原创力文档- 1、本文档共79页，可阅读全部内容。
- 2、原创力文档（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)。
Answerbookletforstudents
July25,2023
PAGE2
PAGE2
Copyright
Copyright?2023SimonPrince.
Copyright
Copyright?2023SimonPrince.
Answerbooklet
ThisdocumentaccompaniesthebookUnderstandingDeepLearning.Itcontainsanswerstoaselectedsubsetoftheproblemsattheendofeachchapterofthemainbook.TheremaininganswersareavailableonlytoinstructorsviatheMITPress.
Thisbooklethasnotyetbeencheckedverycarefully.IreallyneedyourhelpinthisregardandI’dbeverygratefulifyouwouldmailmeatudlbookmail@ifyoucannotunderstandthetextorifyouthinkthatyoufindamistake.Suggestionsforextraproblemswillalsobegratefullyreceived!
SimonPrinceJuly25,2023
PAGE4
PAGE4
Copyright
Copyright?2023SimonPrince.
Copyright
Copyright?2023SimonPrince.
Chapter2
Supervisedlearning
Problem2.1Towalk“downhill”onthelossfunction(equation2.5),wemeasureitsgradientwithrespecttotheparameters?0and?1.Calculateexpressionsfortheslopes?L/??0and?L/??1.
Problem2.2Showthatwecanfindtheminimumofthelossfunctioninclosed-formbysettingtheexpressionforthederivativesfromproblem2.1tozeroandsolvingfor?0and?1.
Problem2.3Considerreformulatinglinearregressionasagenerativemodelsowehavex=g[y,?]=?0+?1y.Whatisthenewlossfunction?Findanexpressionfortheinversefunctiony=g?1[x,?]thatwewouldusetoperforminference.Willthismodelmakethesamepredictionsasthediscriminativeversionforagiventrainingdataset{xi,yi}?Onewaytoestablishthisistowritecodethatfitsalinetothreedatapointsusingbothmethodsandseeiftheresultisthesame.
Answer
Wecantriviallysolveforxbyrearrangingtheequation:
y=x??0.
?1
Theanswersarenotthesame;youcanshowthisbyfindingequationsfortheslopeandinterceptofthefunctionoflinerelatingxtoyandshowingtheyarenotthesame.Oryoucanjusttryfittingbothmodelstos
文档评论（0）