Offloading large language models (LLMs) states to host memory during inference promises to reduce operational costs by supporting larger models, longer prompts, and larger batch sizes. However, the design of existing memory offloading mechanisms does not take latency service-level objectives (SLOs) into consideration. As a result, they either lead to frequent SLO violations or underutilize host memory, thereby incurring economic loss and thus defeating the purpose of memory offloading.

Temporal processing is vital for extracting meaningful information from time-varying signals. Recent advancements in Spiking Neural Networks (SNNs) have shown immense promise in efficiently processing these signals. However, progress in this field has been impeded by the lack of effective and standardized benchmarks, which complicates the consistent measurement of technological advancements and limits the practical applicability of SNNs. To bridge this gap, we introduce the Neuromorphic Sequential Arena (NSA), a comprehensive benchmark that offers an effective, versatile, and application-oriented evaluation framework for neuromorphic temporal processing. The NSA includes seven real-world temporal processing tasks from a diverse range of application scenarios, each capturing rich temporal dynamics across multiple timescales. Utilizing NSA, we conduct extensive comparisons of recently introduced spiking neuron models and neural architectures, presenting comprehensive baselines in terms of task performance, training speed, memory usage, and energy efficiency. Our findings emphasize an urgent need for efficient SNN designs that can consistently deliver high performance across tasks with varying temporal complexities while maintaining low computational costs. NSA enables systematic tracking of advancements in neuromorphic algorithm research and paves the way for developing effective and efficient neuromorphic temporal processing systems.

Accurately predicting the occurrence, amplitude, and propagation speed of traffic flow shockwaves are essential for dynamic traffic control to mitigate traffic congestion. However, traditional approaches primarily rely on macroscopic traffic data from loop detectors, which often fail to capture the fine-grained interactions among vehicles and roadway features, leading to limited prediction accuracy. To address this gap, we propose a novel vehicle trajectory prediction method based on geometric deep learning (GNN-T), integrating a graph neural network with an attention mechanism. The graph neural network captures vehicle and lane features, and the attention mechanism models interactions between vehicles and between vehicles and lanes, significantly enhancing prediction performance. The evaluation results show that GNN-T outperforms conventional car-following models and other baseline methods, achieving improvements in trajectory prediction accuracy by 7.1% to 9.5%. Furthermore, a traffic flow shockwave prediction model, based on wavelet transform and predicted trajectories, is developed to predict shockwave occurrence, timing, propagation speed, and amplitude. Compared to the Intelligent Driver Model and traditional traffic flow detection methods, our approach demonstrates superior performance, with accuracy rates in shockwave prediction up to 93.5% in validation datasets. These results indicate the potential of the proposed model for practical applications in traffic flow management.

Speech enhancement (SE) is critical for improving speech intelligibility and quality in various audio devices. In recent years, deep learning-based methods have significantly improved SE performance, but they often come with a high computational cost, which is prohibitive for a large number of edge devices, such as headsets and hearing aids. This work proposes an ultralow-power SE system based on the brain-inspired spiking neural network (SNN) called Spiking-FullSubNet. Spiking-FullSubNet follows a full-band and subband fusioned approach to effectively capture both global and local spectral information. To enhance the efficiency of computationally expensive subband modeling, we introduce a frequency partitioning method inspired by the sensitivity profile of the human peripheral auditory system. Furthermore, we introduce a novel spiking neuron model that can dynamically control the input information integration and forgetting, enhancing the multiscale temporal processing capability of SNN, which is critical for speech denoising. Experiments conducted on the recent Intel Neuromorphic Deep Noise Suppression (N-DNS) Challenge dataset show that the Spiking-FullSubNet surpasses state-of-the-art (SOTA) methods by large margins in terms of both speech quality and energy efficiency metrics. Notably, our system won the championship of the Intel N-DNS Challenge (algorithmic track), opening up a myriad of opportunities for ultralow-power SE at the edge. Our source code and model checkpoints are publicly available at github.com/haoxiangsnr/spiking-fullsubnet.

The identification of sensory cues associated with potential opportunities and dangers is frequently complicated by unrelated events that separate useful cues by long delays. As a result, it remains a challenging task for state-of-the-art spiking neural networks (SNNs) to establish long-term temporal dependency between distant cues. To address this challenge, we propose a novel biologically inspired Two-Compartment Leaky Integrate-and-Fire spiking neuron model, dubbed TC-LIF. The proposed model incorporates carefully designed somatic and dendritic compartments that are tailored to facilitate learning long-term temporal dependencies. Furthermore, the theoretical analysis is provided to validate the effectiveness of TC-LIF in propagating error gradients over an extended temporal duration. Our experimental results, on a diverse range of temporal classification tasks, demonstrate superior temporal classification capability, rapid training convergence, and high energy efficiency of the proposed TC-LIF model. Therefore, this work opens up a myriad of opportunities for solving challenging temporal processing tasks on emerging neuromorphic computing systems. Our code is publicly available at https://github.com/ZhangShimin1/TC-LIF.

Deep neural networks are typically trained using global error signals that backpropagate (BP) end-to-end, which is not only biologically implausible but also suffers from the update locking problem and requires huge memory consumption. Local learning, which updates each layer independently with a gradient-isolated auxiliary network, offers a promising alternative to address the above problems. However, existing local learning methods are confronted with a large accuracy gap with the BP counterpart, particularly for large-scale networks. This is due to the weak coupling between local layers and their subsequent network layers, as there is no gradient communication across layers. To tackle this issue, we put forward an augmented local learning method, dubbed AugLocal. AugLocal constructs each hidden layer’s auxiliary network by uniformly selecting a small subset of layers from its subsequent network layers to enhance their synergy. We also propose to linearly reduce the depth of auxiliary networks as the hidden layer goes deeper, ensuring sufficient network capacity while reducing the computational cost of auxiliary networks. Our extensive experiments on four image classification datasets (i.e., CIFAR-10, SVHN, STL-10, and ImageNet) demonstrate that AugLocal can effectively scale up to tens of local layers with a comparable accuracy to BP-trained networks while reducing GPU memory usage by around 40%. The proposed AugLocal method, therefore, opens up a myriad of opportunities for training high-performance deep neural networks on resource-constrained platforms. Code is available at \url{https://github.com/ChenxiangMA/AugLocal}.

The brain-inspired Spiking Neural Networks (SNNs) have garnered considerable research interest due to their superior performance and energy efficiency in processing temporal signals. Recently, a novel multi-compartment spiking neuron model, namely the Two-Compartment LIF (TC-LIF) model, has been proposed and exhibited a remarkable capacity for sequential modelling. However, training the TC-LIF model presents challenges stemming from the large memory consumption and the issue of vanishing gradient associated with the Backpropagation Through Time (BPTT) algorithm. To address these challenges, online learning methodologies emerge as a promising solution. Yet, to date, the application of online learning methods in SNNs has been predominantly confined to simplified Leaky Integrate-and-Fire (LIF) neuron models. In this paper, we present a novel online learning method specifically tailored for networks of TC-LIF neurons. Additionally, we propose a refined TC-LIF neuron model called Adaptive TC-LIF, which is carefully designed to enhance temporal information integration in online learning scenarios. Extensive experiments, conducted on various sequential benchmarks, demonstrate that our approach successfully preserves the superior sequential modeling capabilities of the TC-LIF neuron while incorporating the training efficiency and hardware friendliness of online learning. As a result, it offers a multitude of opportunities to leverage neuromorphic solutions for processing temporal signals.

Backpropagation has been successfully generalized to optimize deep spiking neural networks (SNNs), where, nevertheless, gradients need to be propagated back through all layers, resulting in a massive consumption of computing resources and an obstacle to the parallelization of training. A biologically motivated scheme of local learning provides an alternative to efficiently train deep networks but often suffers a low performance of accuracy on practical tasks. Thus, how to train deep SNNs with the local learning scheme to achieve both efficient and accurate performance still remains an important challenge. In this study, we focus on a supervised local learning scheme where each layer is independently optimized with an auxiliary classifier. Accordingly, we first propose a spike-based efficient local learning rule by only considering the direct dependencies in the current time. We then propose two variants that additionally incorporate temporal dependencies through a backward and forward process, respectively. The effectiveness and performance of our proposed methods are extensively evaluated with six mainstream datasets. Experimental results show that our methods can successfully scale up to large networks and substantially outperform the spike-based local learning baselines on all studied benchmarks. Our results also reveal that gradients with temporal dependencies are essential for high performance on temporal tasks, while they have negligible effects on rate-based tasks. Our work is significant as it brings the performance of spike-based local learning to a new level with the computational benefits being retained.

Spiking neural networks (SNNs) are considered as a potential candidate to overcome current challenges, such as the high-power consumption encountered by artificial neural networks (ANNs); however, there is still a gap between them with respect to the recognition accuracy on various tasks. A conversion strategy was, thus, introduced recently to bridge this gap by mapping a trained ANN to an SNN. However, it is still unclear that to what extent this obtained SNN can benefit both the accuracy advantage from ANN and high efficiency from the spike-based paradigm of computation. In this article, we propose two new conversion methods, namely TerMapping and AugMapping. The TerMapping is a straightforward extension of a typical threshold-balancing method with a double-threshold scheme, while the AugMapping additionally incorporates a new scheme of augmented spike that employs a spike coefficient to carry the number of typical all-or-nothing spikes occurring at a time step. We examine the performance of our methods based on the MNIST, Fashion-MNIST, and CIFAR10 data sets. The results show that the proposed double-threshold scheme can effectively improve the accuracies of the converted SNNs. More importantly, the proposed AugMapping is more advantageous for constructing accurate, fast, and efficient deep SNNs compared with other state-of-the-art approaches. Our study, therefore, provides new approaches for further integration of advanced techniques in ANNs to improve the performance of SNNs, which could be of great merit to applied developments with spike-based neuromorphic computing.

Traditional neuron models use analog values for information representation and computation, while all-or-nothing spikes are employed in the spiking ones. With a more brain-like processing paradigm, spiking neurons are more promising for improvements in efficiency and computational capability. They extend the computation of traditional neurons with an additional dimension of time carried by all-or-nothing spikes. Could one benefit from both the accuracy of analog values and the time-processing capability of spikes? In this article, we introduce a concept of augmented spikes to carry complementary information with spike coefficients in addition to spike latencies. New augmented spiking neuron model and synaptic learning rules are proposed to process and learn patterns of augmented spikes. We provide systematic insights into the properties and characteristics of our methods, including classification of augmented spike patterns, learning capacity, construction of causality, feature detection, robustness, and applicability to practical tasks, such as acoustic and visual pattern recognition. Our augmented approaches show several advanced learning properties and reliably outperform the baseline ones that use typical all-or-nothing spikes. Our approaches significantly improve the accuracies of a temporal-based approach on sound and MNIST recognition tasks to 99.38% and 97.90%, respectively, highlighting the effectiveness and potential merits of our methods. More importantly, our augmented approaches are versatile and can be easily generalized to other spike-based systems, contributing to a potential development for them, including neuromorphic computing.

Biological systems under a parallel and spike-based computation endow individuals with abilities to have prompt and reliable responses to different stimuli. Spiking neural networks (SNNs) have thus been developed to emulate their efficiency and to explore principles of spike-based processing. However, the design of a biologically plausible and efficient SNN for image classification still remains as a challenging task. Previous efforts can be generally clustered into two major categories in terms of coding schemes being employed: rate and temporal. The rate-based schemes suffer inefficiency, whereas the temporal-based ones typically end with a relatively poor performance in accuracy. It is intriguing and important to develop an SNN with both efficiency and efficacy being considered. In this article, we focus on the temporal-based approaches in a way to advance their accuracy performance by a great margin while keeping the efficiency on the other hand. A new temporal-based framework integrated with the multispike learning is developed for efficient recognition of visual patterns. Different approaches of encoding and learning under our framework are evaluated with the MNIST and Fashion-MNIST data sets. Experimental results demonstrate the efficient and effective performance of our temporal-based approaches across a variety of conditions, improving accuracies to higher levels that are even comparable to rate-based ones but importantly with a lighter network structure and far less number of spikes. This article attempts to extend the advanced multispike learning to the challenging task of image recognition and bring state of the arts in temporal-based approaches to a novel level. The experimental results could be potentially favorable to low-power and high-speed requirements in the field of artificial intelligence and contribute to attract more efforts toward brain-like computing.

In addition to biological plausibility, spiking neural networks (SNNs) are drawing significant attention recently due to their promising advantages in computational efficiency, which could potentially help to overcome the consumption obstacle in deep learning. Training deep SNNs is of great importance for solving practical tasks. In this paper, we propose a new deep spike learning rule to train deep SNNs to associate input spike patterns with desired output spike numbers. Our proposed rule is able to construct error signals based on a critical time point that is likely close to change the neuron's response toward its desired. We evaluate the performance of our method with both static and dynamic vision datasets. Experimental results show that the proposed rule can effectively learn spike patterns encoded with both rate and temporal codes, and more importantly, achieves impressive accuracies on all benchmark datasets. We further provide a comprehensive analysis of both codes with respect to efficiency and robustness. Our study thus provides an effective rule that is generalized to process information under a broad range of coding schemes, which would be of great merit for spike-based learning and processing.

Spiking neural networks (SNNs) are promising to replicate the efficiency of the brain by utilizing a paradigm of spike-based computation. Training a deep SNN is of great importance for solving practical tasks as well as discovering the fascinating capability of spike-based computation. The biologically plausible scheme of local learning motivates many approaches that enable training deep networks in an efficient parallel way. However, most of the existing spike-based local learning approaches show relatively low performances on challenging tasks. In this paper, we propose a new spike-based temporal dependent local learning (TDLL) algorithm, where each hidden layer of a deep SNN is independently trained with an auxiliary trainable spiking projection layer, and temporal dependency is fully employed to construct local errors for adjusting parameters. We examine the performance of the proposed TDLL with various networks on the MNIST, Fashion-MNIST, SVHN and CIFAR-10 datasets. Experimental results highlight that our method can scale up to larger networks, and more importantly, achieves relatively high accuracies on all benchmarks, which are even competitive with the ones obtained by global backpropagation-based methods. This work therefore contributes to providing an effective and efficient local learning method for deep SNNs, which could greatly benefit the developments of distributed neuromorphic computing.