Switching plasticity in compensated ferrimagnetic multilayers for neuromorphic computing

Current-induced multilevel magnetization switching in ferrimagnetic spintronic devices is highly pursued for the application in neuromorphic computing.In this work,we demonstrate the switching plasticity in Co/Gd ferrimagnetic multilayers where the binary states magnetization switching induced by spin–orbit toque can be tuned into a multistate one as decreasing the domain nucleation barrier.Therefore,the switching plasticity can be tuned by the perpendicular magnetic anisotropy of the multilayers and the in-plane magnetic field.Moreover,we used the switching plasticity of Co/Gd multilayers for demonstrating spike timing-dependent plasticity and sigmoid-like activation behavior.This work gives useful guidance to design multilevel spintronic devices which could be applied in high-performance neuromorphic computing.

Plasticity-induced characteristic changes of pattern dynamics and the related phase transitions in small-world neuronal networks

Phase transitions widely exist in nature and occur when some control parameters are changed. In neural systems, their macroscopic states are represented by the activity states of neuron populations, and phase transitions between different activity states are closely related to corresponding functions in the brain. In particular, phase transitions to some rhythmic synchronous firing states play significant roles on diverse brain functions and disfunctions, such as encoding rhythmical external stimuli, epileptic seizure, etc. However, in previous studies, phase transitions in neuronal networks are almost driven by network parameters （e.g., external stimuli）, and there has been no investigation about the transitions between typical activity states of neuronal networks in a self-organized way by applying plastic connection weights. In this paper, we discuss phase transitions in electrically coupled and lattice-based small-world neuronal networks （LBSW networks） under spike-timing-dependent plasticity （STDP）. By applying STDP on all electrical synapses, various known and novel phase transitions could emerge in LBSW networks, particularly, the phenomenon of self-organized phase transitions （SOPTs）： repeated transitions between synchronous and asynchronous firing states. We further explore the mechanics generating SOPTs on the basis of synaptic weight dynamics.

The first factor affecting dryland winter wheat grain yield under various mulching measures: Spike number

Water is the key factor limiting dryland wheat grain yield.Mulching affects crop yield and yield components by affecting soil moisture.Further research is needed to determine the relationships between yield components and soil moisture with yield,and to identify the most important factor affecting grain yield under various mulching measures.A long-term 9-yearifeld experiment in the Loess Plateau of Northwest China was carried out with three treatments:no mulch (CK),plastic mulch (M_(P)) and straw mulch (M_(S)).Yield factors and soil moisture were measured,and the relationships between them were explored by correlation analysis,structural equation modeling and significance analysis.The results showed that compared with CK,the average grain yields of M_(P) and M_(S) increased by 13.0and 10.6%,respectively.The average annual grain yield of the M_(P) treatment was 134 kg ha^(–1) higher than the M_(S) treatment.There were no significant differences in yield components among the three treatments (P<0.05).Soil water storage of the M_(S) treatment was greater than the M_(P) treatment,although the differences were not statistically signifiant.Soil water storage during the summer fallow period (SWSSF) and soil water storage before sowing (SWSS) of M_(S) were significantly higher than in CK,which increased by 38.5 and 13.6%,respectively.The relationship between M_(P) and CK was not statistically significant for SWSSF,but the SWSS in M_(P) was significantly higher than in CK.In terms of soil water storage after harvest (SWSH) and water consumption in the growth period(ET),there were no signi?cant differences among the three treatments.Based on the three analysis methods,we found that spike number and ET were positively correlated with grain yield.However,the relative importance of spike number to yield was the greatest in the M_(P )and M_(S) treatments,while that of ET was the greatest in CK.Suifcient SWSSF could indirectly increase spike number and ET in the three treatments.Based on these results,mulch can improve yield and soil water storage.The most important factor affecting the grain yield of dryland wheat was spike number under mulching,and ET with CK.These findings may help us to understand the main factors influencing dryland wheat grain yield under mulching conditions compared to CK.

Self-sustained firing activities of the cortical network with plastic rules in weak AC electrical fields

Both external and endogenous electrical fields widely exist in the environment of cortical neurons. The effects of a weak alternating current （AC） field on a neural network model with synaptic plasticity are studied. It is found that self-sustained rhythmic firing patterns, which are closely correlated with the cognitive functions, are significantly modified due to the self-organizing of the network in the weak AC field. The activities of the neural networks are affected by the synaptic connection strength, the exterrtal stimuli, and so on. In the presence of learning rules, the synaptic connections can be modulated by the external stimuli, which will further enhance the sensitivity of the network to the external signal. The properties of the external AC stimuli can serve as control parameters in modulating the evolution of the neural network.

Complementary memtransistors for neuromorphic computing: How, what and why

Memtransistors in which the source-drain channel conductance can be nonvolatilely manipulated through the gate signals have emerged as promising components for implementing neuromorphic computing.On the other side,it is known that the complementary metal-oxide-semiconductor(CMOS)field effect transistors have played the fundamental role in the modern integrated circuit technology.Therefore,will complementary memtransistors(CMT)also play such a role in the future neuromorphic circuits and chips?In this review,various types of materials and physical mechanisms for constructing CMT(how)are inspected with their merits and need-to-address challenges discussed.Then the unique properties(what)and poten-tial applications of CMT in different learning algorithms/scenarios of spiking neural networks(why)are reviewed,including super-vised rule,reinforcement one,dynamic vision with in-sensor computing,etc.Through exploiting the complementary structure-related novel functions,significant reduction of hardware consuming,enhancement of energy/efficiency ratio and other advan-tages have been gained,illustrating the alluring prospect of design technology co-optimization(DTCO)of CMT towards neuro-morphic computing.

基于多子网络预训练的脉冲神经网络分类模型

脉冲神经网络(Spiking Neural Network,SNN)被认为是最符合生物大脑机制的类脑计算模型,凭借其事件驱动、高能效、可解释等特点吸引了越来越多的研究关注。然而,由于脉冲的二值输出与不可微分性,SNN的训练方法仍存在一定空缺。于是借鉴皮层记忆单元通过局部网络存储记忆信息的方式,提出一种基于多子网络预训练的脉冲神经网络分类方法。该方法使用样本标签信息优化了脉冲序列特征提取过程,采用改进的脉冲时间依赖可塑性学习规则预训练多个单类别特征提取子网络,并将预训练后的子网络进行无监督特征融合,有效提高了网络的特征分类能力。此外,在权重可视化与t-SNE可视化工具的帮助下,分析了方法的有效性。所提方法在MNIST与Fashion-MNIST数据集上分别取得了97.40%与88.81%的分类准确度。

基于双阈值的ANN-SNN转换方法优化

脉冲神经网络作为第三代神经网络,能够克服许多人工神经网络中所存在的问题,如高功耗、鲁棒性较差等;通过对预训练好的人工神经网络模型进行转换是获取深度脉冲神经网络模型的一种主要方法,然而通过这种方法获取的脉冲神经网络的延迟较高,无法满足实时性要求;文章在双阈值转换方法的基础上,采用阈值平衡技术对转换过程进行优化,通过理论推导,提出了一种对称阈值LeakyReLU激活函数,并对人工神经网络到脉冲神经网络的转换流程进行了梳理;此外,采用了泄漏机制对转换后的脉冲神经网络模型结构进行了优化,并通过脉冲时序依赖可塑性学习规则对该结构进行训练;最终,在MNIST数据集与CIFAR-10数据集上进行了实验,结果表明,优化后脉冲神经网络的收敛速度与鲁棒性得到了大幅提升。

The Lightweight Edge-Side Fault Diagnosis Approach Based on Spiking Neural Network

Network fault diagnosis methods play a vital role in maintaining network service quality and enhancing user experience as an integral component of intelligent network management.Considering the unique characteristics of edge networks,such as limited resources,complex network faults,and the need for high real-time performance,enhancing and optimizing existing network fault diagnosis methods is necessary.Therefore,this paper proposes the lightweight edge-side fault diagnosis approach based on a spiking neural network(LSNN).Firstly,we use the Izhikevich neurons model to replace the Leaky Integrate and Fire(LIF)neurons model in the LSNN model.Izhikevich neurons inherit the simplicity of LIF neurons but also possess richer behavioral characteristics and flexibility to handle diverse data inputs.Inspired by Fast Spiking Interneurons(FSIs)with a high-frequency firing pattern,we use the parameters of FSIs.Secondly,inspired by the connection mode based on spiking dynamics in the basal ganglia(BG)area of the brain,we propose the pruning approach based on the FSIs of the BG in LSNN to improve computational efficiency and reduce the demand for computing resources and energy consumption.Furthermore,we propose a multiple iterative Dynamic Spike Timing Dependent Plasticity(DSTDP)algorithm to enhance the accuracy of the LSNN model.Experiments on two server fault datasets demonstrate significant precision,recall,and F1 improvements across three diagnosis dimensions.Simultaneously,lightweight indicators such as Params and FLOPs significantly reduced,showcasing the LSNN’s advanced performance and model efficiency.To conclude,experiment results on a pair of datasets indicate that the LSNN model surpasses traditional models and achieves cutting-edge outcomes in network fault diagnosis tasks.

脉冲神经网络的监督学习算法研究综述

脉冲神经网络是进行复杂时空信息处理的有效工具,但由于其内在的不连续和非线性机制,构建高效的脉冲神经网络监督学习算法非常困难,同时也是该研究领域的重要问题.本文介绍了脉冲神经网络监督学习算法的基本框架,以及性能评价原则,包括脉冲序列学习能力、离线与在线处理性能、学习规则的局部特性和对神经网络结构的适用性.此外,对脉冲神经网络监督学习算法的梯度下降学习规则、突触可塑性学习规则和脉冲序列卷积学习规则进行了详细的讨论,通过对比分析指出现有算法存在的优缺点,并展望了该领域未来的研究方向.

Effects of paired associative magnetic stimulation between nerve root and cortex on motor function of lower limbs after spinal cord injury:study protocol for a randomized controlled trial

Classic paired associative stimulation can improve synaptic plasticity,as demonstrated by animal expe riments and human clinical trials in spinal cord injury patients.Paired associative magnetic stimulation(dual-target peripheral and central magnetic stimulation)has been shown to promote neurologic recove ry after stroke.However,it remains unclear whether paired associative magnetic stimulation can promote recovery of lower limb motor dysfunction after spinal cord injury.We hypothesize that the curre nt caused by central and peripheral magnetic stimulation will conve rge at the synapse,which will promote synapse function and improve the motor function of the relevant muscles.Therefore,this study aimed to examine the effects of paired associative magnetic stimulation on neural circuit activation by measuring changes in motor evoked and somatosensory evoked potentials,motor and sensory function of the lower limbs,functional health and activities of daily living,and depression in patients with spinal co rd injury.We will recruit 110 thora cic spinal trauma patients treated in the Department of Spinal Cord Injury,China Rehabilitation Hospital and randomly assign them to expe rimental and control groups in a 1:1 ratio.The trial group(n=55)will be treated with paired associative magnetic stimulation and conventional rehabilitation treatment.The control group(n=55)will be treated with sham stimulation and co nventional rehabilitation treatment.Outcomes will be measured at four time points:baseline and 4,12,and 24 wee ks after the start of inte rvention(active or sham paired associative magnetic stimulation).The primary outcome measure of this trial is change in lower limb American Spinal Injury Association Impairment Scale motor function score from baseline to last follow-up.Secondary outcome measures include changes in lower limb American Spinal Injury Association sensory function sco re,motor evoked potentials,sensory evoked potentials,modified Ashwo rth scale score,Maslach Burnout Invento ry score,and Hamilton Depression Scale score over time.Motor evoked potential latency reflects corticospinal tract transmission time,while amplitude reflects recruitment ability;both measures can help elucidate the mechanism underlying the effect of paired associative magnetic stimulation on synaptic efficiency.Adve rse events will be recorded.Findings from this trial will help to indicate whether paired associative magnetic stimulation(1)promotes recove ry of lower limb sensory and motor function,reduces spasticity,and improves quality of life;(2)promotes neurologic recovery by increasing excitability of spinal cord motor neurons and stimulating synaptic plasticity;and(3)improves rehabilitation outcome in patients with spinal cord injury.Recruitment for this trial began in April 2021 and is currently ongoing.It was approved by the Ethics Committee of Yangzhi Affiliated Rehabilitation Hospital of Tongji University,China(approval No.YZ2020-018)on May 18,2020.The study protocol was registered in the Chinese Clinical Trial Registry(registration number:ChiCTR2100044794)on March 27,2021(protocol version 1.0).This trial will be completed in April 2022.

脉冲神经网络研究现状及展望

脉冲神经网络(Spiking Neural Network,SNN)包含具有时序动力学特性的神经元节点、稳态-可塑性平衡的突触结构、功能特异性的网络环路等,高度借鉴了生物启发的局部非监督(如脉冲时序依赖可塑性、短时突触可塑性、局部稳态调节等)、全局弱监督(如多巴胺奖赏学习、基于能量的函数优化等)的生物优化方法,因此具有强大的时空信息表征、异步事件信息处理、网络自组织学习等能力.SNN的研究属于交叉学科,将深入融合脑科学和计算机科学,因此对其研究也可以主要分为两大类:一类是以更好地理解生物系统为最终目的;另一类是以追求卓越计算性能为优化目标.本文首先对当前这两大类SNN的研究进展、研究特点等进行分析,重点介绍基于Spike的多类异步信息编码、基于Motif分布的多亚型复杂网络结构、多层时钟网络自组织计算、神经形态计算芯片的软硬结合等.同时,介绍一种融合生物多尺度、多类型神经可塑性的高效SNN优化策略,使得SNN中的信度分配可以从宏观尺度有效覆盖到微观尺度,如全部的网络输出、网络隐层状态、局部的各个神经节点等,并部分解答生物系统是如何通过局部参数的调优而实现全局网络优化的问题.这将不仅为现有人工智能模型提高其认知能力指明一种可能的生物类优化方向,还为反向促进生命科学中生物神经网络的可塑性研究新发现提供启发.本文认为,脉冲神经网络的发展目标不是构建人工神经网络的生物版本替代品,而是通过突破生物启发的多尺度可塑性优化理论,去粗取精,最终实现具有生物认知计算特色的新一代高效脉冲神经网络模型,使其有望获得更快的学习速度、更小的能量消耗、更强的适应性和更好的可解释性等.

基于突触可塑性的自适应脉冲神经网络在高斯白噪声刺激下的抗扰功能研究（英文）

电磁环境的复杂多变使得传统电磁抗扰方式的不足日益凸显。生物体在神经系统调节下具有自组织、自适应和抗扰性等优势。因此,探索一种借鉴生物体优势的新的电磁抗扰方式具有重要意义,电磁仿生防护应运而生。该文基于电磁仿生防护的理念开展了基于兴奋性和抑制性突触可塑性共同调控的脉冲神经网络的抗扰功能的研究。构建了十层前馈脉冲神经网络,对高斯白噪声干扰下的脉冲神经网络进行仿真实验,重点研究了高斯白噪声对输出层神经元放电率和膜电位间相关性的影响。实验结果表明,一定强度范围的高斯白噪声干扰,对输出层神经元的放电率影响较小,输出层神经元放电率的相对变化率较小;对输出层神经元膜电位的影响较小,膜电位间的相关性较高。即在突触可塑性机制的调控下,脉冲神经网络具有一定的抗噪声干扰的能力。该研究成果为提高电子系统在复杂电磁环境下的防护能力奠定了理论基础。

θ相移在单次学习过程中促进神经网络对空间位置顺序记忆的研究

θ相移是在大鼠海马中发现的位置细胞放电的特殊模式。随着大鼠在某个位置场中行进,相应位置细胞发放脉冲的相位(相对于局部电位中的θ节律)会逐渐提前。一些学者认为,该现象可以将大鼠在运动中所经过的一系列位置场的顺序编码成时间上压缩,并且多次重复出现的脉冲模式,因此可以促进大鼠对其在运动中经过的空间位置的顺序的记忆。本文建立了一个模型,对该现象进行了研究。首先,本文建立了能够产生θ相移现象的单个海马神经元模型。这一模型建立在HarrisKD等及MageeJC的电生理实验研究的基础上,根据神经元真实的生理特性来建模。并且以整合与发放的脉冲神经元模型取代H-H模型,大大简化了计算量。而模拟结果又能较好的重现实验中真实神经元的表现。为了研究θ相移对空间位置顺序记忆的作用,在单神经元模型的基础上,又建立了一个基于STDP的学习型神经网络。通过对网络的研究发现,空间位置顺序的信息在模拟中只要输入一次,就可以使该网络对这一顺序形成一定程度的记忆,并且有一定的比率能达到很高的准确率。而如果在单神经元模型中去除θ相移功能,则在单次学习过程中,根本无法形成对空间位置顺序的记忆,代表各个空间位置的神经元几乎同时发放,基本上不能代表顺序信息。

螺纹道钉抗拔力对再生塑性复合轨枕的影响分析

再生塑性复合轨枕是一种新型材料轨枕。为研究螺纹道钉上拔力对轨枕的影响,通过轨枕材料轴向拉伸试验数据得到轨枕材料真实应力应变曲线,道钉使用木枕用螺纹道钉。假设轨枕各向同性,应变与速率无关,道钉与轨枕接触良好,建立螺纹道钉与轨枕1/4有限元模型。由钢轨在横向作用力下的转动计算出上拔力,对道钉施加上拔力,分析在螺纹道钉抗拔过程中轨枕的位移与应力应变状态。结果表明:上拔力将影响整个轨枕宽度范围,螺栓底部以及螺纹端部与轨枕接触区域应力较大,也最容易产生塑性区域;加大上拔力会在螺纹端部形成塑性区域带最终导致破坏。

双层结构突触仿生忆阻器的时空信息传递及稳定性

现有计算机体系架构下的神经网络难以对多任务复杂数据进行高效处理,成为制约人工智能技术发展的瓶颈之一,而人脑的并行运算方式具有高效率、低功耗和存算一体的特点,被视为打破传统冯·诺依曼计算体系最具潜力的运算体系.突触仿生器件是指从硬件层面上实现人脑神经拟态的器件,它可以模拟脑神经对信息的处理方式,即“记忆”和“信息处理”过程在同一硬件上实现,这对于构建新的运算体系具有重要的意义.近年,制备仿生突触器件的忆阻材料已获得进展,但多聚焦于神经突触功能的模拟,对于时空信息感知和传递的关键研究较为缺乏.本文通过制备一种双层结构忆阻器,实现了突触仿生器件的基本功能包括双脉冲易化和抑制、脉冲时间依赖突触可塑性(spiking time dependent plasticity,STDP)和经验式学习等,还对器件的信息感知、传递特性和稳定性进行了研究,发现该器件脉冲测试结果满足神经网络处理时空信息的基本要求,这一结果可以为忆阻器在类脑芯片中的应用提供参考.

抑制性STDP突触机制对皮层网络的调节

为研究抑制性突触的脉冲时间依赖可塑性(spike-timing-dependent-plasticity,STDP)突触机制对大脑皮层网络的调节作用,构建了脑皮层神经网络的局部回路模型。通过模型观察到,在兴奋性与抑制性突触的共同作用下,不同类型突触连接的平均强度均维持稳定,保证了皮层网络自身的平稳放电;随着抑制性STDP突触规则学习率的增大,网络中神经元集群的平均放电率和同步指数均增大,兴奋性突触的整体强度减弱,抑制性突触的整体强度增强;揭示了皮层网络中兴奋性与抑制性的调节过程,有助于认识抑制性突触可塑性在皮层网络功能机制中的重要作用。

抑制性STDP不同学习窗口下的神经元放电特性

为了深入研究不同抑制性突触可塑性机制的功能,在大脑皮层神经网络模型中对Hebbian、anti-Hebbian和Symmetric这3种不同学习窗口的抑制性脉冲时间依赖可塑性(spike timing-dependent plasticity,STDP)突触机制作用下的皮层神经元放电特性进行对比分析。通过分析不同窗口下神经元集群的平均放电频率、同步性、突触权值和突触电流,发现抑制性STDP机制的学习窗口类型会影响该机制对网络神经元放电特性的调节,anti-Hebbian类型学习窗口能根据网络中神经元放电率的变化,自适应调节突触权值以维持神经元放电;Hebbian和Symmetric类型学习窗口对神经元放电的抑制作用较强,不利于皮层神经元的放电。

不同重连概率的小世界脉冲神经网络抗扰功能研究

现今各种电磁干扰对电子系统造成的不良影响越来越严重,传统防护方式的局限性日益凸显。电磁仿生学由此被提出,目的是通过借鉴生物体的自适应抗扰的优良特性,以期建立新的防护模式。构建了以Izhikevich神经元模型为节点,兴奋性和抑制性突触可塑性模型共同调节基于小世界网络拓扑的小世界脉冲神经网络;基于复杂网络理论对比分析了不同重连概率的小世界网络的拓扑特性;对比分析了不同重连概率的小世界脉冲神经网络在高斯白噪声刺激下的抗扰功能。实验结果表明:小世界网络的平均路径长度和全局效率值受重连概率的影响较小,平均聚类系数和小世界属性受重连概率的影响较大;构建的不同重连概率的脉冲神经网络均具有一定抗扰功能且高聚类系数和低平均路径长度显著的小世界脉冲神经网络抗扰功能最优。

神经系统识别爆发性锋电位序列的机制讨论

讨论神经系统识别爆发型锋电位序列信息的机制,认为序列的信息存储在爆发锋电位组内间隔和组间间隔2个时间变量中,建立了一个神经回路,通过突触传递过程中的易化、反馈调节机制以及时间依赖的学习机制等突触可塑性机制,给出了神经系统识别爆发性锋电位序列信息的一种可能机制,其中包括分解机制和整合机制两部分.首先通过神经元选择性响应的动力学性质,将锋电位序列的信息分解,并将每组锋电位内部间隔的信息通过不同神经元学习存储.通过突触延迟时间的动力学调整,将2组锋电位之间的时间间隔学习、存储在回路中.经过多次学习训练,神经回路对输入信号形成特定的突触连接结构以及时空响应输出模式,实现对爆发性锋电位序列信息的识别.

多种连接模型的忆阻神经网络学习

忆阻器以其独特的非易失性、天然的记忆功能以及纳米级尺寸,在人工神经网络、信号处理和模式识别等方面展现了巨大的应用前景。采用了基于STDP学习规则的忆阻神经网络,运用了网络自适应突变以及网络拓扑结构变化的基因算法,其中包括隐藏层神经元个数,连接权重以及神经网络突触模型的变化。比较了基于HP线性忆阻器模型,非线性忆阻器模型以及阈值模型这3种不同忆阻器模型的忆阻神经网络,并提出了学习效果更好的混合型忆阻神经网络。