Firing Threshold Research Articles

Recurrent activity in neuronal avalanches

A new statistical analysis of large neuronal avalanches observed in mouse and rat brain tissues reveals a substantial degree of recurrent activity and cyclic patterns of activation not seen in smaller avalanches. To explain these observations, we adapted a model of structural weakening in materials. In this model, dynamical weakening of neuron firing thresholds closely replicates experimental avalanche size distributions, firing number distributions, and patterns of cyclic activity. This agreement between model and data suggests that a mechanism like dynamical weakening plays a key role in recurrent activity found in large neuronal avalanches. We expect these results to illuminate the causes and dynamics of large avalanches, like those seen in seizures.

Constrain Bias Addition to Train Low-Latency Spiking Neural Networks.

In recent years, a third-generation neural network, namely, spiking neural network, has received plethora of attention in the broad areas of Machine learning and Artificial Intelligence. In this paper, a novel differential-based encoding method is proposed and new spike-based learning rules for backpropagation is derived by constraining the addition of bias voltage in spiking neurons. The proposed differential encoding method can effectively exploit the correlation between the data and improve the performance of the proposed model, and the new learning rule can take complete advantage of the modulation properties of bias on the spike firing threshold. We experiment with the proposed model on the environmental sound dataset RWCP and the image dataset MNIST and Fashion-MNIST, respectively, and assign various conditions to test the learning ability and robustness of the proposed model. The experimental results demonstrate that the proposed model achieves near-optimal results with a smaller time step by maintaining the highest accuracy and robustness with less training data. Among them, in MNIST dataset, compared with the original spiking neural network with the same network structure, we achieved a 0.39% accuracy improvement.

Hyperexcited limbic neurons represent sexual satiety and reduce mating motivation.

Transient sexual experiences can have long-lasting effects on behavioral decisions, but the neural coding that accounts for this change is unclear. We found that the ejaculation experience selectively activated estrogen receptor 2 (Esr2)-expressing neurons in the bed nucleus of the stria terminalis (BNST)-BNSTEsr2-and led to persistent decreases in firing threshold for days, during which time the mice displayed sexual satiety. Inhibition of hyperexcited BNSTEsr2 elicited fast mating recovery in satiated mice of both sexes. In males, such hyperexcitability reduced mating motivation and was partially mediated by larger HCN (hyperpolarization-activated cyclic nucleotide-gated) currents. Thus, BNSTEsr2 not only encode a specific mating action but also represent a persistent state of sexual satiety, and alterations in a neuronal ion channel contribute to sexual experience-dependent long-term changes to mating drive.

Mechanisms Underlying the Recruitment of Inhibitory Interneurons in Fictive Swimming in Developing Xenopus laevis Tadpoles.

Developing spinal circuits generate patterned motor outputs while many neurons with high membrane resistances are still maturing. In the spinal cord of hatchling frog tadpoles of unknown sex, we found that the firing reliability in swimming of inhibitory interneurons with commissural and ipsilateral ascending axons was negatively correlated with their cellular membrane resistance. Further analyses showed that neurons with higher resistances had outward rectifying properties, low firing thresholds, and little delay in firing evoked by current injections. Input synaptic currents these neurons received during swimming, either compound, unitary current amplitudes, or unitary synaptic current numbers, were scaled with their membrane resistances, but their own synaptic outputs were correlated with membrane resistances of their postsynaptic partners. Analyses of neuronal dendritic and axonal lengths and their activities in swimming and cellular input resistances did not reveal a clear correlation pattern. Incorporating these electrical and synaptic properties into a computer swimming model produced robust swimming rhythms, whereas randomizing input synaptic strengths led to the breakdown of swimming rhythms, coupled with less synchronized spiking in the inhibitory interneurons. We conclude that the recruitment of these developing interneurons in swimming can be predicted by cellular input resistances, but the order is opposite to the motor-strength-based recruitment scheme depicted by Henneman's size principle. This form of recruitment/integration order in development before the emergence of refined motor control is progressive potentially with neuronal acquisition of mature electrical and synaptic properties, among which the scaling of input synaptic strengths with cellular input resistance plays a critical role.SIGNIFICANCE STATEMENT The mechanisms on how interneurons are recruited to participate in circuit function in developing neuronal systems are rarely investigated. In 2-d-old frog tadpole spinal cord, we found the recruitment of inhibitory interneurons in swimming is inversely correlated with cellular input resistances, opposite to the motor-strength-based recruitment order depicted by Henneman's size principle. Further analyses showed the amplitude of synaptic inputs that neurons received during swimming was inversely correlated with cellular input resistances. Randomizing/reversing the relation between input synaptic strengths and membrane resistances in modeling broke down swimming rhythms. Therefore, the recruitment or integration of these interneurons is conditional on the acquisition of several electrical and synaptic properties including the scaling of input synaptic strengths with cellular input resistances.

Toward robust and scalable deep spiking reinforcement learning.

Deep reinforcement learning (DRL) combines reinforcement learning algorithms with deep neural networks (DNNs). Spiking neural networks (SNNs) have been shown to be a biologically plausible and energy efficient alternative to DNNs. Since the introduction of surrogate gradient approaches that allowed to overcome the discontinuity in the spike function, SNNs can now be trained with the backpropagation through time (BPTT) algorithm. While largely explored on supervised learning problems, little work has been done on investigating the use of SNNs as function approximators in DRL. Here we show how SNNs can be applied to different DRL algorithms like Deep Q-Network (DQN) and Twin-Delayed Deep Deteministic Policy Gradient (TD3) for discrete and continuous action space environments, respectively. We found that SNNs are sensitive to the additional hyperparameters introduced by spiking neuron models like current and voltage decay factors, firing thresholds, and that extensive hyperparameter tuning is inevitable. However, we show that increasing the simulation time of SNNs, as well as applying a two-neuron encoding to the input observations helps reduce the sensitivity to the membrane parameters. Furthermore, we show that randomizing the membrane parameters, instead of selecting uniform values for all neurons, has stabilizing effects on the training. We conclude that SNNs can be utilized for learning complex continuous control problems with state-of-the-art DRL algorithms. While the training complexity increases, the resulting SNNs can be directly executed on neuromorphic processors and potentially benefit from their high energy efficiency.

Scaling Limit of a Generalized Contact Process

We derive macroscopic equations for a generalized contact process that is inspired by a neuronal integrate and fire model on the lattice $$\mathbb {Z}^d$$ . The states at each lattice site can take values in $$0,\ldots ,k$$ . These can be interpreted as neuronal membrane potential, with the state k corresponding to a firing threshold. In the terminology of the contact processes, which we shall use in this paper, the state k corresponds to the individual being infectious (all other states are noninfectious). In order to reach the firing threshold, or to become infectious, the site must progress sequentially from 0 to k. The rate at which it climbs is determined by other neurons at state k, coupled to it through a Kac-type potential, of range $$\gamma ^{-1}$$ . The hydrodynamic equations are obtained in the limit $$\gamma \rightarrow 0$$ . Extensions of the microscopic model to include excitatory and inhibitory neuron types, as well as other biophysical mechanisms, are also considered.

A Compact Artificial Spiking Neuron Using a Sharp-Switching FET With Ultra-Low Energy Consumption Down to 0.45 fJ/Spike

In this article, we first utilize a zero subthreshold swing and zero impact ionization FET (Z2-FET) as an innovative artificial spiking neuron and demonstrate it with CMOS-compatible technology. Owing to the sharp-switching and hysteresis characteristics of Z2-FET, the artificial neuron successfully emulates the key biological neuronal behaviors based on a single device, including threshold-driven spiking and stimulus strength-modulated frequency response. Furthermore, the firing threshold of the neuron is conveniently tuned by the gate voltage, which is helpful in the application of spiking neural networks (SNN). A preliminary endurance up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${2}\times {10}^{{8}}$ </tex-math></inline-formula> cycles is obtained in the neuron. TCAD simulations further verify the scaling capability of the Z2-FET neuron and its energy consumption is estimated. The results suggest that <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{Z}^{\vphantom {D^{f}}{2}}$ </tex-math></inline-formula> -FET has great potential for highly compact and energy-efficient artificial spiking neurons.

Fire on a tropical floodplain: a fine-scale fire history of coastal floodplains in the Northern Territory, Australia

Background The coastal floodplains of northern Australia are fire-prone, but the impact of fire on floodplain biota is not well understood. Aims In this study, we sought to characterise the fire history of six adjacent floodplains in coastal Northern Territory, Australia. Methods We built a fine-scale 31-year fire history (1988–2018) to compare fire regimes on floodplains across land tenures and floodplain fire regimes with the surrounding savanna fire regime, determine the extent to which current fire regimes are meeting existing ecological fire thresholds, and investigate the relationship between rainfall totals and the extent of burning. Key results Floodplains in conservation reserves burnt more frequently than those on pastoral lands, and savannas burnt more often than floodplains. Current floodplain fire regimes comfortably meet existing ecological fire thresholds. The proportion of floodplain burning is inversely proportional to the amount of rain in the previous wet season. Conclusions Floodplain fire regimes vary markedly between land-use types, and floodplain fire regimes differ to those of savannas. The current management thresholds for floodplain fire regimes would benefit from further evidence of conservation outcomes. Implications For more effective floodplain fire management, research is needed to generate floodplain-specific thresholds that best conserve their considerable conservation value.

ERR2 and ERR3 promote the development of gamma motor neuron functional properties required for proprioceptive movement control

The ability of terrestrial vertebrates to effectively move on land is integrally linked to the diversification of motor neurons into types that generate muscle force (alpha motor neurons) and types that modulate muscle proprioception, a task that in mammals is chiefly mediated by gamma motor neurons. The diversification of motor neurons into alpha and gamma types and their respective contributions to movement control have been firmly established in the past 7 decades, while recent studies identified gene expression signatures linked to both motor neuron types. However, the mechanisms that promote the specification of gamma motor neurons and/or their unique properties remained unaddressed. Here, we found that upon selective loss of the orphan nuclear receptors ERR2 and ERR3 (also known as ERRβ, ERRγ or NR3B2, NR3B3, respectively) in motor neurons in mice, morphologically distinguishable gamma motor neurons are generated but do not acquire characteristic functional properties necessary for regulating muscle proprioception, thus disrupting gait and precision movements. Complementary gain-of-function experiments in chick suggest that ERR2 and ERR3 could operate via transcriptional activation of neural activity modulators to promote a gamma motor neuron biophysical signature of low firing thresholds and high firing rates. Our work identifies a mechanism specifying gamma motor neuron functional properties essential for the regulation of proprioceptive movement control.

Targeting the vascular endothelial growth factor A/neuropilin 1 axis for relief of neuropathic pain.

Vascular endothelial growth factor A (VEGF-A) is a pronociceptive factor that causes neuronal sensitization and pain. We reported that blocking the interaction between the membrane receptor neuropilin 1 (NRP1) and VEGF-A-blocked VEGF-A-mediated sensory neuron hyperexcitability and reduced mechanical hypersensitivity in a rodent chronic neuropathic pain model. These findings identified the NRP1-VEGF-A signaling axis for therapeutic targeting of chronic pain. In an in-silico screening of approximately 480 K small molecules binding to the extracellular b1b2 pocket of NRP1, we identified 9 chemical series, with 6 compounds disrupting VEGF-A binding to NRP1. The small molecule with greatest efficacy, 4'-methyl-2'-morpholino-2-(phenylamino)-[4,5'-bipyrimidin]-6(1H)-one, designated NRP1-4, was selected for further evaluation. In cultured primary sensory neurons, VEGF-A enhanced excitability and decreased firing threshold, which was blocked by NRP1-4. In addition, NaV1.7 and CaV2.2 currents and membrane expression were potentiated by treatment with VEGF-A, and this potentiation was blocked by NRP1-4 cotreatment. Neuropilin 1-4 reduced VEGF-A-mediated increases in the frequency and amplitude of spontaneous excitatory postsynaptic currents in dorsal horn of the spinal cord. Neuropilin 1-4 did not bind to more than 300 G-protein-coupled receptors and receptors including human opioids receptors, indicating a favorable safety profile. In rats with spared nerve injury-induced neuropathic pain, intrathecal administration of NRP1-4 significantly attenuated mechanical allodynia. Intravenous treatment with NRP1-4 reversed both mechanical allodynia and thermal hyperalgesia in rats with L5/L6 spinal nerve ligation-induced neuropathic pain. Collectively, our findings show that NRP1-4 is a first-in-class compound targeting the NRP1-VEGF-A signaling axis to control voltage-gated ion channel function, neuronal excitability, and synaptic activity that curb chronic pain.

Optimization of Water Network Topology and Pipe Sizing to Aid Water Utilities in Deciding on a Design Philosophy: A Real Case Study in Belgium

Numerical optimization is gradually finding its way into drinking water practice. For successful introduction of optimization into the sector, it is important that researchers and utility experts work together on the problem formulation with the water utility experts. Water utilities heed the solutions provided by optimization techniques only when the underlying approach and performance criteria match their specific goals. In this contribution, we demonstrate the application of numerical optimization on a real-life problem. The Belgian utility De Watergroep is looking to not only reinforce its distribution networks but to also structurally modify the network’s topology to enhance the quality of water delivered in the future. To help the utility explore the possibilities of these far-reaching changes in the most flexible way possible, an optimization problem was formulated to optimize topology and pipe sizing simultaneously for the distribution network of a Belgian city. The objective of the problem is to minimize the volume of the looped network and thereby work towards a situation where most of the customers are fed by branched extremities of the network. This objective is constrained by pressure and fire flow requirements and thresholds on the number of customers on the branched sections. The requirements for continuity of supply under failure scenarios are guaranteed by these constraints, as verified in the final solution. The results of the optimization process show that it is possible to design a network which is 18.5% cheaper than the currently existing network. Moreover, it turns out the—previously completely meshed—topology can be restructured so that 67% of the network length is turned into branched clusters, with a meshed superstructure of 33% of the length remaining.

Tuning Convolutional Spiking Neural Network With Biologically Plausible Reward Propagation.

Spiking neural networks (SNNs) contain more biologically realistic structures and biologically inspired learning principles than those in standard artificial neural networks (ANNs). SNNs are considered the third generation of ANNs, powerful on the robust computation with a low computational cost. The neurons in SNNs are nondifferential, containing decayed historical states and generating event-based spikes after their states reaching the firing threshold. These dynamic characteristics of SNNs make it difficult to be directly trained with the standard backpropagation (BP), which is also considered not biologically plausible. In this article, a biologically plausible reward propagation (BRP) algorithm is proposed and applied to the SNN architecture with both spiking-convolution (with both 1-D and 2-D convolutional kernels) and full-connection layers. Unlike the standard BP that propagates error signals from postsynaptic to presynaptic neurons layer by layer, the BRP propagates target labels instead of errors directly from the output layer to all prehidden layers. This effort is more consistent with the top-down reward-guiding learning in cortical columns of the neocortex. Synaptic modifications with only local gradient differences are induced with pseudo-BP that might also be replaced with the spike-timing-dependent plasticity (STDP). The performance of the proposed BRP-SNN is further verified on the spatial (including MNIST and Cifar-10) and temporal (including TIDigits and DvsGesture) tasks, where the SNN using BRP has reached a similar accuracy compared to other state-of-the-art (SOTA) BP-based SNNs and saved 50% more computational cost than ANNs. We think that the introduction of biologically plausible learning rules to the training procedure of biologically realistic SNNs will give us more hints and inspiration toward a better understanding of the biological system's intelligent nature.

A Vertical Single Transistor Neuron with Core-Shell Dual-Gate for Excitatory-Inhibitory Function and Tunable Firing Threshold Voltage.

A novel inhibitable and firing threshold voltage tunable vertical nanowire (NW) single transistor neuron device with core–shell dual-gate (CSDG) was realized and verified by TCAD simulation. The CSDG NW neuron is enclosed by an independently accessed shell gate and core gate to serve an excitatory–inhibitory transition and a firing threshold voltage adjustment, respectively. By utilizing the shell gate, the firing of specific neuron can be inhibited for winner-takes-all learning. It was confirmed that the independently accessed core gate can be used for adjustment of the firing threshold voltage to compensate random conductance variation before the learning and to fix inference error caused by unwanted synapse conductance change after the learning. This threshold voltage tuning can also be utilized for homeostatic function during the learning process. Furthermore, a myelination function which controls the transmission rate was obtained based on the inherent asymmetry between the source and drain in vertical NW structure. Finally, using the CSDG NW neuron device, a letter recognition test was conducted by SPICE simulation for a system-level validation. This multi-functional neuron device can contribute to construct a high-density monolithic SNN hardware combining with the previously developed vertical synapse MOSFET devices.

Patch-clamp analysis of nicotinic synapses whose strength straddles the firing threshold of rat sympathetic neurons.

Neurons in paravertebral sympathetic ganglia are innervated by converging nicotinic synapses of varying strength. Based upon intracellular recordings of excitatory postsynaptic potentials (EPSPs) with sharp microelectrodes these synapses were classified in the past as either primary (strong) or secondary (weak) by their ability to trigger postsynaptic action potentials. Here we present an analysis of 22 synapses whose strength straddled threshold, thereby distinguishing them from the original classification scheme for primary and secondary synapses. Recordings at 36°C were made from intact superior cervical ganglia isolated from 13 male and 3 female Sprague-Dawley rats and from 4 male spontaneously hypertensive (SHR) rats. Ganglia were pretreated with collagenase to permit patch recording. By dissecting a 1 cm length of the presynaptic cervical sympathetic nerve as part of the preparation and through use of graded presynaptic stimulation it was possible to fractionate synaptic inputs by their distinct latencies and magnitudes, and by the presynaptic stimulus threshold for each component. Comparison of cell-attached extracellular recordings with intracellular recordings of synaptic potentials and synaptic currents indicated that straddling EPSPs are not an artifact of shunting damage caused by intracellular recording. The results also showed that in cells where a single presynaptic shock elicits multiple action potentials, the response is driven by multiple synapses and not by repetitive postsynaptic firing. The conductance of straddling synapses also provides a direct estimate of the threshold synaptic conductance (9.8 nS ± 7.6 nS, n = 22, mean ± SD). The results are discussed in terms of their implications for ganglionic integration and an existing model of synaptic amplification.

Synchronous firing of dorsal horn neurons at the origin of dorsal root reflexes in naïve and paw-inflamed mice.

Spinal interneurons located in the dorsal horn induce primary afferent depolarization (PAD) controlling the excitability of the afferent’s terminals. Following inflammation, PAD may reach firing threshold contributing to maintain inflammation and pain. Our aim was to study the collective behavior of dorsal horn neurons, its relation to backfiring of primary afferents and the effects of a peripheral inflammation in this system. Experiments were performed on slices of spinal cord obtained from naïve adult mice or mice that had suffered an inflammatory pretreatment. Simultaneous recordings from groups of dorsal horn neurons and primary afferents were obtained and machine-learning methodology was used to analyze effective connectivity between them. Dorsal horn recordings showed grouping of spontaneous action potentials from different neurons in “population bursts.” These occurred at irregular intervals and were formed by action potentials from all classes of neurons recorded. Compared to naïve, population bursts from treated animals concentrated more action potentials, had a faster onset and a slower decay. Population bursts were disrupted by perfusion of picrotoxin and held a strong temporal correlation with backfiring of afferents. Effective connectivity analysis allowed pinpointing specific neurons holding pre- or post-synaptic relation to the afferents. Many of these neurons had an irregular fast bursting pattern of spontaneous firing. We conclude that population bursts contain action potentials from neurons presynaptic to the afferents which are likely to control their excitability. Peripheral inflammation may enhance synchrony in these neurons, increasing the chance of triggering action potentials in primary afferents and contributing toward central sensitization.

Nanoplasma-Based Millimeter-Wave Modulators on a Single Metal Layer

Fundamental constraints imposing power-frequency trade-offs in conventional electronics have stimulated research on alternative technologies for millimeter-wave and sub-millimeter-wave applications. In this work, we use the picosecond threshold firing of nanoplasma switches to demonstrate on-chip millimeter-wave modulators that rely only on a single metal layer. We show amplitude shift keying (ASK) modulation with self-synthesized carrier frequencies up to 66 GHz (limited by the bandwidth of our experimental setup), with output powers up to ~30 dBm. These all-metal nanoplasma modulators are low cost, and generally compatible with different platforms, from CMOS and III-V compounds to flexible substrates. Our work paves the way towards future terahertz communication circuits with large output powers, which otherwise are not practical using high-power amplifiers at frequencies over 100 GHz. In a more general context, the proposed all-metal circuits can potentially synthesize arbitrarily-shaped ultra-wide-band (UWB) signals with applications in advanced wireless communications, radars, and imaging systems.

Synaptic NMDA receptor activity at resting membrane potentials.

NMDA receptors (NMDARs) are crucial for glutamatergic synaptic signaling in the mammalian central nervous system. When activated by glutamate and glycine/D-serine, the NMDAR ion channel can open, but current flux is further regulated by voltage-dependent block conferred by extracellular Mg2+ ions. The unique biophysical property of ligand- and voltage-dependence positions NMDARs as synaptic coincidence detectors, controlling a major source of synaptic Ca2+ influx. We measured synaptic currents in layer 2/3 neurons after stimulation in layer 4 of somatosensory cortex and found measurable NMDAR currents at all voltages tested. This NMDAR current did not require concurrent AMPAR depolarization. In physiological ionic conditions, the NMDAR current response at negative potentials was enhanced relative to ionic conditions typically used in slice experiments. NMDAR activity was also seen in synaptic recordings from hippocampal CA1 neurons, indicating a general property of NMDAR signaling. Using a fluorescent Ca2+ indicator, we measured responses to stimulation in layer 4 at individual synaptic sites, and Ca2+ influx could be detected even with AMPARs blocked. In current clamp recordings, we found that resting membrane potential was hyperpolarized by ∼7 mV and AP firing threshold depolarized by ∼4 mV in traditional compared to physiological ionic concentrations, and that NMDARs contribute to EPSPs at resting membrane potentials. These measurements demonstrate that, even in the presence of extracellular Mg2+ and absence of postsynaptic depolarization, NMDARs contribute to synaptic currents and Ca2+ influx.

Teasing Out the Interplay Between Natural Killer Cells and Nociceptor Neurons.

Somatosensory neurons have evolved to detect noxious stimuli and activate defensive reflexes. By sharing means of communication, nociceptor neurons also tune host defenses by controlling the activity of the immune system. The communication between these systems is mostly adaptive, helping to protect homeostasis, it can also lead to, or promote, the onset of chronic diseases. Both systems co-evolved to allow for such local interaction, as found in primary and secondary lymphoid tissues and mucosa. Recent studies have demonstrated that nociceptors directly detect and respond to foreign antigens, immune cell-derived cytokines, and microbes. Nociceptor activation not only results in pain hypersensitivity and itching, but lowers the nociceptor firing threshold, leading to the local release of neuropeptides. The peptides that are produced by, and released from, the peripheral terminals of nociceptors can block the chemotaxis and polarization of lymphocytes, controlling the localization, duration, and type of inflammation. Recent evidence shows that sensory neurons interact with innate immune cells via cell-cell contact, for example, engaging group 2D (NKG2D) receptors on natural killer (NK) cells. Given that NK cells express the cognate receptors for various nociceptor-produced mediators, it is conceivable that nociceptors use neuropeptides to control the activity of NK cells. Here, we devise a co-culture method to study nociceptor neuron-NK cell interactions in a dish. Using this approach, we found that lumbar nociceptor neurons decrease NK cell cytokine expression. Overall, such a reductionist method could be useful to study how tumor-innervating neurons control the anticancer function of NK cells and how NK cells control the elimination of injured neurons.

Non-monotonic effects of GABAergic synaptic inputs on neuronal firing.

GABA is generally known as the principal inhibitory neurotransmitter in the nervous system, usually acting by hyperpolarizing membrane potential. However, GABAergic currents sometimes exhibit non-inhibitory effects, depending on the brain region, developmental stage or pathological condition. Here, we investigate the diverse effects of GABA on the firing rate of several single neuron models, using both analytical calculations and numerical simulations. We find that GABAergic synaptic conductance and output firing rate exhibit three qualitatively different regimes as a function of GABA reversal potential, EGABA: monotonically decreasing for sufficiently low EGABA (inhibitory), monotonically increasing for EGABA above firing threshold (excitatory); and a non-monotonic region for intermediate values of EGABA. In the non-monotonic regime, small GABA conductances have an excitatory effect while large GABA conductances show an inhibitory effect. We provide a phase diagram of different GABAergic effects as a function of GABA reversal potential and glutamate conductance. We find that noisy inputs increase the range of EGABA for which the non-monotonic effect can be observed. We also construct a micro-circuit model of striatum to explain observed effects of GABAergic fast spiking interneurons on spiny projection neurons, including non-monotonicity, as well as the heterogeneity of the effects. Our work provides a mechanistic explanation of paradoxical effects of GABAergic synaptic inputs, with implications for understanding the effects of GABA in neural computation and development.

Effective Conversion of a Convolutional Neural Network into a Spiking Neural Network for Image Recognition Tasks

Due to energy efficiency, spiking neural networks (SNNs) have gradually been considered as an alternative to convolutional neural networks (CNNs) in various machine learning tasks. In image recognition tasks, leveraging the superior capability of CNNs, the CNN–SNN conversion is considered one of the most successful approaches to training SNNs. However, previous works assume a rather long inference time period called inference latency to be allowed, while having a trade-off between inference latency and accuracy. One of the main reasons for this phenomenon stems from the difficulty in determining proper a firing threshold for spiking neurons. The threshold determination procedure is called a threshold balancing technique in the CNN–SNN conversion approach. This paper proposes a CNN–SNN conversion method with a new threshold balancing technique that obtains converted SNN models with good accuracy even with low latency. The proposed method organizes the SNN models with soft-reset IF spiking neurons. The threshold balancing technique estimates the thresholds for spiking neurons based on the maximum input current in a layerwise and channelwise manner. The experiment results have shown that our converted SNN models attain even higher accuracy than the corresponding trained CNN model for the MNIST dataset with low latency. In addition, for the Fashion-MNIST and CIFAR-10 datasets, our converted SNNs have shown less conversion loss than other methods in low latencies. The proposed method can be beneficial in deploying efficient SNN models for recognition tasks on resource-limited systems because the inference latency is strongly associated with energy consumption.