简介：SeveralextensionsofthelogicprogramminglanguagePrologtononHornclausesusecaseanalysistohandlenon-Hornclauses.Inthispaper,analyticalandempriricalevidencesarepresentedtoshowthat,bymakingasetofclausesless'non-Horn'usingpredicatedrenaming.theperformanceofthesecase-analysisbasedprocedurescanbeimprovedsignificantly.Inaddition,thepaperalsoinvestigatedtheproblemofefficeientlyconstructingapredicaterenamingthatreducesthedegreeof'non-Hornness'ofaclausesetmaximally.Itisshownthatthisproblemoffindingaredicaterenamingtoachieveminimal'non-Hornness'isNP-complete.

简介：Recentextensivemeasurementsofreal-lifetrafficdemonstratethattheprobabilitydensityfunctionofthetrafficinnon-Gaussian.Ifatrafficmodeldoesnotcapturethischaracteristics,anyanalyticalorsimulationresultswillnotbeaccurate.Inthiswork,westudytheimpactofnon-Gaussiantrafficonnetworkperformance,andpresentanapproachthatcanaccuratelymodelthemarginaldistributionofreal-lifetraffic.Boththelong-andshort-rangeautocorrelationsarealsoaccounted.Weshowthattheremovalofnon-Gaussiancomponentsoftheprocessdoesnotchangeitscorrelationstructure,andwevalidateourpromisingprocedurebysimulations.

简介：两倍缓冲是有效机制隐藏在在薄片上和离开薄片记忆之间的数据转移的潜伏。在dataflow建筑学，因为dataflow加速器的重复充满并且排干，然而，交换二在许多瓦减少的执行期间缓冲性能。在这个工作，我们为dataflow建筑学建议连续双的缓冲机制。没有停止通过在dataflow建筑学优化控制逻辑处理元素的执行，建议不停的机制把瓦分到处理元素数组。而且，我们建议一个工作流节目与连续双的缓冲机制合作。在控制逻辑上并且在工作流节目上的优化以后，充满并且排干数组需要越过属于一样的dataflow图的所有瓦的执行被做仅仅一次。试验性的结果证明没有优化，为dataflow建筑学的建议双缓冲机制在那上完成16.2%平均效率改进。

简介：数据立方体计算前是为支持OLAP(处理的OnlineAnalytical)的一个重要概念并且广泛地被学习了。由于巨大的存储器需求计算一个完全的数据立方体经常不是可行的。最近建议的商立方体通过组织立方体房间进等价分区的分割法处理了这个问题。如此的一条途径不仅为象和那样的分发的聚合函数是有用的而且能被用于整体的聚合函数象一样的维护中部它将为每个等价班要求一套元组的存储。不幸地，当变化被做到数据来源，自从划分立方体房间必须也被更新，维持商立方体是重要的。在这篇论文，作者设计增量算法为和和中部的聚合函数高效地更新一个商立方体。为聚合函数和，概念从Galois的原则被借开发中央处理器有效的算法更新一个商立方体。为聚合函数中部，一个假班的概念被介绍进一步减少商立方体的尺寸。结合了一种新奇滑动窗口技术，一个有效算法为维持那收起的一个中部的商立方体被开发相当小的存储空间。建议算法在大数据库上有效、可伸缩的性能研究表演。

简介：Inthispaper,aneffectiveandrobustactivespeechdetectionmethodisproposedbasedonthe1/fprocesstechniqueforsignalsundernon-stationarynoisyenvironments.TheGaussian1/fprocess,amathematicalmodelforstatisticallyself-similarradomprocessesbasedonfractals,isselectedtomodelthespeechandthebackgroundnoise.AnoptimalBayesiantwo-classclassifierisdevelopedtodiscriminatethembytheir1/fwaveletcoefficientswithKarhunen-Loeve-typeproperties.Multipletemplatesaretrainedforthespeechsignal,andtheparametersofthebackgroundnoisecanbedynamicallyadaptedinruntimetomodelthevariationofboththespeechandthenoise.Inourexperiments,a10-minutelongspeechwithdifferenttypesofnoisesrangingfrom20dBto5dBistestedusingthisnewdetectionmethod.Ahighperformancewithover90%detectionaccuracyisachievedwhenaverageSNRisabout10dB.

简介：这份报纸建议为非奉献的簇设计的一台预言引擎，它能为平行应用估计重组时间，甚至面对工作站主人的连续工作量。预言引擎能被设置用三个不同评价核工作：一个历史的核，一个模拟核基于分析模型和两个的集成，命名的混合的核。这些评价建议集成于一个安排系统，说出CISNE，它能在一个联机或离线的模式被执行。建议评价方法的精确性在与安排政策在的不同工作的关系被评估一真实并且模仿的簇环境。在两环境，因为和历史的方法的精确性捕获非奉献的环境的精力就资源的状态而言估计应用程序运行时刻联合一台模拟引擎的能力，我们观察到混合系统给最好的结果。

简介：Newnon-volatilememory(NVM)technologiesareexpectedtoreplacemainmemoryDRAM(dynamicrandomaccessmemory)inthenearfuture.NANDflashtechnologicalbreakthroughshaveenabledwideadoptionofsolidstatedrives(SSDs)instoragesystems.However,flash-basedSSDs,bynature,cannotavoidlowenduranceproblemsbecauseeachcellonlyallowsalimitednumberoferasures.ThiscangiverisetocriticalSSDreliabilityissues.SincemanySSDwriteoperationseventuallycausemanySSDeraseoperations,reducingSSDwritetrafficplaysacrucialroleinSSDreliability.ThispaperproposestwoNVM-basedbuffercachepolicieswhichcanworktogetherindifferentlayerstomaximallyreduceSSDwritetraffic:amainmemorybuffercachedesignnamedHierarchicalAdaptiveReplacementCache(H-ARC)andaninternalSSDwritebufferdesignnamedWriteTrafficReductionBuffer(WRB).H-ARCconsidersfourfactors(dirty,clean,recency,andfrequency)toreducewritetrafficandimprovecachehitratiosinthehost.WRBreducesblockerasuresandwritetrafficfurtherinsideanSSDbyeffectivelyexploitingtemporalandspatiallocalities.ThesetwocomprehensiveschemessignificantlyreducetotalSSDwritetrafficateachdifferentlayer(i.e.,hostandSSD)byupto3x.Consequently,theyhelpextendSSDlifespanwithoutsystemperformancedegradation.

简介：Inthispaper,wepresentanovelapproachtosynthesizingfrontalandsemi-frontalcartoon-likefacialcaricaturesfromanimage.Thecaricatureisgeneratedbywarpingtheinputfacefromtheoriginalfeaturepointstothecorrespondingexaggeratedfeaturepoints.A3Dmeanfacemodelisincorporatedtofacilitatefacetocaricaturesbyinferringthedepthof3Dfeaturepointsandthespatialtransformation.Thenthe3Dfaceisdeformedbyusingnon-negativematrixfactorizationandprojectedbacktoimageplaneforfuturewarping.Toefficientlysolvethenonlinearspatialtransformation,weproposeanovelinitializationschemetosetupLevenberg-Marquardtoptimization.Accordingtothespatialtransformation,exaggerationisappliedtothemostsalientfeaturesbyexaggeratingtheirnormalizeddifferencefromthemean.Non-photorealisticrendering(NPR)basedstylizationcompletesthecartooncaricature.Experimentsdemonstratethatourmethodoutperformsexistingmethodsintermsofviewanglesandaestheticvisualquality.

简介：Whilecloud-basedBPM(BusinessProcessManagement)showspotentialsofinherentscalabilityandexpenditurereduction,suchissuesasuserautonomy,privacyprotectionandefficiencyhavepoppedupasmajorconcerns.Usersmayhavetheirownrudimentaryorevenfull-edgedBPMsystems,whichmaybeembodiedbylocalEAIsystems,attheirend,butstillintendtomakeuseofcloud-sideinfrastructureservicesandBPMcapabilities,whichmayappearasPaaS(Platform-as-a-Service)services,atthesametime.Awholebusinessprocessmaycontainanumberofnon-compute-intensiveactivities,forwhichcloudcomputingisover-provision.Moreover,someusersfeardataleakageandlossofprivacyiftheirsensitivedataisprocessedinthecloud.Thispaperproposesandanalyzesanovelarchitectureofcloud-basedBPM,whichsupportsuser-enddistributionofnon-compute-intensiveactivitiesandsensitivedata.Anapproachtooptimaldistributionofactivitiesanddataforsyntheticallyutilizingbothuser-endandcloud-sideresourcesisdiscussed.Experimentalresultsshowthatwiththehelpofsuitabledistributionschemes,dataprivacycanbesatisfactorilyprotected,andresourcesonbothsidescanbeutilizedatlowercost.