基于自适应环吸引子网络的基础数值计算模型
打开文本图片集
关键词：头部方向细胞（HDC）；自适应神经网络；Hebbian学习；生物启发计算；神经形态计算中图分类号：TP183 文献标志码：A 文章编号：1001-3695（2025）12-013-3628-09doi：10.19734/j.issn.1001-3695.2025.05.0166
Fundamental numerical computational modeling based on adaptive loop attractor network
Chen Xinyu'，Chen Zugang2+ （1.SchoolofuterecedrifalItellgence，ZenoUniersitZegzou4oin；erospaceo search Institute，Chinese Academy of Sciences，Beijing 1Ooo94，China）
Abstract：Thisarticleproposedabio-inspiredadaptiveHDCnetwork model.Itmimicked theoperationlawof mammalian brainheaddirectioncels，ndrealizedtheautonomous leaninganderorcorrctionfunctionsof thenetworkbyusingself-adjusting weightsand Hebianlearning law.Theexperimentalresultsfoundthatin terms ofautonomous learning，the network weightscouldbeadaptivelyadjustedbasedonthecomputationaleror，andtheaverageconvergencetimewas2.5iterations， which improved the learning efficiency by 60% compared with the fixed-weight method. In terms of error correction，the system was able to achieve an error correction rate of 94% ，and the final error was stably controlled to within 3% . The ring topologymodelwas formed withthehelpofLIFneurons，whichrealizedexcitatoryand inhibitorysignallinkagestateregulation ability，andafterallkindsofneurologicalarithmetictaskswereprocessedthroughit.Comparedwiththeraditionalfixedweight method，the convergence time of adaptive learning was shortened by 60% ，and the system showed an average of 35% reductionoferrorrate，whetheritwasintheprocessingofresultsbeyondtherangeoftheoverflowofthe[O，99]，orinthe procesing ofthedifferent spansof theadditiveandsubtractivearithmetictasks.Thisstudyprovidesanewresearchidea for thedevelopmentofbio-inspiredintellgentcomputation，whichisanimportantreferencefortheadvancementoftificialieligence induced by brain science and has the possibility of popularization and application.
Key words：headdirectioncell（HDC）；adaptive neural network；Hebbianlearning；bio-inspiredcomputation；neuromorphic computation
0 引言
生物启发式神经计算已经应用到边缘计算、智能机器人、神经形态芯片等领域[1]，潜移默化地影响着计算技术的发展方向。（剩余21860字）
