有哪些信誉好的足球投注网站Competition Between Feed-Forward and Lateral Information Processing in Layered Neural Netwo.pdf
- 1、本文档共8页，可阅读全部内容。
- 2、有哪些信誉好的足球投注网站（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
- 3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
查看更多
Competition Between Feed-Forward and Lateral Information Processing in Layered Neural Netwo
Competition Between Feed-Forward and LateralInformation Processing in Layered NeuralNetworksA.C.C. Coolen1 and L. Viana21 Department of Mathematics, Kings College LondonStrand, London WC2R 2LS, U.K.2 Laboratorio de Ensenada, Instituto de Fisica, UNAMA. Postal 2681, Ensenada B.C., MexicoAbstract. We use replica theory to analyse layered networks of binaryneurons, with lateral Hebbian-type synapses within individual layers, incombination with strictly feed-forward Hebbian-type synapses betweensuccessive layers. The competition between two qualitatively di erentmodes of operation, feed-forward versus lateral, induces interesting tran-sitions, and allows for the identi cation of an optimal value for the bal-ance between the two synapse types.1 IntroductionThe operation of networks with purely feed-foward interactions di ers fundamen-tally from that of networks with recurrent interactions. In feed-forward networksthe nal state depends only on stimuli from previous layers, not on the initialstate. The nal state of recurrent networks, on the other hand, often dependsstrongly on the initial state. Both have speci c advantages and disadvantages,and it seems sensible to try to have the best of both worlds by building net-works with feed-forward and lateral interactions. Feed-forward interactions pro-vide information transport, but do not enforce temporal continuity; inputs areresponded to strictly, even when wildly changing over time. Lateral interactionsintroduce a dependence on the systems history, and thereby enforce tempo-ral continuity. The optimal balance between feed-forward and lateral processingappears to be the onset of ergodicity breaking, where temporal continuity ismaximised under the contraint that no attractors have yet been created. Herewe present and solve a model in which this competition between feed-forwardand lateral processing can be studied analytically.2 Model De nitionsWe study feedforward chains of L recurrent layers, with N binary neurons each.Th
您可能关注的文档
- characteristics of stemflow for tall stewartia growing on a hillslope.pdf
- Characteristics and sources of chromophoric dissolved organic matter in lakes of the Yungui Plateau.pdf
- Characterization of Electrochemically Grown Crystalline CuInSe2 Thin Films.pdf
- Characterization of PM2.5 in the ambientair of Shanghaicity by analyzingindividualparticles.pdf
- Characterizations of essential ideals as operator modules over C-algebras.pdf
- Characterizing packet audio streams from internet multimedia applications.pdf
- Charge amplification concepts for direction-sensitive dark matter detectors.pdf
- Charge carriers of different origin in cuprates as revealed by experiment.pdf
- Charge collective modes in an incommensurately modulated cuprate.pdf
- Charge exchange for medium energy He and Ne ions in a large-angle collision at solid surfaces.pdf
文档评论（0）