检索结果-内蒙古大学图书馆

An Energy Efficient Residual Spiking Neural Network Accelerator With Ternary Spikes

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGrateD CIRCUITS AND SYSTEMS 2025年第1期44卷 395-400页

作者： Sun, Congyi Song, Wenqing Chen, Qinyu Dai, Chenyang Fu, Yuxiang Li, Li Nanjing Univ Sch Elect Sci & Engn Nanjing 210000 Peoples R China Leiden Univ Leiden Inst Adv Comp Sci NL-2300 RA Leiden Netherlands Nanjing Univ Sch Integrated Circuits Nanjing 210000 Peoples R China

Spiking neural networks (SNNs) use discrete binary spikes to transfer information between neurons, which is different from artificial neural networks (ANNs). Although event-based characteristics bring potential computation power and efficiency to SNNs, the long processing time window of discrete spikes leads to high latency. In this brief, a spike version of the residual network using ternary spikes is proposed. A shorter time window is required to achieve competitive performance because the ability to transfer information of the ternary spikes is strengthened. An SNN accelerator based on the proposed residual network with ternary spikes is designed and implemented with 28 nm CMOS technology, and the core area is 0.63 mm2. The proposed SNN accelerator achieves the classification accuracy of 92.07% on CIFAR-10 dataset with SResNet20 and only 6 time steps. The accelerator achieves 0.39 mJ energy consumption per frame with a throughput of 165.7 FPS when running at 500 MHz.

关键词： Neurons Residual neural networks Encoding Membrane potentials Computer architecture Convolutional codes Accuracy Hardware implementation rate coding residual network spiking neural network (SNN) ternary spikes

来源：评论

学校读者我要写书评

暂无评论

Distinct Cellular Mechanisms underlying rate-coding Principles in Primate Auditory Cortex Revealed by Intracellular Recordings in Awake Marmosets

Distinct Cellular Mechanisms underlying Rate-Coding Principl...

引用

第九届海内外华人神经科学家研讨会

作者： Lixia Gao Xiaoqin Wang Department of Biomedical Engineering Johns Hopkins University School of Medicine

A key computational principle for encoding time-varying signals in auditory and somatosensory cortices of monkeys is the opponent model of rate-coding by two distinct populations of neurons. However, the subthreshold mechanisms that give rise to this computation have not been revealed. Because the rate-coding neurons are only observed in awake conditions, it is especially challenging to probe their underlying cellular mechanisms. By using a novel intracellular recording technique that we developed in awake marmosets, we found that the two types of rate-coding neurons in auditory cortex exhibited distinct subthreshold responses. While the positive-monotonic neurons(monotonically increasing firing rate with increasing stimulus repetition frequency) displayed sustained depolarization at high repetition frequency, the negative-monotonic neurons(opposite trend) exhibited instead hyperpolarization at high repetition frequency but sustained depolarization at low repetition frequency. The combination of excitatory and inhibitory subthreshold events allows the cortex to represent time-varying signals through these two opponent neuronal populations.

关键词： temporal processing rate coding in vivo intracellular recording auditory cortex marmoset

来源：评论

学校读者我要写书评

暂无评论

Distinct Cellular Mechanisms underlying rate-coding Principles in Primate Auditory Cortex Revealed by Intracellular Recordings in Awake Marmosets

Distinct Cellular Mechanisms underlying Rate-Coding Principl...

引用

第九届海内外华人神经科学家研讨会(The 9th Symposium for Chinese Neuroscientists Worldwide)

作者： LixiaGao Xiaoqin Wang Department of Biomedical Engineering Johns Hopkins University School of MedicineBaltimoreMD 21025USA

A key computational principle for encoding time-varying signals in auditory and somatosensory cortices of monkeys is the opponent model of rate-coding by two distinct populations of ***,the subthreshold mechanisms that give rise to this computation have not been *** the rate-coding neurons are only observed in awake conditions,it is especially challenging to probe their underlying cellular *** using a novel intracellular recording technique that we developed in awake marmosets,we found that the two types of rate-coding neurons in auditory cortex exhibited distinct subthreshold *** the positive-monotonic neurons(monotonically increasing firing rate with increasing stimulus repetition frequency)displayed sustained depolarization at high repetition frequency,the negative-monotonic neurons(opposite trend)exhibited instead hyperpolarization at high repetition frequency but sustained depolarization at low repetition *** combination of excitatory and inhibitory subthreshold events allows the cortex to represent time-varying signals through these two opponent neuronal populations.

关键词： temporal processing rate coding in vivo intracellular recording auditory cortex marmoset

来源：评论

学校读者我要写书评

暂无评论

Self-Rectifying Dynamic Memristor Circuits for Periodic LIF Refractory Period Emulation and TTFS/rate Signal Encoding

引用

SMALL 2025年第15期21卷 e2408233页

作者： You, Song-Xian Hong, Sheng-Jie Chen, Kuan-Ting Shih, Li-Chung Chen, Jen-Sue Natl Cheng Kung Univ Dept Mat Sci & Engn Tainan Taiwan Natl Cheng Kung Univ Acad Innovat Semicond & Sustainable Mfg Tainan Taiwan

Spiking Neural Networks (SNNs) have gained attention due to their potential to improve computational efficiency compared to traditional artificial neural networks. This study investigates the use of dynamic memristors Ta/IGZO/TaOx/Pt combined with peripheral circuits to emulate the leaky integrate-and-fire behavior of neurons, focusing on incorporating a refractory period. The refractory period is crucial as it prevents neurons from becoming overactive and ensures precise timing in signal processing. This improvement allows the memristor to mimic biological neuron behavior more accurately. The memristor's transient resistance exhibits nonlinear I-V hysteresis and changes in response to pulses, enabling functions of integration, leakage, and firing. Additionally, the memristor is configured as an encoder, converting external signals into voltage pulse sequences. Using coding methods, including rate coding and time-to-first-spike (TTFS) coding, the encoder demonstrates improved signal processing, with TTFS occurring within 21 to 62 ms and encoder frequencies from 2500 to 9500 Hz. Experimental results show that this approach enhances SNN performance, making it more suitable for real-time applications and complex temporal signal processing tasks. This research highlights the potential of dynamic memristors to bridge the gap between neurons and artificial neurons, paving the way for more efficient neuromorphic computing systems.

关键词： dynamic memristors encoders leaky integrate-and-fire (LIF) rate coding refractory period time-to-first-spike (TTFS)

来源：评论

学校读者我要写书评

暂无评论

Timing is everything: The effect of early-life seizures on developing neuronal circuits subserving spatial memory

引用

EPILEPSIA OPEN 2025年第0期 2025 Mar 20页

作者： Holmes, Gregory L. Univ Vermont Coll Med Dept Neurol Sci 95 Carrigan DrStafford 118 Burlington VT 05405 USA

Spatial memory, the aspect of memory involving encoding and retrieval of information regarding one's environment and spatial orientation, is a complex biological function incorporating multiple neuronal networks. Hippocampus-dependent spatial memory is not innate and emerges during development in both humans and rodents. For spatial memory to occur, the hippocampus forms highly associative networks integrating external inputs conveying multi-sensory, proprioceptive, contextual, and emotional information onto internally generated dynamics. Hippocampal cognitive maps are produced by sequences of transient ordered neuronal activations that represent not only spatial information but also the temporal order of events in a memory episode. This patterned activity fine-tunes synaptic connectivity of the network and drives the emergence of specific firing necessary for spatial memory. In the rodent hippocampus, there is a sequence of spontaneous activities that are precisely timed, starting with early sharp waves progressing to theta and gamma oscillations, place and grid cell firing, and sharp wave-ripples that must occur for spatial memory to develop. Whereas normal activity patterns are required for circuit maturation, aberrant neuronal activity during development can have major adverse consequences, disrupting the development of spatial memory. Seizures during infancy, involving massive bursts of synchronized network activity, result in impaired spatial memory when animals are tested as adolescents or adults. This impaired spatial memory is accompanied by alterations in theta and gamma oscillations and spatial and temporal coding of place cells. Conversely, enhancement of oscillatory activity following early-life seizures can improve cognitive impairment. The plasticity of developing oscillatory activity in the immature brain provides exciting opportunities for therapeutic intervention in childhood *** Language SummaryChildren with epilepsy often struggle with

关键词： circuit cognition development memory network oscillations place cells rate coding temporal coding

来源：评论

学校读者我要写书评

暂无评论

Motor unit discharge behavior in human muscles throughout force gradation: a systematic review and meta-analysis with meta-regression

引用

JOURNAL OF APPLIED PHYSIOLOGY 2025年第4期138卷 1050-1065页

作者： Inglis, J. Greig Cabral, Helio V. Cosentino, Caterina Bonardi, Alberto Negro, Francesco Univ Brescia Dept Clin & Expt Sci Brescia Italy

The analysis of motor unit (MU) discharge behavior provides an effective way of assembling information about the generation and control of movement. In this systematic review and meta-analysis, we identified and summarized the literature investigating MU discharge rate and discharge rate variability (CoV-ISI) during voluntary isometric contractions at various force levels. Databases were searched up to January 7, 2025, and a total of 262 studies were included. The meta-means of MU discharge rate and CoV-ISI were estimated and compared across human muscles. The influence of contraction intensity on MU discharge behavior was assessed through linear meta-regressions. At low-to-moderate forces [<60% maximal voluntary contraction (MVC)], the first dorsal interosseous, biceps brachii (BB), and forearm extensors (FEs) had the highest discharge rate, whereas the soleus had the lowest. At high force levels (>60% MVC), the tibialis anterior (TA) had the highest mean discharge rate compared with all other muscles, with the soleus maintaining the lowest. Regarding CoV-ISI results at low forces (<30% MVC), the TA had the lowest CoV-ISI values, except in comparison with the vastii. In addition, the vastii had lower CoV-ISI values than the FE, gastrocnemius medialis, and soleus. Contraction intensity was positively associated with the mean discharge rates in all muscles investigated, although some muscles showed steeper slopes than others. Similar results were observed for CoV-ISI meta-regressions, with muscle-specific differences in slope. These findings suggest potential variations in neural control strategies across muscles during force gradation, such as differences in the relative contribution of rate coding to facilitate increasing force demands.

关键词： coefficient of variation of the interspike interval HDsEMG intramuscular electromyography neural control rate coding

来源：评论

学校读者我要写书评

暂无评论

SAR-TinySNN: A Lightweight Spiking Neural Network for SAR Target Recognition

引用

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 2025年 22卷

作者： Wang, Junyu Sun, Hao Sun, Yuli Tang, Tao Lei, Lin Ji, Kefeng Natl Univ Def Technol State Key Lab Complex Electromagnet Environm Effec Changsha 410073 Peoples R China Natl Univ Def Technol Coll Elect Sci & Technol Changsha 410073 Peoples R China

Spiking neural networks (SNNs) are the third generation of neural networks that offer the advantages of low computational requirements, fast inference speed, and strong biological interpretability. This makes SNNs suitable for synthetic aperture radar (SAR) target recognition tasks, which are often constrained by limited computational power. This letter proposes SAR-TinySNN, a lightweight SNN architecture designed for SAR target recognition. Unlike existing SAR-related studies that predominantly rely on rate coding, SAR-TinySNN uses direct coding to encode SAR images, allowing for a more efficient coding method adapted to SAR images and achieving high target recognition accuracy, especially in scenarios with limited training samples. By integrating direct coding into a trainable SNN framework, SAR-TinySNN achieves competitive performance compared with traditional deep neural networks (DNNs) and deep SNNs on vehicle, aircraft, and ship SAR target recognition datasets, with faster inference times. The experimental results demonstrate the effectiveness of SAR-TinySNN for SAR target recognition.

关键词： Encoding Image coding Neurons Target recognition Training Radar polarimetry Synthetic aperture radar Computer architecture Computational modeling Membrane potentials Direct coding rate coding spiking neural network (SNN) synthetic aperture radar (SAR) target recognition

来源：评论

学校读者我要写书评

暂无评论

ASIC Implementation of Efficient 512-Neuron 256K-Synapses Digital Neuromorphic Processor with On-Chip Encoding in 22nmFDX

ASIC Implementation of Efficient 512-Neuron 256K-Synapses Di...

引用

IEEE International Symposium on Circuits and Systems (ISCAS)

作者： Ghonem, Ahmed Zaky Tawfik, Eslam Yahya Ohio State Univ Columbus OH 43210 USA

ISBN: (纸本)9798350330991;9798350331004

There is a rising demand for AI workloads running on the edge that are increasingly complex, calling for more efficient computing platforms. Neuromorphic hardware is a class of electronic circuits and devices that aim to mimic a biological brain's functionality and computational efficiency. In this work, we propose a 512 neuron 256K synapses LIF-based digital neuromorphic processor. Each neurosynaptic core can be configured to act as either an encoding neuron or as a processing neuron, chosen to be Leaky-Integrate-and-Fire (LIF). We have also implemented a 1-1 software code for the processor, in which Pytorch and SNNtorch are used to build the spiking neural network, and surrogate gradients are used for training. An image classification task is used to evaluate the performance of the neuromorphic processor using the MNIST dataset, achieving an accuracy of 96.12% for 4-bit quantized synaptic weights. Implemented using GF22nmFDX technology, the processor is 0.72mm(2) in area, of which synaptic memories occupy 68% of the area, and achieves an average power efficiency of 14.8 nJ/SOP. This large area is justified by the configuration flexibility of the processor, where each neuron can be connected to any other. This allows the mapping of different and more sophisticated tasks.

关键词： Leaky-Integrate-and-Fire Neuromorphic Computing rate coding Spiking Neural Networks (SNNs)

来源：评论

学校读者我要写书评

暂无评论

Spiking Neural Networks Trained via Proxy

引用

IEEE ACCESS 2022年 10卷 70769-70778页

作者： Kheradpisheh, Saeed Reza Mirsadeghi, Maryam Masquelier, Timothee Shahid Beheshti Univ Fac Math Sci Dept Comp & Data Sci Tehran *** Iran Amirkabir Univ Technol Dept Elect Engn Tehran 158754413 Iran Univ Toulouse 3 Ctr Natl Rech Sci CerCo UMR 5549 F-31062 Toulouse France

We propose a new learning algorithm to train spiking neural networks (SNN) using conventional artificial neural networks (ANN) as proxy. We couple two SNN and ANN networks, respectively, made of integrate-and-fire (IF) and ReLU neurons with the same network architectures and shared synaptic weights. The forward passes of the two networks are totally independent. By assuming IF neuron with rate-coding as an approximation of ReLU, we backpropagate the error of the SNN in the proxy ANN to update the shared weights, simply by replacing the ANN final output with that of the SNN. We applied the proposed proxy learning to deep convolutional SNNs and evaluated it on two benchmarked datasets of Fashion-MNIST and Cifar10 with 94.56% and 93.11% classification accuracy, respectively. The proposed networks could outperform other deep SNNs trained with tandem learning, surrogate gradient learning, or converted from deep ANNs. Converted SNNs require long simulation times to reach reasonable accuracies while our proxy learning leads to efficient SNNs with much smaller simulation times. The source codes of the proposed method are publicly available at https://***/SRKH/ProxyLearning.

关键词： Neurons Artificial neural networks Encoding Backpropagation Biological neural networks Learning (artificial intelligence) Firing Spiking neural networks supervised learning proxy learning rate coding

来源：评论

学校读者我要写书评

暂无评论

Capacity per Unit Cost-Achieving Input Distribution of rated-Inverse Gaussian Biological Neuron

引用

IEEE TRANSACTIONS ON COMMUNICATIONS 2022年第6期70卷 3788-3803页

作者： Nasiraee, Mohammad Montazery Kordy, Hussain Kazemitabar, Javad Babol Noshirvani Univ Technol Fac Elect & Comp Engn Babol *** Iran

Maximizing the information per the expected unit energy cost in biological neurons is vital in understanding the functionality of our nervous system and improving the bio-inspired nano-networks. Hence, toward a better comprehension of neuronal information processing and communication from an information-energy standpoint, this paper presents the following novel results on the capacity-achieving distribution: 1) For the first time, the paper presents the probability mass function of the number of spikes observed in a given time window where inter-spike intervals follow the Inverse Gaussian distribution. To extend this result, the first and second moments of the underlying probability mass function will be determined. 2) The paper proves that the relationship between the underlying probability mass function and the marginal distribution of the intensity parameter is one-to-one when the parameter is a random variable. 3) Proof that a unique capacity-achieving distribution for the stimulus intensity exists in a discrete form with a finite number of mass points that varies with the energy budget. Then, the paper presents an algorithm that numerically computes a trade-off between the maximum transmit bits and the coding-duration energy expenditure. Subsequently, the optimal multiplier along with the optimal distribution and the corresponding capacity will be obtained.

关键词： Neurons Encoding Axons Random variables Information processing Synapses Nanobioscience Information theory nano-scale communications rate coding inverse Gaussian distribution count distribution

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：