Spike Threshold Research Articles

Effective multispike learning in a spiking neural network with a new temporal feedback backpropagation for breast cancer detection

This paper presents an effective learning multi-spike deep spiking neural network with temporal feedback backpropagation for breast cancer detection using contrast-enhanced MRI images. The learning in the spiking network is a new universal temporal feedback called Temporal_Feedback_SpikeProp (TF_SpikeProp) and Temporal_Feedback_ReSuMe (TF_ReSuMe). Thus, it can be implemented on all kinds of algorithms such as MultiSpikeProp and MultiReSuMe algorithms and is compatible with all temporal codings. The presented spiking network is a functional network with high accuracy and convergence in low epochs. The new algorithm explores the influence of all presynaptic neurons on the output error and reduces the role of inactive neurons. Therefore, the error propagation of the spiking network is such that it affects not only the spike time of each neuron in the output spike time, but also the refractory time and the presynaptic spikes of the neuron during the spiking time of the output neuron. The spiking network can correct and adjust the weights, spike delay, spike threshold, and also the time constant of the spike collection kernel. In order to diagnose cancer tissue, time–frequency features such as STFT and packet wavelet transform (WPT) have been used along with texture recognition features such as co-occurrence matrix. The presented network achieved an accuracy of 98.3% in 23 epochs in SpikeProp algorithm, 97.4% in ReSuMe with Duke MRI dataset, 98.64% on MNIST, and 96.1% on Iris. The results have shown that the training algorithm used in this study can achieve high accuracy in low epochs compared to traditional algorithms, while solving the challenge of getting stuck in local minima and poor convergence, as well as reducing the problem of time confusion of output spikes to an acceptable level and providing a fully functional network in practice and reality.

Approximating Nonlinear Functions With Latent Boundaries in Low-Rank Excitatory-Inhibitory Spiking Networks.

Deep feedforward and recurrent neural networks have become successful functional models of the brain, but they neglect obvious biological details such as spikes and Dale's law. Here we argue that these details are crucial in order to understand how real neural circuits operate. Towards this aim, we put forth a new framework for spike-based computation in low-rank excitatory-inhibitory spiking networks. By considering populations with rank-1 connectivity, we cast each neuron's spiking threshold as a boundary in a low-dimensional input-output space. We then show how the combined thresholds of a population of inhibitory neurons form a stable boundary in this space, and those of a population of excitatory neurons form an unstable boundary. Combining the two boundaries results in a rank-2 excitatory-inhibitory (EI) network with inhibition-stabilized dynamics at the intersection of the two boundaries. The computation of the resulting networks can be understood as the difference of two convex functions and is thereby capable of approximating arbitrary non-linear input-output mappings. We demonstrate several properties of these networks, including noise suppression and amplification, irregular activity and synaptic balance, as well as how they relate to rate network dynamics in the limit that the boundary becomes soft. Finally, while our work focuses on small networks (5-50 neurons), we discuss potential avenues for scaling up to much larger networks. Overall, our work proposes a new perspective on spiking networks that may serve as a starting point for a mechanistic understanding of biological spike-based computation.

Leaky-integrate-fire neuron based on vertically extended drain Si[formula omitted]Ge[formula omitted] source TFET: Ultra-energy-efficient and high speed design

This article investigates the performance of a vertically extended drain double gate Si1−x Gex source tunnel field effect transistor (VD-DG-Si1−x Gex S-TFET) as a leaky integrate-and-fire (LIF) neuron. The proposed VD-DG-Si1−x Gex S-TFET structure has been designed and optimized by deploying the commercially available Silvaco TCAD. The LIF neuron using VD-DG-Si1−x Gex S-TFET requires a low spiking threshold voltage (i.e. 0.20 V) which is 5x, 3x, 5x, 1.3x, 1.55x, 7.5x, and 1.44x times less in comparison to MOSFET, Bulk FinFET, phase change device, SOI MOSFET, DG-JLFET, silicon nanowire, and DL-TFET silicon neuron. The proposed device based neuron demonstrates the least spiking energy (177 ZJ) reported to date which is 18192 times less as compared to BTBT based PDSOI MOSFET neuron. The threshold value of current is achieved at 0.20 V drain voltage which is 15x, 14x, 10x, and 1.5x less as compared to FinFET, PD-SOI MOSFET, LBIMOS, and DL-TFET, respectively. Moreover, the effects of gate metal work function, germanium mole fraction, and temperature on the spike current variations have also been investigated. Here, impact ionization is a key method for generating a neuron spike due to holes accumulated in the potential well of VD-DG-Si1−xGex S-TFET. In addition, a LIF neuron based on VD-DG-Si0.5Ge0.5 S-TFET device exhibits a 9 THz spiking frequency which is approximately 12 times higher than real nerves present in the human body. This work shows a remarkable reduction in the power consumption and the proposed device neuron is a potential start-up solution for a large scale spiking neural networks (SNN) implementation because of its better performance and compact circuitry.

A Synapse-Threshold Synergistic Learning Approach for Spiking Neural Networks

Spiking neural networks (SNNs) have demonstrated excellent capabilities in various intelligent scenarios. Most existing methods for training SNNs are based on the concept of synaptic plasticity; however, learning in the realistic brain also utilizes intrinsic non-synaptic mechanisms of neurons. The spike threshold of biological neurons is a critical intrinsic neuronal feature that exhibits rich dynamics on a millisecond timescale and has been proposed as an underlying mechanism that facilitates neural information processing. In this study, we develop a novel synergistic learning approach that involves simultaneously training synaptic weights and spike thresholds in SNNs. SNNs trained with synapse-threshold synergistic learning (STL-SNNs) achieve significantly superior performance on various static and neuromorphic datasets than SNNs trained with two degenerated single-learning models. During training, the synergistic learning approach optimizes neural thresholds, providing the network with stable signal transmission via appropriate firing rates. Further analysis indicates that STL-SNNs are robust to noisy data and exhibit low energy consumption for deep network structures. Additionally, the performance of STL-SNN can be further improved by introducing a generalized joint decision framework. Overall, our findings indicate that biologically plausible synergies between synaptic and intrinsic non-synaptic mechanisms may provide a promising approach for developing highly efficient SNN learning methods.

A brainstem–hypothalamus neuronal circuit reduces feeding upon heat exposure

Empirical evidence suggests that heat exposure reduces food intake. However, the neurocircuit architecture and the signalling mechanisms that form an associative interface between sensory and metabolic modalities remain unknown, despite primary thermoceptive neurons in the pontine parabrachial nucleus becoming well characterized1. Tanycytes are a specialized cell type along the wall of the third ventricle2 that bidirectionally transport hormones and signalling molecules between the brain’s parenchyma and ventricular system3–8. Here we show that tanycytes are activated upon acute thermal challenge and are necessary to reduce food intake afterwards. Virus-mediated gene manipulation and circuit mapping showed that thermosensing glutamatergic neurons of the parabrachial nucleus innervate tanycytes either directly or through second-order hypothalamic neurons. Heat-dependent Fos expression in tanycytes suggested their ability to produce signalling molecules, including vascular endothelial growth factor A (VEGFA). Instead of discharging VEGFA into the cerebrospinal fluid for a systemic effect, VEGFA was released along the parenchymal processes of tanycytes in the arcuate nucleus. VEGFA then increased the spike threshold of Flt1-expressing dopamine and agouti-related peptide (Agrp)-containing neurons, thus priming net anorexigenic output. Indeed, both acute heat and the chemogenetic activation of glutamatergic parabrachial neurons at thermoneutrality reduced food intake for hours, in a manner that is sensitive to both Vegfa loss-of-function and blockage of vesicle-associated membrane protein 2 (VAMP2)-dependent exocytosis from tanycytes. Overall, we define a multimodal neurocircuit in which tanycytes link parabrachial sensory relay to the long-term enforcement of a metabolic code.

A biologically inspired repair mechanism for neuronal reconstructions with a focus on human dendrites.

Investigating and modelling the functionality of human neurons remains challenging due to the technical limitations, resulting in scarce and incomplete 3D anatomical reconstructions. Here we used a morphological modelling approach based on optimal wiring to repair the parts of a dendritic morphology that were lost due to incomplete tissue samples. In Drosophila, where dendritic regrowth has been studied experimentally using laser ablation, we found that modelling the regrowth reproduced a bimodal distribution between regeneration of cut branches and invasion by neighbouring branches. Interestingly, our repair model followed growth rules similar to those for the generation of a new dendritic tree. To generalise the repair algorithm from Drosophila to mammalian neurons, we artificially sectioned reconstructed dendrites from mouse and human hippocampal pyramidal cell morphologies, and showed that the regrown dendrites were morphologically similar to the original ones. Furthermore, we were able to restore their electrophysiological functionality, as evidenced by the recovery of their firing behaviour. Importantly, we show that such repairs also apply to other neuron types including hippocampal granule cells and cerebellar Purkinje cells. We then extrapolated the repair to incomplete human CA1 pyramidal neurons, where the anatomical boundaries of the particular brain areas innervated by the neurons in question were known. Interestingly, the repair of incomplete human dendrites helped to simulate the recently observed increased synaptic thresholds for dendritic NMDA spikes in human versus mouse dendrites. To make the repair tool available to the neuroscience community, we have developed an intuitive and simple graphical user interface (GUI), which is available in the TREES toolbox (www.treestoolbox.org).

Excitatory action of low frequency depolarizing GABA/glycine synaptic inputs is prevalent in prenatal spinal SOD1G93A motoneurons.

Amyotrophic lateral sclerosis (ALS) is a fatal adult-onsetneurodegenerative disease characterized by progressive motor neuron degeneration and muscle paralysis. Recent evidence suggests the dysfunction of inhibitory signalling in ALS motor neurons. We have shown that embryonic day (E)17.5spinal motoneurons (MNs) of the SOD1G93A mouse model of ALS exhibit an altered chloride homeostasis. At this prenatal stage, inhibition of spinal motoneurons (MNs) is mediated by depolarizing GABAergic/glycinergic postsynaptic potentials (dGPSPs). Here, using an ex vivo preparation and patch clamp recording from MNs with a chloride equilibrium set below spike threshold, we report that low input resistance (Rin ) E17.5 MNs from the SOD1G93A ALS mouse model do not correctly integrate dGPSPs evoked by electrical stimulations of GABA/glycine inputs at different frequencies. Indeed, firing activity of most wild-type (WT) MNs with low Rin was inhibited by incoming dGPSPs, whereas low Rin SOD1G93A MNs were excited or exhibited a dual response (excited by low frequency dGPSPs and inhibited by high frequency dGPSPs). Simulation highlighted the importance of the GABA/glycine input density and showed that pure excitation could be obtained in SOD-like MNs by moving GABA/glycine input away from the cell body to dendrites. This was in agreement with confocal imaging showing a lack of peri-somatic inhibitory terminals in SOD1G93A MNs compared to WT littermates. Putative fast ALS-vulnerable MNs with low Rin are therefore lacking functional inhibition at the near-term prenatal stage. KEY POINTS: We analysed the integration of GABAergic/glycinergic synaptic events by embryonic spinal motoneurons (MNs) in a mouse model of the amyotrophic lateral sclerosis (ALS) neurodegenerative disease. We found that GABAergic/glycinergic synaptic events do not properly inhibit ALS MNs with low input resistance, most probably corresponding to future vulnerable MNs. We used a neuron model to highlight the importance of the GABA/glycine terminal location and density in the integration of the GABAergic/glycinergic synaptic events. Confocal imaging showed a lack of GABA/glycine terminals on the cell body of ALS MNs. The present study suggests that putative ALS vulnerable MNs with low Rin lack functional inhibition at the near-term stage.

GABAergic regulation of striatal spiny projection neurons depends upon their activity state.

Synaptic transmission mediated by GABAA receptors (GABAARs) in adult, principal striatal spiny projection neurons (SPNs) can suppress ongoing spiking, but its effect on synaptic integration at subthreshold membrane potentials is less well characterized, particularly those near the resting down-state. To fill this gap, a combination of molecular, optogenetic, optical, and electrophysiological approaches were used to study SPNs in mouse ex vivo brain slices, and computational tools were used to model somatodendritic synaptic integration. In perforated patch recordings, activation of GABAARs, either by uncaging of GABA or by optogenetic stimulation of GABAergic synapses, evoked currents with a reversal potential near -60 mV in both juvenile and adult SPNs. Transcriptomic analysis and pharmacological work suggested that this relatively positive GABAAR reversal potential was not attributable to NKCC1 expression, but rather to HCO3- permeability. Regardless, from down-state potentials, optogenetic activation of dendritic GABAergic synapses depolarized SPNs. This GABAAR-mediated depolarization summed with trailing ionotropic glutamate receptor (iGluR) stimulation, promoting dendritic spikes and increasing somatic depolarization. Simulations revealed that a diffuse dendritic GABAergic input to SPNs effectively enhanced the response to dendritic iGluR signaling and promoted dendritic spikes. Taken together, our results demonstrate that GABAARs can work in concert with iGluRs to excite adult SPNs when they are in the resting down-state, suggesting that their inhibitory role is limited to brief periods near spike threshold. This state-dependence calls for a reformulation for the role of intrastriatal GABAergic circuits.

Neural heterogeneity controls computations in spiking neural networks.

The brain is composed of complex networks of interacting neurons that express considerable heterogeneity in their physiology and spiking characteristics. How does this neural heterogeneity influence macroscopic neural dynamics, and how might it contribute to neural computation? In this work, we use a mean-field model to investigate computation in heterogeneous neural networks, by studying how the heterogeneity of cell spiking thresholds affects three key computational functions of a neural population: the gating, encoding, and decoding of neural signals. Our results suggest that heterogeneity serves different computational functions in different cell types. In inhibitory interneurons, varying the degree of spike threshold heterogeneity allows them to gate the propagation of neural signals in a reciprocally coupled excitatory population. Whereas homogeneous interneurons impose synchronized dynamics that narrow the dynamic repertoire of the excitatory neurons, heterogeneous interneurons act as an inhibitory offset while preserving excitatory neuron function. Spike threshold heterogeneity also controls the entrainment properties of neural networks to periodic input, thus affecting the temporal gating of synaptic inputs. Among excitatory neurons, heterogeneity increases the dimensionality of neural dynamics, improving the network's capacity to perform decoding tasks. Conversely, homogeneous networks suffer in their capacity for function generation, but excel at encoding signals via multistable dynamic regimes. Drawing from these findings, we propose intra-cell-type heterogeneity as a mechanism for sculpting the computational properties of local circuits of excitatory and inhibitory spiking neurons, permitting the same canonical microcircuit to be tuned for diverse computational tasks.

Ethanol’s interaction with BK channel α subunit residue K361 does not mediate behavioral responses to alcohol in mice

Large conductance potassium (BK) channels are among the most sensitive molecular targets of ethanol and genetic variations in the channel-forming α subunit have been nominally associated with alcohol use disorders. However, whether the action of ethanol at BK α influences the motivation to drink alcohol remains to be determined. To address this question, we first tested the effect of systemically administered BK channel modulators on voluntary alcohol consumption in C57BL/6J males. Penitrem A (blocker) exerted dose-dependent effects on moderate alcohol intake, while paxilline (blocker) and BMS-204352 (opener) were ineffective. Because pharmacological manipulations are inherently limited by non-specific effects, we then sought to investigate the behavioral relevance of ethanol’s direct interaction with BK α by introducing in the mouse genome a point mutation known to render BK channels insensitive to ethanol while preserving their physiological function. The BK α K361N substitution prevented ethanol from reducing spike threshold in medial habenula neurons. However, it did not alter acute responses to ethanol in vivo, including ataxia, sedation, hypothermia, analgesia, and conditioned place preference. Furthermore, the mutation did not have reproducible effects on alcohol consumption in limited, continuous, or intermittent access home cage two-bottle choice paradigms conducted in both males and females. Notably, in contrast to previous observations made in mice missing BK channel auxiliary β subunits, the BK α K361N substitution had no significant impact on ethanol intake escalation induced by chronic intermittent alcohol vapor inhalation. It also did not affect the metabolic and locomotor consequences of chronic alcohol exposure. Altogether, these data suggest that the direct interaction of ethanol with BK α does not mediate the alcohol-related phenotypes examined here in mice.

M-type potassium currents differentially affect activation of motoneuron subtypes and tune recruitment gain.

The size principle is a key mechanism governing the orderly recruitment of motor units and is believed to be dependent on passive properties of the constituent motoneurons. However, motoneurons are endowed with voltage-sensitive ion channels that create non-linearities in their input-output functions. Here we describe a role for the M-type potassium current, conducted by KCNQ channels, in the control of motoneuron recruitment in mice. Motoneurons were studied with whole-cell patch clamp electrophysiology in transverse spinal slices and identified based on delayed (fast) and immediate (slow) onsets of repetitive firing. M-currents were larger in delayed compared to immediate firing motoneurons, which was not reflected by variations in the presence of Kv7.2 or Kv7.3 subunits. Instead, a more depolarized spike threshold in delayed-firing motoneurons afforded a greater proportion of the total M-current to become activated within the subthreshold voltage range, which translated to a greater influence on their recruitment with little influence on their firing rates. Pharmacological activation of M-currents also influenced motoneuron recruitment at the population level, producing a rightward shift in the recruitment curve of monosynaptic reflexes within isolated mouse spinal cords. These results demonstrate a prominent role for M-type potassium currents in regulating the function of motor units, which occurs primarily through the differential control of motoneuron subtype recruitment. More generally, these findings highlight the importance of active properties mediated by voltage-sensitive ion channels in the differential control of motoneuron recruitment, which is a key mechanism for the gradation of muscle force. KEY POINTS: M-currents exert an inhibitory influence on spinal motor output. This inhibitory influence is exerted by controlling the recruitment, but not the firing rate, of high-threshold fast-like motoneurons, with limited influence on low-threshold slow-like motoneurons. Preferential control of fast motoneurons may be linked to a larger M-current that is activated within the subthreshold voltage range compared to slow motoneurons. Larger M-currents in fast compared to slow motoneurons are not accounted for by differences in Kv7.2 or Kv7.3 channel composition. The orderly recruitment of motoneuron subtypes is shaped by differences in the contribution of voltage-gated ion channels, including KCNQ channels. KCNQ channels may provide a target to dynamically modulate the recruitment gain across the motor pool and readily adjust movement vigour.

Effect of applied electric fields on supralinear dendritic integration of interneuron

Evidences show that electric fields (EFs) induced by the magnetic stimulation could modulates brain activities by regulating the excitability of GABAergic interneuron. However, it is still unclear how and why the EF-induced polarization affects the interneuron response as the interneuron receives NMDA synaptic inputs. Considering the key role of NMDA receptor-mediated supralinear dendritic integration in neuronal computations, we suppose that the applied EFs could functionally modulate interneurons’ response via regulating dendritic integration. At first, we build a simplified multi-dendritic circuit model with inhomogeneous extracellular potentials, which characterizes the relationship among EF-induced spatial polarizations, dendritic integration, and somatic output. By performing model-based singular perturbation analysis, it is found that the equilibrium point of fast subsystem can be used to asymptotically depict the subthreshold input–output (sI/O) relationship of dendritic integration. It predicted that EF-induced strong depolarizations on the distal dendrites reduce the dendritic saturation output by reducing driving force of synaptic input, and it shifts the steep change of sI/O curve left by reducing stimulation threshold of triggering NMDA spike. Also, the EF modulation prefers the global dendritic integration with asymmetric scatter distribution of NMDA synapses. Furthermore, we identify the respective contribution of EF-regulated dendritic integration and EF-induced somatic polarization to an action potential generation and find that they have an antagonistic effect on AP generation due to the varied NMDA spike threshold under EF stimulation.

Isoliquiritigenin modulates the activity of LTS and non-LTS cells in the ventrolateral preoptic area via GABAA receptors

ObjectiveIsoliquiritigenin (ILTG) is a chalcone compound that exhibits hypnotic effects via gamma-aminobutyric acid type A (GABAA) receptors. The ventrolateral preoptic area (VLPO) is a sleep-promoting center that contains a large number of GABA-releasing cells. There are two cell types in the VLPO: one generates a low-threshold spike (LTS), whereas the other lacks an LTS (non-LTS). MethodWhole-cell patch-clamp technology was used to detect the firing and currents of LTS and non-LTS cells in the VLPO. ResultsBath administration of ILTG (10 μM) increased the firing rate of VLPO LTS cells, reversed by flumazenil (5 μM), a GABAA benzodiazepine site antagonist. However, the firing rate of VLPO non-LTS cells was inhibited by ILTG (10 μM), also reversed by flumazenil (5 μM). No differences were detected regarding resting membrane potential (RMP) amplitude, spike threshold, afterhyperpolarization (AHP) amplitude, or action potential duration (APD50) after ILTG (10 μM) perfusion in VLPO LTS cells. RMP amplitude was more hyperpolarized and spike threshold was higher after ILTG (10 μM) application in VLPO non-LTS cells. In addition, ILTG significantly reduced the frequency of miniature inhibitory postsynaptic currents (mIPSCs) in VLPO LTS cells. ILTG significantly increased the amplitude of mIPSCs in VLPO non-LTS cells. ConclusionsThis study revealed that ILTG suppresses presynaptic GABA release on VLPO LTS cells, thereby increasing their excitability. ILTG enhances postsynaptic GABAA receptor function on VLPO non-LTS cells, thereby decreasing their excitability. These results suggest that ILTG may produce hypnotic effects by modulating the GABAergic synaptic transmission properties of these two cell types.

Adaptive spike threshold dynamics associated with sparse spiking of hilar mossy cells are captured by a simple model.

Hilar mossy cells (hMCs) in the dentate gyrus (DG) receive inputs from DG granule cells (GCs), CA3 pyramidal cells and inhibitory interneurons, and provide feedback input to GCs. Behavioural and in vivo recording experiments implicate hMCs in pattern separation, navigation and spatial learning. Our experiments link hMC intrinsic excitability to their synaptically evoked in vivo spiking outputs. We performed electrophysiological recordings from DG neurons and found that hMCs displayed an adaptative spike threshold that increased both in proportion to the intensity of injected currents, and in response to spiking itself, returning to baseline over a long time scale, thereby instantaneously limiting their firing rate responses. The hMC activity is additionally limited by a prominent medium after-hyperpolarizing potential (AHP) generated by small conductance K+ channels. We hypothesize that these intrinsic hMC properties are responsible for their low in vivo firing rates. Our findings extend previous studies that compare hMCs, CA3 pyramidal cells and hilar inhibitory cells and provide novel quantitative data that contrast the intrinsic properties of these cell types. We developed a phenomenological exponential integrate-and-fire model that closely reproduces the hMC adaptive threshold nonlinearities with respect to their threshold dependence on input current intensity, evoked spike latency and long-lasting spike-induced increase in spike threshold. Our robust and computationally efficient model is amenable to incorporation into large network models of the DG that will deepen our understanding of the neural bases of pattern separation, spatial navigation and learning. KEY POINTS: Previous studies have shown that hilar mossy cells (hMCs) are implicated in pattern separation and the formation of spatial memory, but how their intrinsic properties relate to their in vivo spiking patterns is still unknown. Here we show that the hMCs display electrophysiological properties that distinguish them from the other hilar cell types including a highly adaptive spike threshold that decays slowly. The spike-dependent increase in threshold combined with an after-hyperpolarizing potential mediated by a slow K+ conductance is hypothesized to be responsible for the low-firing rate of the hMC observed in vivo. The hMC's features are well captured by a modified stochastic exponential integrate-and-fire model that has the unique feature of a threshold intrinsically dependant on both the stimulus intensity and the spiking history. This computational model will allow future work to study how the hMCs can contribute to spatial memory formation and navigation.

High calcium concentrations reduce cellular excitability of mouse MNTB neurons

Calcium, a universal intracellular signaling molecule, plays essential roles in neural functions. Historically, in most in vitro brain slice electrophysiology studies, the extracellular calcium concentration ([Ca2+]e) in artificial cerebrospinal fluid is of a wide range and typically higher than the physiological value. At high [Ca2+]e, synaptic transmission is generally enhanced. However, the effects and the underlying mechanisms of calcium on intrinsic neuronal properties are diverse. Using whole-cell patch clamp in acute brainstem slices obtained from mice of either sex, we investigated the effects and the underlying mechanisms of high [Ca2+]e on intrinsic neuronal properties of neurons in the medial nucleus of the trapezoid body (MNTB), an auditory brainstem component in the sound localization circuitry. Compared to the physiological [Ca2+]e (1.2 mM), high [Ca2+]e at 1.8 and 2.4 mM significantly reduced the cellular excitability of MNTB neurons, resulting in decreased spike firing rate, depolarized spike threshold, and decreased the ability to follow high frequency inputs. High extracellular magnesium concentrations at 1.8 and 2.4 mM produced similar but less robust effects, due to surface charge screening. Upon high calcium application, voltage-gated sodium channel currents remained largely unchanged. Calcium-sensing receptors were detected in MNTB neurons, but blocking these receptors did not eliminate the effects of high calcium on spontaneous spiking. We attribute the lack of significant effects in these last two experiments to the moderate changes in calcium we tested. Our results call for the use of physiological [Ca2+]e in brain slice experiments.

Unsupervised Spiking Neural Network with Dynamic Learning of Inhibitory Neurons.

A spiking neural network (SNN) is a type of artificial neural network that operates based on discrete spikes to process timing information, similar to the manner in which the human brain processes real-world problems. In this paper, we propose a new spiking neural network (SNN) based on conventional, biologically plausible paradigms, such as the leaky integrate-and-fire model, spike timing-dependent plasticity, and the adaptive spiking threshold, by suggesting new biological models; that is, dynamic inhibition weight change, a synaptic wiring method, and Bayesian inference. The proposed network is designed for image recognition tasks, which are frequently used to evaluate the performance of conventional deep neural networks. To manifest the bio-realistic neural architecture, the learning is unsupervised, and the inhibition weight is dynamically changed; this, in turn, affects the synaptic wiring method based on Hebbian learning and the neuronal population. In the inference phase, Bayesian inference successfully classifies the input digits by counting the spikes from the responding neurons. The experimental results demonstrate that the proposed biological model ensures a performance improvement compared with other biologically plausible SNN models.

Mutual information measure of visual perception based on noisy spiking neural networks.

Note that images of low-illumination are weak aperiodic signals, while mutual information can be used as an effective measure for the shared information between the input stimulus and the output response of nonlinear systems, thus it is possible to develop novel visual perception algorithm based on the principle of aperiodic stochastic resonance within the frame of information theory. To confirm this, we reveal this phenomenon using the integrate-and-fire neural networks of neurons with noisy binary random signal as input first. And then, we propose an improved visual perception algorithm with the image mutual information as assessment index. The numerical experiences show that the target image can be picked up with more easiness by the maximal mutual information than by the minimum of natural image quality evaluation (NIQE), which is one of the most frequently used indexes. Moreover, the advantage of choosing quantile as spike threshold has also been confirmed. The improvement of this research should provide large convenience for potential applications including video tracking in environments of low illumination.

Photonic integrated spiking neuron chip based on a self-pulsating DFB laser with a saturable absorber

We proposed and experimentally demonstrated a simple and novel photonic spiking neuron based on a distributed feedback (DFB) laser chip with an intracavity saturable absorber (SA). The DFB laser with an intracavity SA (DFB-SA) contains a gain region and an SA region. The gain region is designed and fabricated by the asymmetric equivalent π-phase shift based on the reconstruction-equivalent-chirp technique. Under properly injected current in the gain region and reversely biased voltage in the SA region, periodic self-pulsation was experimentally observed due to the Q-switching effect. The self-pulsation frequency increases with the increase of the bias current and is within the range of several gigahertz. When the bias current is below the self-pulsation threshold, neuronlike spiking responses appear when external optical stimulus pulses are injected. Experimental results show that the spike threshold, temporal integration, and refractory period can all be observed in the fabricated DFB-SA chip. To numerically verify the experimental findings, a time-dependent coupled-wave equation model was developed, which described the physics processes inside the gain and SA regions. The numerical results agree well with the experimental measurements. We further experimentally demonstrated that the weighted sum output can readily be encoded into the self-pulsation frequency of the DFB-SA neuron. We also benchmarked the handwritten digit classification task with a simple single-layer fully connected neural network. By using the experimentally measured dependence of the self-pulsation frequency on the bias current in the gain region as an activation function, we can achieve a recognition accuracy of 92.2%, which bridges the gap between the continuous valued artificial neural networks and spike-based neuromorphic networks. To the best of our knowledge, this is the first experimental demonstration of a photonic integrated spiking neuron based on a DFB-SA, which shows great potential to realizing large-scale multiwavelength photonic spiking neural network chips.

A Hybrid Model for Multi-Day-Ahead Electricity Price Forecasting considering Price Spikes

This paper proposes a new hybrid model to forecast electricity market prices up to four days ahead. The components of the proposed model are combined in two dimensions. First, on the “vertical” dimension, long short-term memory (LSTM) neural networks and extreme gradient boosting (XGBoost) models are stacked up to produce supplementary price forecasts. The final forecasts are then picked depending on how the predictions compare to a price spike threshold. On the “horizontal” dimension, five models are designed to extend the forecasting horizon to four days. This is an important requirement to make forecasts useful for market participants who trade energy and ancillary services multiple days ahead. The horizontally cascaded models take advantage of the availability of specific public data for each forecasting horizon. To enhance the forecasting capability of the model in dealing with price spikes, we deploy a previously unexplored input in the proposed methodology. That is, to use the recent variations in the output power of thermal units as an indicator of unplanned outages or shift in the supply stack. The proposed method is tested using data from Alberta’s electricity market, which is known for its volatility and price spikes. An economic application of the developed forecasting model is also carried out to demonstrate how several market players in the Alberta electricity market can benefit from the proposed multi-day ahead price forecasting model. The numerical results demonstrate that the proposed methodology is effective in enhancing forecasting accuracy and price spike detection.

Synaptic variance and action potential firing of cerebellar output neurons during motor learning in larval zebrafish

The cerebellum regulates both reflexive and acquired movements. Here, by recording voltage-clamped synaptic currents and spiking in cerebellar output (eurydendroid) neurons in immobilized larval zebrafish, we investigated synaptic integration during reflexive movements and throughout associative motor learning. Spiking coincides with the onset of reflexive fictive swimming but precedes learned swimming, suggesting that eurydendroid signals may facilitate the initiation of acquired movements. Although firing rates increase during swimming, mean synaptic inhibition greatly exceeds mean excitation, indicating that learned responses cannot result solely from changes in synaptic weight or upstream excitability that favor excitation. Estimates of spike threshold crossings based on measurements of intrinsic properties and the time course of synaptic currents demonstrate that noisy excitation can transiently outweigh noisy inhibition enough to increase firing rates at swimming onset. Thus, the millisecond-scale variance of synaptic currents can regulate cerebellar output, and the emergence of learned cerebellar behaviors may involve a time-based code.