Real-time execution of SNN models with synaptic plasticity for handwritten digit recognition on SIMD hardware.

Recent advancements in neuromorphic computing have led to the development of hardware architectures inspired by Spiking Neural Networks (SNNs) to emulate the efficiency and parallel processing capabilities of the human brain. This work focuses on testing the HEENS architecture, specifically designed for high parallel processing and biological realism in SNN emulation, implemented on a ZYNQ family FPGA. The study applies this architecture to the classification of digits using the well-known MNIST database. The image resolutions were adjusted to match HEENS' processing capacity. Results were compared with existing work, demonstrating HEENS' performance comparable to other solutions. This study highlights the importance of balancing accuracy and efficiency in the execution of applications. HEENS offers a flexible solution for SNN emulation, allowing for the implementation of programmable neural and synaptic models. It encourages the exploration of novel algorithms and network architectures, providing an alternative for real-time processing with efficient energy consumption.

Recall tempo of Hebbian sequences depends on the interplay of Hebbian kernel with tutor signal timing.

Understanding how neural circuits generate sequential activity is a longstanding challenge. While foundational theoretical models have shown how sequences can be stored as memories in neural networks with Hebbian plasticity rules, these models considered only a narrow range of Hebbian rules. Here, we introduce a model for arbitrary Hebbian plasticity rules, capturing the diversity of spike-timing-dependent synaptic plasticity seen in experiments, and show how the choice of these rules and of neural activity patterns influences sequence memory formation and retrieval. In particular, we derive a general theory that predicts the tempo of sequence replay. This theory lays a foundation for explaining how cortical tutor signals might give rise to motor actions that eventually become "automatic." Our theory also captures the impact of changing the tempo of the tutor signal. Beyond shedding light on biological circuits, this theory has relevance in artificial intelligence by laying a foundation for frameworks whereby slow and computationally expensive deliberation can be stored as memories and eventually replaced by inexpensive recall.

Coordinated reset stimulation of plastic neural networks with spatially dependent synaptic connections.

Background: Abnormal neuronal synchrony is associated with several neurological disorders, including Parkinson's disease (PD), essential tremor, dystonia, and epilepsy. Coordinated reset (CR) stimulation was developed computationally to counteract abnormal neuronal synchrony. During CR stimulation, phase-shifted stimuli are delivered to multiple stimulation sites. Computational studies in plastic neural networks reported that CR stimulation drove the networks into an attractor of a stable desynchronized state by down-regulating synaptic connections, which led to long-lasting desynchronization effects that outlasted stimulation. Later, corresponding long-lasting desynchronization and therapeutic effects were found in animal models of PD and PD patients. To date, it is unclear how spatially dependent synaptic connections, as typically observed in the brain, shape CR-induced synaptic downregulation and long-lasting effects. Methods: We performed numerical simulations of networks of leaky integrate-and-fire neurons with spike-timing-dependent plasticity and spatially dependent synaptic connections to study and further improve acute and long-term responses to CR stimulation. Results: The characteristic length scale of synaptic connections relative to the distance between stimulation sites plays a key role in CR parameter adjustment. In networks with short synaptic length scales, a substantial synaptic downregulation can be achieved by selecting appropriate stimulus-related parameters, such as the stimulus amplitude and shape, regardless of the employed spatiotemporal pattern of stimulus deliveries. Complex stimulus shapes can induce local connectivity patterns in the vicinity of the stimulation sites. In contrast, in networks with longer synaptic length scales, the spatiotemporal sequence of stimulus deliveries is of major importance for synaptic downregulation. In particular, rapid shuffling of the stimulus sequence is advantageous for synaptic downregulation. Conclusion: Our results suggest that CR stimulation parameters can be adjusted to synaptic connectivity to further improve the long-lasting effects. Furthermore, shuffling of CR sequences is advantageous for long-lasting desynchronization effects. Our work provides important hypotheses on CR parameter selection for future preclinical and clinical studies.

Bidirectional modulation of synaptic transmission by insulin-like growth factor-I.

Insulin-like growth factor-I (IGF-I) plays a key role in the modulation of synaptic plasticity and is an essential factor in learning and memory processes. However, during aging, IGF-I levels are decreased, and the effect of this decrease in the induction of synaptic plasticity remains unknown. Here we show that the induction of N-methyl-D-aspartate receptor (NMDAR)-dependent long-term potentiation (LTP) at layer 2/3 pyramidal neurons (PNs) of the mouse barrel cortex is favored or prevented by IGF-I (10 nM) or IGF-I (7 nM), respectively, when IGF-I is applied 1 h before the induction of Hebbian LTP. Analyzing the cellular basis of this bidirectional control of synaptic plasticity, we observed that while 10 nM IGF-I generates LTP (LTPIGF-I) of the post-synaptic potentials (PSPs) by inducing long-term depression (LTD) of the inhibitory post-synaptic currents (IPSCs), 7 nM IGF-I generates LTD of the PSPs (LTDIGF-I) by inducing LTD of the excitatory post-synaptic currents (EPSCs). This bidirectional effect of IGF-I is supported by the observation of IGF-IR immunoreactivity at both excitatory and inhibitory synapses. Therefore, IGF-I controls the induction of Hebbian NMDAR-dependent plasticity depending on its concentration, revealing novel cellular mechanisms of IGF-I on synaptic plasticity and in the learning and memory machinery of the brain.

The spike-timing-dependent plasticity of VIP interneurons in motor cortex.

The plasticity of inhibitory interneurons (INs) plays an important role in the organization and maintenance of cortical microcircuits. Given the many different IN types, there is an even greater diversity in synapse-type-specific plasticity learning rules at excitatory to excitatory (E→I), I→E, and I→I synapses. I→I synapses play a key disinhibitory role in cortical circuits. Because they typically target other INs, vasoactive intestinal peptide (VIP) INs are often featured in I→I→E disinhibition, which upregulates activity in nearby excitatory neurons. VIP IN dysregulation may thus lead to neuropathologies such as epilepsy. In spite of the important activity regulatory role of VIP INs, their long-term plasticity has not been described. Therefore, we characterized the phenomenology of spike-timing-dependent plasticity (STDP) at inputs and outputs of genetically defined VIP INs. Using a combination of whole-cell recording, 2-photon microscopy, and optogenetics, we explored I→I STDP at layer 2/3 (L2/3) VIP IN outputs onto L5 Martinotti cells (MCs) and basket cells (BCs). We found that VIP IN→MC synapses underwent causal long-term depression (LTD) that was presynaptically expressed. VIP IN→BC connections, however, did not undergo any detectable plasticity. Conversely, using extracellular stimulation, we explored E→I STDP at inputs to VIP INs which revealed long-term potentiation (LTP) for both causal and acausal timings. Taken together, our results demonstrate that VIP INs possess synapse-type-specific learning rules at their inputs and outputs. This suggests the possibility of harnessing VIP IN long-term plasticity to control activity-related neuropathologies such as epilepsy.

Simulated dataset on coordinated reset stimulation of homogeneous and inhomogeneous networks of excitatory leaky integrate-and-fire neurons with spike-timing-dependent plasticity.

We present simulated data on coordinated reset stimulation (CRS) of plastic neuronal networks. The neuronal network consists of excitatory leaky integrate-and-fire neurons and plasticity is implemented as spike-timing-dependent plasticity (STDP). A synchronized state with strong synaptic connectivity and a desynchronized state with weak synaptic connectivity coexist. CRS may drive the network from the synchronized state into a desynchronized state inducing long-lasting desynchronization effects that persist after cessation of stimulation. This is used to model brain stimulation-induced transitions between a pathological state, with abnormally strong neuronal synchrony, and a physiological state, e.g., in Parkinson's disease. During CRS, a sequence of stimuli is delivered to multiple stimulation sites - called CR sequence. We present simulated data for the analysis of long-lasting desynchronization effects of CRS with shuffled CR sequences versus non-shuffled CR sequences in which the order of stimulus deliveries to the sites remains unchanged throughout the entire stimulation period. Such data are presented for networks with homogeneous synaptic connectivity and networks with inhomogeneous synaptic connectivity. Homogeneous synaptic connectivity refers to a network in which the probability of a synaptic connection does not depend on the pre- and postsynaptic neurons' locations. In contrast, inhomogeneous synaptic connectivity refers to a network in which the probability of a synaptic connection depends on the neurons' locations. The presented neuronal network model was used to analyse the impact of the CR sequences and their shuffling on the long-lasting effects of CRS [1].

Enhanced representation learning with temporal coding in sparsely spiking neural networks.

Current representation learning methods in Spiking Neural Networks (SNNs) rely on rate-based encoding, resulting in high spike counts, increased energy consumption, and slower information transmission. In contrast, our proposed method, Weight-Temporally Coded Representation Learning (W-TCRL), utilizes temporally coded inputs, leading to lower spike counts and improved efficiency. To address the challenge of extracting representations from a temporal code with low reconstruction error, we introduce a novel Spike-Timing-Dependent Plasticity (STDP) rule. This rule enables stable learning of relative latencies within the synaptic weight distribution and is locally implemented in space and time, making it compatible with neuromorphic processors. We evaluate the performance of W-TCRL on the MNIST and natural image datasets for image reconstruction tasks. Our results demonstrate relative improvements of 53% for MNIST and 75% for natural images in terms of reconstruction error compared to the SNN state of the art. Additionally, our method achieves significantly higher sparsity, up to 900 times greater, when compared to related work. These findings emphasize the efficacy of W-TCRL in leveraging temporal coding for enhanced representation learning in Spiking Neural Networks.

Synaptic plasticity through a naturalistic lens.

From the myriad of studies on neuronal plasticity, investigating its underlying molecular mechanisms up to its behavioral relevance, a very complex landscape has emerged. Recent efforts have been achieved toward more naturalistic investigations as an attempt to better capture the synaptic plasticity underpinning of learning and memory, which has been fostered by the development of in vivo electrophysiological and imaging tools. In this review, we examine these naturalistic investigations, by devoting a first part to synaptic plasticity rules issued from naturalistic in vivo-like activity patterns. We next give an overview of the novel tools, which enable an increased spatio-temporal specificity for detecting and manipulating plasticity expressed at individual spines up to neuronal circuit level during behavior. Finally, we put particular emphasis on works considering brain-body communication loops and macroscale contributors to synaptic plasticity, such as body internal states and brain energy metabolism.

The Enhanced Performance of Neuromorphic Computing Hardware in an ITO/ZnO/HfOx/W Bilayer-Structured Memory Device.

This study discusses the potential application of ITO/ZnO/HfOx/W bilayer-structured memory devices in neuromorphic systems. These devices exhibit uniform resistive switching characteristics and demonstrate favorable endurance (>102) and stable retention (>104 s). Notably, the formation and rupture of filaments at the interface of ZnO and HfOx contribute to a higher ON/OFF ratio and improve cycle uniformity compared to RRAM devices without the HfOx layer. Additionally, the linearity of potentiation and depression responses validates their applicability in neural network pattern recognition, and spike-timing-dependent plasticity (STDP) behavior is observed. These findings collectively suggest that the ITO/ZnO/HfOx/W structure holds the potential to be a viable memory component for integration into neuromorphic systems.

Improved Resistive and Synaptic Characteristics in Neuromorphic Systems Achieved Using the Double-Forming Process.

In this study, we investigate the electrical properties of ITO/ZrOx/TaN RRAM devices for neuromorphic computing applications. The thickness and material composition of the device are analyzed using transmission electron microscopy. Additionally, the existence of TaON interface layers was confirmed using dispersive X-ray spectroscopy and X-ray photoelectron analysis. The forming process of the ZrOx-based device can be divided into two categories, namely single- and double forming, based on the initial lattice oxygen vacancies. The resistive switching behaviors of the two forming methods are compared in terms of the uniformity properties of endurance and retention. The rationale behind each I-V forming process was determined as follows: in the double-forming method case, an energy band diagram was constructed using F-N tunneling; conversely, in the single-forming method case, the ratio of oxygen vacancies was extracted based on XPS analysis to identify the conditions for filament formation. Subsequently, synaptic simulations for the applications of neuromorphic systems were conducted using a pulse scheme to achieve potentiation and depression with a deep neural network-based pattern recognition system to display the achieved recognition accuracy. Finally, high-order synaptic plasticity (spike-timing-dependent plasticity (STDP)) is emulated based on the Hebbian rule.

TiN/Ti/HfO2/TiN memristive devices for neuromorphic computing: from synaptic plasticity to stochastic resonance.

We characterize TiN/Ti/HfO2/TiN memristive devices for neuromorphic computing. We analyze different features that allow the devices to mimic biological synapses and present the models to reproduce analytically some of the data measured. In particular, we have measured the spike timing dependent plasticity behavior in our devices and later on we have modeled it. The spike timing dependent plasticity model was implemented as the learning rule of a spiking neural network that was trained to recognize the MNIST dataset. Variability is implemented and its influence on the network recognition accuracy is considered accounting for the number of neurons in the network and the number of training epochs. Finally, stochastic resonance is studied as another synaptic feature. It is shown that this effect is important and greatly depends on the noise statistical characteristics.

The Memristive Properties and Spike Timing-Dependent Plasticity in Electrodeposited Copper Tungstates and Molybdates.

Memristors possess non-volatile memory, adjusting their electrical resistance to the current that flows through them and allowing switching between high and low conducting states. This technology could find applications in fields such as IT, consumer electronics, computing, sensors, and medicine. In this paper, we report successful electrodeposition of thin-film materials consisting of copper tungstate and copper molybdate (CuWO4 and Cu3Mo2O9), which showed notable memristive properties. Material characterisation was performed with techniques such as XRD, XPS, and SEM. The electrodeposited materials exhibited the ability to switch between low and high resistive states during varied cyclic scans and short-term impulses. The retention time of these switched states was also explored. Using these materials, the effects seen in biological systems, specifically spike timing-dependent plasticity, were simulated, being based on analogue operation of the memristors to achieve multiple conductivity states. Bio-inspired simulations performed directly on the material could possibly offer energy and time savings for classical computations. Memristors could be crucial for the advancement of high-efficiency, low-energy neuromorphic electronic devices and technologies in the future.

Improved Uniformity of TaOx-Based Resistive Switching Memory Device by Inserting Thin SiO2 Layer for Neuromorphic System.

RRAM devices operating based on the creation of conductive filaments via the migration of oxygen vacancies are widely studied as promising candidates for next-generation memory devices due to their superior memory characteristics. However, the issues of variation in the resistance state and operating voltage remain key issues that must be addressed. In this study, we propose a TaOx/SiO2 bilayer device, where the inserted SiO2 layer localizes the conductive path, improving uniformity during cycle-to-cycle endurance and retention. Transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS) confirm the device structure and chemical properties. In addition, various electric pulses are used to investigate the neuromorphic system properties of the device, revealing its good potential for future memory device applications.

Double-Forming Mechanism of TaOx-Based Resistive Memory Device and Its Synaptic Applications.

The bipolar resistive switching properties of Pt/TaOx/InOx/ITO-resistive random-access memory devices under DC and pulse measurement conditions are explored in this work. Transmission electron microscopy and X-ray photoelectron spectroscopy were used to confirm the structure and chemical compositions of the devices. A unique two-step forming process referred to as the double-forming phenomenon and self-compliance characteristics are demonstrated under a DC sweep. A model based on oxygen vacancy migration is proposed to explain its conduction mechanism. Varying reset voltages and compliance currents were applied to evaluate multilevel cell characteristics. Furthermore, pulses were applied to the devices to demonstrate the neuromorphic system's application via testing potentiation, depression, spike-timing-dependent plasticity, and spike-rate-dependent plasticity.

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.

Incorporating structural plasticity into self-organization recurrent networks for sequence learning.

Introduction: Spiking neural networks (SNNs), inspired by biological neural networks, have received a surge of interest due to its temporal encoding. Biological neural networks are driven by multiple plasticities, including spike timing-dependent plasticity (STDP), structural plasticity, and homeostatic plasticity, making network connection patterns and weights to change continuously during the lifecycle. However, it is unclear how these plasticities interact to shape neural networks and affect neural signal processing. Method: Here, we propose a reward-modulated self-organization recurrent network with structural plasticity (RSRN-SP) to investigate this issue. Specifically, RSRN-SP uses spikes to encode information, and incorporate multiple plasticities including reward-modulated spike timing-dependent plasticity (R-STDP), homeostatic plasticity, and structural plasticity. On the one hand, combined with homeostatic plasticity, R-STDP is presented to guide the updating of synaptic weights. On the other hand, structural plasticity is utilized to simulate the growth and pruning of synaptic connections. Results and discussion: Extensive experiments for sequential learning tasks are conducted to demonstrate the representational ability of the RSRN-SP, including counting task, motion prediction, and motion generation. Furthermore, the simulations also indicate that the characteristics arose from the RSRN-SP are consistent with biological observations.

Skewed distribution of spines is independent of presynaptic transmitter release and synaptic plasticity, and emerges early during adult neurogenesis.

Dendritic spines are crucial for excitatory synaptic transmission as the size of a spine head correlates with the strength of its synapse. The distribution of spine head sizes follows a lognormal-like distribution with more small spines than large ones. We analysed the impact of synaptic activity and plasticity on the spine size distribution in adult-born hippocampal granule cells from rats with induced homo- and heterosynaptic long-term plasticity in vivo and CA1 pyramidal cells from Munc13-1/Munc13-2 knockout mice with completely blocked synaptic transmission. Neither the induction of extrinsic synaptic plasticity nor the blockage of presynaptic activity degrades the lognormal-like distribution but changes its mean, variance and skewness. The skewed distribution develops early in the life of the neuron. Our findings and their computational modelling support the idea that intrinsic synaptic plasticity is sufficient for the generation, while a combination of intrinsic and extrinsic synaptic plasticity maintains lognormal-like distribution of spines.

Spiking neural networks produce informational closure by stimulus avoidance.

The concept of Learning by Stimulus Avoidance (LSA) has been proposed in recent literature, and the methods of avoiding stimuli: action, prediction, and separation appear to align well with the formation of Bertschinger's informational closure. In this study, we provide experimental evidence demonstrating that spiking neural networks, which avoid stimuli, can indeed facilitate the emergence of informational closure. The established link between LSA and informational closure lays the foundation for further exploration of autopoietic relationships and the self-organization of closure within neural networks.

Controlling neocortical epileptic seizures using forced temporal spike-time stimulation: an in silico computational study.

Epileptic seizure is typically characterized by highly synchronized episodes of neural activity. Existing stimulation therapies focus purely on suppressing the pathologically synchronized neuronal firing patterns during the ictal (seizure) period. While these strategies are effective in suppressing seizures when they occur, they fail to prevent the re-emergence of seizures once the stimulation is turned off. Previously, we developed a novel neurostimulation motif, which we refer to as "Forced Temporal Spike-Time Stimulation" (FTSTS) that has shown remarkable promise in long-lasting desynchronization of excessively synchronized neuronal firing patterns by harnessing synaptic plasticity. In this paper, we build upon this prior work by optimizing the parameters of the FTSTS protocol in order to efficiently desynchronize the pathologically synchronous neuronal firing patterns that occur during epileptic seizures using a recently published computational model of neocortical-onset seizures. We show that the FTSTS protocol applied during the ictal period can modify the excitatory-to-inhibitory synaptic weight in order to effectively desynchronize the pathological neuronal firing patterns even after the ictal period. Our investigation opens the door to a possible new neurostimulation therapy for epilepsy.

Unsupervised learning in hexagonal boron nitride memristor-based spiking neural networks.

Resistive random access memory (RRAM) is an emerging non-volatile memory technology that can be used in neuromorphic computing hardware to exceed the limitations of traditional von Neumann architectures by merging processing and memory units. Two-dimensional (2D) materials with non-volatile switching behavior can be used as the switching layer of RRAMs, exhibiting superior behavior compared to conventional oxide-based devices. In this study, we investigate the electrical performance of 2D hexagonal boron nitride (h-BN) memristors towards their implementation in spiking neural networks (SNN). Based on experimental behavior of the h-BN memristors as artificial synapses, we simulate the implementation of unsupervised learning in SNN for image classification on the Modified National Institute of Standards and Technology dataset. Additionally, we propose a simple spike-timing-dependent-plasticity (STDP)-based dropout technique to enhance the recognition rate in h-BN memristor-based SNN. Our results demonstrate the viability of using 2D-material-based memristors as artificial synapses to perform unsupervised learning in SNN using hardware-friendly methods for online learning.