOC.

Functional Integration and Segregation in Multiplex Brain Networks for Alzheimer's Disease.

Growing evidence links impairment of brain functions in Alzheimer's disease (AD) with disruptions of brain functional connectivity. However, whether the AD brain shows similar changes from a dynamic or cross-frequency view remains poorly explored. This paper provides an effective framework to investigate the properties of multiplex brain networks in AD considering inter-frequency and temporal dynamics. Using resting-state EEG signals, two types of multiplex networks were reconstructed separately considering the network interactions between different frequency bands or time points. We further applied multiplex network features to characterize functional integration and segregation of the cross-frequency or time-varying networks. Finally, machine learning methods were employed to evaluate the performance of multiplex-network-based indexes for detection of AD. Results revealed that the brain networks of AD patients are disrupted with reduced segregation particularly in the left occipital area for both cross-frequency and time-varying networks. However, the alteration of integration differs among brain regions and may show an increasing trend in the frontal area of AD brain. By combining the features of integration and segregation in time-varying networks, the best classification performance was achieved with an accuracy of 92.5%. These findings suggest that our multiplex framework can be applied to explore functional integration and segregation of brain networks and characterize the abnormalities of brain function. This may shed new light on the brain network analysis and extend our understanding of brain function in patients with neurological diseases.

Comments and Corrections Corrections to "Application of Reinforcement Learning to Deep Brain Stimulation in a Computational Model of Parkinson's Disease".

In the above article [1], financial support was incorrectly published. The correct information is as follows: This work was supported in part by the National Natural Science Foundation of China under Grant 61501330 and Grant 61771330, and in part by the Tianjin Municipal Special Program of Talents Development for Excellent Youth Scholars under Grant TJTZJH-QNBJRC-2-2.

Characterization of network switching in disorder of consciousness at multiple time scales.

OBJECTIVE: Recent works have shown that flexible information processing is closely related to the reconfiguration of human brain networks underlying brain functions. However, the role of network switching for consciousness is poorly explored and whether such transition can indicate the behavioral performance of patients with disorders of consciousness (DOC) remains unknown. Here, we investigate the relationship between the switching of brain networks (states) over time and the consciousness levels. APPROACH: By applying multilayer network methods, we calculated time-resolved functional connectivity from source-level EEG data in different frequency bands. At various time scales, we explored how the human brain changes its community structure and traverses across defined network states (integrated and segregated states) in subjects with different consciousness levels. MAIN RESULTS: Network switching in the human brain is decreased with increasing time scale opposite to that in random systems. Transitions of community assignment (denoted by flexibility) are negatively correlated with the consciousness levels (particularly in the alpha band) at short time scales. At long time scales, the opposite trend is found. Compared to healthy controls, patients show a new balance between dynamic segregation and integration, with decreased proportion and mean duration of segregated state (contrary to those of integrated state) at small scales. SIGNIFICANCE: These findings may contribute to the development of EEG-based network analysis and shed new light on the pathological mechanisms of neurological disorders like DOC.

Training Spiking Neural Networks for Cognitive Tasks: A Versatile Framework Compatible With Various Temporal Codes.

Recent studies have demonstrated the effectiveness of supervised learning in spiking neural networks (SNNs). A trainable SNN provides a valuable tool not only for engineering applications but also for theoretical neuroscience studies. Here, we propose a modified SpikeProp learning algorithm, which ensures better learning stability for SNNs and provides more diverse network structures and coding schemes. Specifically, we designed a spike gradient threshold rule to solve the well-known gradient exploding problem in SNN training. In addition, regulation rules on firing rates and connection weights are proposed to control the network activity during training. Based on these rules, biologically realistic features such as lateral connections, complex synaptic dynamics, and sparse activities are included in the network to facilitate neural computation. We demonstrate the versatility of this framework by implementing three well-known temporal codes for different types of cognitive tasks, namely, handwritten digit recognition, spatial coordinate transformation, and motor sequence generation. Several important features observed in experimental studies, such as selective activity, excitatory-inhibitory balance, and weak pairwise correlation, emerged in the trained model. This agreement between experimental and computational results further confirmed the importance of these features in neural function. This work provides a new framework, in which various neural behaviors can be modeled and the underlying computational mechanisms can be studied.