Search

Spiking neurons encode information through their spiking temporal patterns. Although the precise spike-timing based encoding scheme has long been recognised, the exact mechanism that underlies the learning of such precise spike-timing in the brain remains an open question. Most of the existing learning methods for spiking neurons are based on synaptic weight adjustment. However, biological evidences suggest that synaptic delays can also be modulated to play an important role in the learning process. This paper investigates the viability of integrating synaptic delay plasticity into supervised learning and proposes a novel learning method that adjusts both the synaptic delays and weights of the learning neurons to make them fire precisely timed spikes, that is referred to as synaptic delay-weight plasticity. Remote Supervised Method (ReSuMe) and Perceptron Based Spiking Neuron Learning Rule (PBSNLR), two representative supervised learning methods, are studied to illustrate how the synaptic delay-weight plasticity works. The performance of the proposed learning method is thoroughly evaluated on synthetic data and is further demonstrated on real-world classification tasks. The experiments show that the synaptic delay-weight learning method outperforms the traditional synaptic weight learning methods in many ways. [ABSTRACT FROM AUTHOR]

With a similar capability of processing spikes as biological neural systems, networks of spiking neurons are expected to achieve a performance similar to that of living brains. Despite the achievement of spiking neuron based applications, most of them assume noise-free condition for learning and testing. This assumption, though fairly general, ignores the fact that noise widely exists in spiking neural networks (SNNs) and the neural response can be significantly disturbed by noise. Therefore, how to deal with noise is an important issue in the applications of SNNs. Here, by analyzing strategies employed to make spiking neurons robust to noise, also inspired by biological neurons, we propose a strategy that train spiking neurons with a dynamic firing threshold named noise-threshold. The noise-threshold can be applied by the existing supervised learning methods to improve the noise tolerance of them. Experimental results show that, with a combination of noise-threshold, the anti-noise capability of the existing supervised learning methods improves significantly, and the trained neuron can precisely and reliably reproduce target sequences of spikes even under highly noisy conditions. More importantly, the SNNs-based computational model equipped with a noise-threshold is more robust and can achieve a good performance even with different types of noise. Therefore, the noise-threshold is significant to practical applications and theoretical researches of SNNs. [ABSTRACT FROM AUTHOR]

Spiking Neural Networks (SNNs) hold great potential for mimicking the brain's efficient processing of information. Although biological evidence suggests that precise spike timing is crucial for effective information encoding, contemporary SNN research mainly concentrates on adjusting connection weights. In this work, we introduce Delay Learning based on Temporal Coding (DLTC), an innovative approach that integrates delay learning with a temporal coding strategy to optimize spike timing in SNNs. DLTC utilizes a learnable delay shift, which assigns varying levels of importance to different informational elements. This is complemented by an adjustable threshold that regulates firing times, allowing for earlier or later neuron activation as needed. We have tested DLTC's effectiveness in various contexts, including vision and auditory classification tasks, where it consistently outperformed traditional weight-only SNNs. The results indicate that DLTC achieves remarkable improvements in accuracy and computational efficiency, marking a step forward in advancing SNNs towards real-world applications. Our codes are accessible at https://github.com/sunpengfei1122/DLTC. [ABSTRACT FROM AUTHOR]

Decoding visual and auditory stimuli from brain activities, such as electroencephalography (EEG), offers promising advancements for enhancing machine-to-human interaction. However, effectively representing EEG signals remains a significant challenge. In this paper, we introduce a novel Delayed Knowledge Transfer (DKT) framework that employs spiking neurons for attention detection, using our experimental EEG dataset. This framework extracts patterns from audiovisual stimuli to model brain responses in EEG signals, while accounting for inherent response delays. By aligning audiovisual features with EEG signals through a shared embedding space, our approach improves the performance of brain–computer interface (BCI) systems. We also present WithMeAttention, a multimodal dataset designed to facilitate research in continuously distinguishing between target and distractor responses. Our methodology demonstrates a 3% improvement in accuracy on the WithMeAttention dataset compared to a baseline model that decodes EEG signals from scratch. This significant performance increase highlights the effectiveness of our approach Comprehensive analysis across four distinct conditions shows that rhythmic enhancement of visual information can optimize multi-sensory information processing. Notably, the two conditions featuring rhythmic target presentation – with and without accompanying beeps – achieved significantly superior performance compared to other scenarios. Furthermore, the delay distribution observed under different conditions indicates that our delay layer effectively emulates the neural processing delays in response to stimuli. [ABSTRACT FROM AUTHOR]

The deep learning era endows immense opportunities for ubiquitous robotic applications by leveraging big data generated from widespread sensors and ever-growing computing capability. While the growing demands for natural human-robot interaction (HRI) as well as concerns for energy efficiency, real-time performance, and data security motive novel solutions. In this paper, we present a brain-inspired spiking neural network (SNN) based Human-Robot Auditory Interface, namely HuRAI. The HuRAI integrates the voice activity detection, speaker localization and voice command recognition systems into a unified framework that can be implemented on the emerging low-power neuromorphic computing (NC) devices. Our experimental results demonstrate superior modeling capabilities of SNNs, achieving accurate and rapid prediction for each task. Moreover, the energy efficiency analysis reveals a compelling prospect, with up to three orders of magnitude energy savings, over the equivalent artificial neural networks that running on the state-of-the-art Nvidia graphics processing unit (GPU). Therefore, integrating the algorithmic power of large-scale SNN models and the energy efficiency of NC devices offers an attractive solution for real-time, low-power robotic applications. [ABSTRACT FROM AUTHOR]

Spiking Neural Networks (SNNs) have become one of the most prominent next-generation computational models owing to their biological plausibility, low power consumption, and the potential for neuromorphic hardware implementation. Among the various methods for obtaining available SNNs, converting Artificial Neural Networks (ANNs) into SNNs is the most cost-effective approach. The early challenges in ANN-to-SNN conversion work revolved around the susceptibility of converted SNNs to conversion errors. Some recent endeavors have attempted to mitigate these conversion errors by altering the original ANNs. Despite their ability to enhance the accuracy of SNNs, these methods lack generality and cannot be directly applied to convert the majority of existing ANNs. In this paper, we present a framework named DNISNM for converting ANN to SNN, with the aim of addressing conversion errors arising from differences in the discreteness and asynchrony of network transmission between ANN and SNN. The DNISNM consists of two mechanisms, Data-based Neuronal Initialization (DNI) and Signed Neuron with Memory (SNM), designed to respectively address errors stemming from discreteness and asynchrony disparities. This framework requires no additional modifications to the original ANN and can result in SNNs with improved accuracy performance, simultaneously ensuring universality, high precision, and low inference latency. We verify it experimentally on challenging object recognition datasets, including CIFAR10, CIFAR100, and ImageNet-1k. Experimental results show that the SNN converted by our framework has very high accuracy even at extremely low latency. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) are brain-inspired models that utilize discrete and sparse spikes to transmit information, thus having the property of energy efficiency. Recent advances in learning algorithms have greatly improved SNN performance due to the automation of feature engineering. While the choice of neural architecture plays a significant role in deep learning, the current SNN architectures are mainly designed manually, which is a time-consuming and error-prone process. In this paper, we propose a spiking neural architecture search (NAS) method that can automatically find efficient SNNs. To tackle the challenge of long search time faced by SNNs when utilizing NAS, the proposed NAS encodes candidate architectures in a branchless spiking supernet which significantly reduces the computation requirements in the search process. Considering that real-world tasks prefer efficient networks with optimal accuracy under a limited computational budget, we propose a Synaptic Operation (SynOps)-aware optimization to automatically find the computationally efficient subspace of the supernet. Experimental results show that, in less search time, our proposed NAS can find SNNs with higher accuracy and lower computational cost than state-of-the-art SNNs. We also conduct experiments to validate the search process and the trade-off between accuracy and computational cost. [ABSTRACT FROM AUTHOR]

Recommender systems have been comprehensively applied in many industries, such as social Web sites, e-commerce, tourism service and so on, although suffering from the data sparsity and cold start problems. Currently, due to the advantage of online social networking, many social network-based recommendation scenarios have been developed to improve the recommendation accuracy, via exploring hidden social relations between users from the social network. In this article, focusing on addressing these problems, a novel social network and preference-based recommendation method—SRMP, is proposed, which incorporates the social network information and user’s preference in matrix factorization for recommendation. In contrast to previous approaches, to improve the recommendation accuracy, SRMP performs recommendation in each independent sub-community, which is derived from the initial social community according to different category tags. The experimental analysis on large real-world datasets demonstrates that the proposed method SRMP outperforms state-of-the-art approaches, especially in recommendation accuracy and solving the cold start problem. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) have shown clear advantages over traditional artificial neural networks (ANNs) for low latency and high computational efficiency, due to their event-driven nature and sparse communication. However, the training of deep SNNs is not straightforward. In this paper, we propose a novel ANN-to-SNN conversion and layer-wise learning framework for rapid and efficient pattern recognition, which is referred to as progressive tandem learning. By studying the equivalence between ANNs and SNNs in the discrete representation space, a primitive network conversion method is introduced that takes full advantage of spike count to approximate the activation value of ANN neurons. To compensate for the approximation errors arising from the primitive network conversion, we further introduce a layer-wise learning method with an adaptive training scheduler to fine-tune the network weights. The progressive tandem learning framework also allows hardware constraints, such as limited weight precision and fan-in connections, to be progressively imposed during training. The SNNs thus trained have demonstrated remarkable classification and regression capabilities on large-scale object recognition, image reconstruction, and speech separation tasks, while requiring at least an order of magnitude reduced inference time and synaptic operations than other state-of-the-art SNN implementations. It, therefore, opens up a myriad of opportunities for pervasive mobile and embedded devices with a limited power budget. [ABSTRACT FROM AUTHOR]

Document-level relation extraction focuses on identifying the relations between entity pairs across the entirety of a document. However, current mainstream methods mainly have two drawbacks: (a) Separating the context encoding stage and the graph reasoning stage. (b) Limiting the reasoning between entity mentions, which ignores the relation correlations in context. In this paper, we introduce the SE-REPI model, which consists of a Structural-enhanced Encoding (SE) module and a Relation-based Entity-Pair-Level Interaction (REPI) module. The SE module effectively injects structural information into the architecture of the Transformer, integrating the graph reasoning stage into the encoding stage. The REPI module constructs entity-pair-level information interaction, which effectively captures the relation correlations among entity pairs. Comprehensive experimental results on four public DocRE datasets DocRED, CRD, GDA, and DWIE underscore the superior performance and efficiency of our model compared to previous leading-edge methods. Further experimental analyses reveal the interpretability of our approach. • We propose SE module and REPI module for document-level RE. • SE module merges graph reasoning with transformer encoding. • REPI module enhances interaction at the entity-pair level. • Our model outshines previous methods in performance and efficiency. [ABSTRACT FROM AUTHOR]

The attention mechanism has been proven to be an effective way to improve the performance of spiking neural networks (SNNs). However, from the perspective of tensor decomposition to examine the existing attention modules, we find that the rank of the attention maps generated by previous methods is fixed at 1, lacking the flexibility to adjust for specific tasks. To tackle this problem, we propose an attention module, namely Projected-full Attention (PFA), where the rank of the generated attention maps can be determined based on the characteristics of different tasks. Additionally, the parameter count of PFA grows linearly with the data scale. PFA is composed of the linear projection of spike tensor (LPST) module and attention map composing (AMC) module. In LPST, we start by compressing the original spike tensor into three projected tensors with learnable parameters for each dimension. Then, in AMC, we exploit the inverse procedure of the tensor decomposition process to combine the three tensors into the attention map using a so-called connecting factor. To validate the effectiveness of the proposed PFA module, we integrate it into the widely used VGG and ResNet architectures for classification tasks. Our method achieves state-of-the-art performance on both static and dynamic benchmark datasets, surpassing the existing SNN models with Transformer-based and CNN-based backbones. Code for PFA is available at https://github.com/RisingEntropy/PFA. [ABSTRACT FROM AUTHOR]

Existing multi-agent reinforcement learning methods employ a paradigm of centralized training with decentralized execution (CTDE) to learn cooperative policy among agents via coordination. However, within continuous destruction conditions, the inclusion of information from dead agents significantly undermines the ability to effectively learn cooperative policies in multi-agent systems. In this paper, we first analyze the bias introduced by dead agents under the CTDE paradigm and how it affects cooperation among agents. Following this, we propose q-learning based downsizing adaptive policy (QDAP) framework for cooperative multi-agent reinforcement learning. QDAP actively discerns relevant values from dead agents and utilizes an innovative approach to convert historical trajectories into weighting factors, thereby aiding remaining active agents in learning more appropriate cooperative policies. Moreover, we extend our proposed framework into the CTDE paradigm, facilitating seamless adaptation with the methods of value decomposition. Experimental results demonstrate that QDAP significantly improves learning speed and achieves superior cooperation performance on challenging Starcraft II micromanagement benchmark tasks. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) are regarded as effective models for processing spatio-temporal information. However, their inherent complexity of temporal coding makes it an arduous task to put forward an effective supervised learning algorithm, which still puzzles researchers in this area. In this paper, we propose a Recursive Least Squares-Based Learning Rule (RLSBLR) for SNN to generate the desired spatio-temporal spike train. During the learning process of our method, the weight update is driven by the cost function defined by the difference between the membrane potential and the firing threshold. The amount of weight modification depends not only on the impact of the current error function, but also on the previous error functions which are evaluated by current weights. In order to improve the learning performance, we integrate a modified synaptic delay learning to the proposed RLSBLR. We conduct experiments in different settings, such as spiking lengths, number of inputs, firing rates, noises and learning parameters, to thoroughly investigate the performance of this learning algorithm. The proposed RLSBLR is compared with competitive algorithms of Perceptron-Based Spiking Neuron Learning Rule (PBSNLR) and Remote Supervised Method (ReSuMe). Experimental results demonstrate that the proposed RLSBLR can achieve higher learning accuracy, higher efficiency and better robustness against different types of noise. In addition, we apply the proposed RLSBLR to open source database TIDIGITS, and the results show that our algorithm has a good practical application performance. [ABSTRACT FROM AUTHOR]

As the third generation of neural networks, Spiking Neural Networks (SNNs) have made great success in pattern recognition fields. However, the existing training methods for SNNs are not efficient enough because of the temporal encoding mechanism. To improve the training efficiency of the supervised SNNs and keep the useful temporal information, the Maximum Points-based Supervised Learning Rule (MPSLR) is proposed in this paper. Three training strategies are adopted in MPSLR to improve the learning performance. Firstly, only the target points and maximum voltage points are trained. By theoretical analyses, we find that the maximum points are effective for the voltage controlling of the non-target points, and the analytic solutions for all maximum voltage points are parallelly obtainable. This improves the training efficiency significantly by avoiding the successive voltage detecting. Secondly, the weight modification for each presynaptic neuron is normalized by a rate function to resizing the output scale. Thirdly, the spiking rates accumulated in a time window are utilized to involve more useful knowledge. Extensive experiments on both synthetic data and four real-world UCI datasets demonstrate that our algorithm achieves significantly better performance and higher efficiency than traditional methods in various situations, including different multi-spike rates and time lengths. Besides, it is more stable to hyper-parameter variations. [ABSTRACT FROM AUTHOR]

In recent years, deep spiking neural networks (SNNs) have demonstrated promising performance across various applications, owing to their low-power characteristics. Research on SNN meta-learning has enabled SNNs to reduce both label cost and computational power consumption in few-shot classification tasks. However, current SNN meta-learning methods still lag behind traditional artificial neural networks (ANNs) in terms of accuracy. In this work, we explore a two-stage metric-based SNN meta-learning framework that achieves the highest accuracy performance in SNN. This framework comprises a pre-training stage and a meta-training stage. During pre-training, a classification embedding SNN model (CESM) is trained to extract image features. Subsequently, in the meta-training stage, the meta embedding SNN model (MESM) employs the centered kernel alignment (CKA) method to measure the similarity between these learned features for meta-learning. We conduct extensive experiments on the Omniglot, tieredImageNet, and miniImageNet datasets, evaluating both CESM and MESM models. Experimental results demonstrate that the proposed framework improves performance by 5% on average compared to previous SNN meta-learning approaches. The proposed method surpasses the early classical ANN methods and further closes the gap with ANN state-of-the-art methods. • We explore an effective two-stage SNN meta-learning framework. • A CKA-based non-parametric classifier is proposed for better capturing spiking temporal feature similarities. • The framework outperforms the existing SNN meta-learning methods and has lower energy consumption than ANN. [ABSTRACT FROM AUTHOR]

The spiking neural networks (SNNs) are the third generation of artificial neural networks, which have made great achievements in the field of pattern recognition. However, the existing supervised training methods of SNNs are not efficient enough to meet the real-time requirement in most cases. To address this issue, the normalized perceptron based learning rule (NPBLR) is proposed in this paper for the supervised training of the multi-layer SNNs. Different from traditional methods, our algorithm only trains the selected misclassified time points and the target ones, employing the perceptron based neuron. Furthermore, the weight modification in our algorithm is normalized by a voltage based function, which is more efficient than the traditional time based method because the firing time is calculated by the voltage value. Superior to the traditional multi-layer algorithm ignoring the time accumulation of spikes, our algorithm defines the spiking activity of the postsynaptic neuron as the rate accumulation function of all presynaptic neurons in a specific time-frame. By these strategies, our algorithm overcomes some difficulties in the training of SNNs, e.g., the inefficient and no-fire problems. Comprehensive simulations are conducted both in single and multi-layer networks to investigate the learning performance of our algorithm, whose results demonstrate that our algorithm possesses higher learning efficiency and stronger parameter robustness than traditional algorithms. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) have drawn increasing attention in recent years and achieved remarkable performance in many graph-based tasks, especially in semi-supervised learning on graphs. However, most existing GNNs are based on the message-passing paradigm to iteratively aggregate neighborhood information in a single topology space. Despite their success, the expressive power of GNNs is limited by some drawbacks, such as inflexibility of message source expansion, negligence of node-level message output discrepancy, and restriction of single message space. To address these drawbacks, we present a novel message-passing paradigm, based on the properties of multi-step message source, node-specific message output, and multi-space message interaction. To verify its validity, we instantiate the new message-passing paradigm as a Dual-Perception Graph Neural Network (DPGNN), which applies a node-to-step attention mechanism to aggregate node-specific multi-step neighborhood information adaptively. Our proposed DPGNN can capture the structural neighborhood information and the feature-related information simultaneously for graph representation learning. Experimental results on six benchmark datasets with different topological structures demonstrate that our method outperforms the latest state-of-the-art models, which proves the superiority and versatility of our method. To our knowledge, we are the first to consider node-specific message passing in the GNNs. [ABSTRACT FROM AUTHOR]

The spiking neural networks (SNNs) are the third generation of neural networks and perform remarkably well in cognitive tasks such as pattern recognition. The spike emitting and information processing mechanisms found in biological cognitive systems motivate the application of the hierarchical structure and temporal encoding mechanism in spiking neural networks, which have exhibited strong computational capability. However, the hierarchical structure and temporal encoding approach require neurons to process information serially in space and time respectively, which reduce the training efficiency significantly. For training the hierarchical SNNs, most existing methods are based on the traditional back-propagation algorithm, inheriting its drawbacks of the gradient diffusion and the sensitivity on parameters. To keep the powerful computation capability of the hierarchical structure and temporal encoding mechanism, but to overcome the low efficiency of the existing algorithms, a new training algorithm, the Normalized Spiking Error Back Propagation (NSEBP) is proposed in this paper. In the feedforward calculation, the output spike times are calculated by solving the quadratic function in the spike response model instead of detecting postsynaptic voltage states at all time points in traditional algorithms. Besides, in the feedback weight modification, the computational error is propagated to previous layers by the presynaptic spike jitter instead of the gradient decent rule, which realizes the layer-wised training. Furthermore, our algorithm investigates the mathematical relation between the weight variation and voltage error change, which makes the normalization in the weight modification applicable. Adopting these strategies, our algorithm outperforms the traditional SNN multi-layer algorithms in terms of learning efficiency and parameter sensitivity, that are also demonstrated by the comprehensive experimental results in this paper. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) have gained a lot of attention and achievements recently because of their low-power advantages on neuromorphic hardware. However, training deep SNNs still requires a large number of labeled data which are expensive to obtain. To address this issue, we propose an effective Bio-inspired Active Learning (BAL) method in this paper to reduce the training cost of SNN models. Specifically, bio-inspired behavior patterns of spiking neurons are defined to represent the internal states of SNN models for active learning. Then, an active learning sample selection strategy is proposed by leveraging the empirical and generalization pattern divergence in SNNs. By labeling selected samples and adding them to training, behavioral patterns can be optimized to improve the performance of neural networks. Comprehensive experiments are conducted on the CIFAR-10, SVHN, and Fashion-MNIST datasets with various sample proportions. The experimental results demonstrate that the proposed BAL achieves state-of-the-arts performance in SNNs compared with the existing active learning methods. • BAL exploits the active learning feasibility in spiking neural networks. • We propose the neuron behavior patterns based on the inner states of spiking neurons. • Experiments are conducted to demonstrate the effectiveness of BAL. [ABSTRACT FROM AUTHOR]

Spiking Neural Network (SNN) has recently gained significant momentum in the neuromorphic low-power systems. However, the existing SNN models have limited use in time-sequential feature learning, and the exhausting spike encoding and decoding make the SNNs not straightforward to use. Inspired by the functional organization in the primate visual system, we propose an end-to-end functional spiking model in this paper to address these issues. Specifically, we propose the functional spike response to make each neuron special, and the dynamic synaptic efficiency to make the transmission of each input signal controllable. We represent inputs by a simple two-tuple set instead of conventional complex encoding, which achieves end-to-end learning. Experiments on synthetic datasets demonstrate that employing the two-tuple encoding strategy, our method improves the accuracy of the traditional SNN model significantly. In addition, we apply our method to seven real-world datasets and one human motion prediction dataset to investigate its performance. Experimental results show that the proposed functional spike response organization saves the running time of our model compared with the LSTM, GRU and one of the state-of-the-art time series processing methods. [ABSTRACT FROM AUTHOR]