2025年计算机专业英语考试试卷及答案.docxVIP

2025年计算机专业英语考试试卷及答案

第一部分阅读理解（共15题，每题2分，满分30分）

阅读以下三篇短文，从每题所给的A、B、C、D四个选项中选出最佳答案。

Passage1

Inrecentyears,theconvergenceofgenerativeAIandedgecomputinghasredefinedthelandscapeofintelligentdevices.TraditionalcloudbasedAImodels,whilepowerful,sufferfromlatencyissueswhenprocessingrealtimedata—forexample,aselfdrivingcarrequiringinstantdecisionmakingonroadconditionscannotaffordtowaitfordatatotraveltoaremoteserverandback.EdgeAI,whichdeployslightweightmodelsdirectlyonlocaldevices(e.g.,smartphones,IoTsensors,orinvehiclesystems),addressesthischallengebyenablingondeviceinference.

Akeybreakthroughinthisfieldisthedevelopmentofmodelquantizationtechniques.Byreducingtheprecisionofnumericalrepresentations(e.g.,from32bitfloatingpointsto8bitintegers),researchershavesuccessfullycompressedmodelsizesby48timeswithoutsignificantlossofaccuracy.ThisallowsevenlowenddevicestoruncomplexAItaskslikeobjectdetectionorspeechrecognition.Forinstance,thelatestgenerationofsmartphonechipsetsnowincludesdedicatedneuralprocessingunits(NPUs)optimizedforquantizedmodels,achievingover10TOPS(teraoperationspersecond)ofcomputepoweratsub2Wpowerconsumption.

However,edgeAIisnotwithoutchallenges.Thelimitedstorageandcomputationalresourcesofedgedevicesrestrictthecomplexityofmodelsthatcanbedeployed.Tomitigatethis,researchersareexploringfederatedlearning,wheremodelsaretrainedcollaborativelyacrossdeviceswithoutsharingrawdata.Thisapproachnotonlypreservesuserprivacybutalsoleveragesthecollectivedatafrommillionsofdevicestoimprovemodelgeneralization.

1.WhatistheprimaryproblemwithcloudbasedAImodelsmentionedinthepassage?

A.Highenergyconsumption

B.Latencyinrealtimeprocessing

C.Inabilitytohandlelargedatasets

D.Lackofprivacyprotection

2.Howdomodelqua