雷鋒網消息，谷歌大腦團隊的IanGoodfellow今日在研究院官網上撰文，總結了谷歌在ICLR2017上所做的學術貢獻。雷鋒網編譯全文如下，未經許可不得轉載。
本周，第五屆國際學習表徵會議（ICLR2017）在法國土倫召開，這是一個關注機器學習領域如何從數據中習得具有意義及有用表徵的會議。ICLR包括conferencetrack及workshoptrack兩個項目，邀請了獲得oral及poster的研究者們進行分享，涵蓋深度學習、度量學習、核學習、組合模型、非線性結構化預測，及非凸優化問題。
站在神經網絡及深度學習領域浪潮之巔，谷歌關注理論與實踐，並致力於開發理解與總結的學習方法。作為ICLR2017的白金贊助商，谷歌有超過50名研究者出席本次會議（大部分為谷歌大腦團隊及谷歌歐洲研究分部的成員），通過在現場展示論文及海報的方式，為建設一個更完善的學術研究交流平臺做出了貢獻，也是一個互相學習的過程。此外，谷歌的研究者也是workshops及組委會構建的中堅力量。
如果你來到ICLR2017，我們希望你能在我們的展臺前駐足，並與我們的研究者進行交流，探討如何為數十億人解決有趣的問題。
以下為谷歌在ICLR2017展示的論文內容（其中的谷歌研究者已經加粗表示）
區域主席
GeorgeDahl,SlavPetrov,VikasSindhwani
程序主席（雷鋒網此前已經做過相關介紹）
HugoLarochelle,TaraSainath
受邀演講論文
UnderstandingDeepLearningRequiresRethinkingGeneralization(BestPaperAward)
ChiyuanZhang*,SamyBengio,MoritzHardt,BenjaminRecht*,OriolVinyals
Semi-SupervisedKnowledgeTransferforDeepLearningfromPrivateTrainingData(BestPaperAward)
NicolasPapernot*,MartínAbadi,lfarErlingsson,IanGoodfellow,KunalTalwar
Q-Prop:Sample-EfficientPolicyGradientwithAnOff-PolicyCritic
Shixiang(Shane)Gu*,TimothyLillicrap,ZoubinGhahramani,RichardE.Turner,SergeyLevine
NeuralArchitectureSearchwithReinforcementLearning
BarretZoph,QuocLe
Poster論文
AdversarialMachineLearningatScale
AlexeyKurakin,IanJ.Goodfellow,SamyBengio
CapacityandTrainabilityinRecurrentNeuralNetworks
JasmineCollins,JaschaSohl-Dickstein,DavidSussillo
ImprovingPolicyGradientbyExploringUnder-AppreciatedRewards
OfirNachum,MohammadNorouzi,DaleSchuurmans
OutrageouslyLargeNeuralNetworks:TheSparsely-GatedMixture-of-ExpertsLayer
NoamShazeer,AzaliaMirhoseini,KrzysztofMaziarz,AndyDavis,QuocLe,GeoffreyHinton,JeffDean
UnrolledGenerativeAdversarialNetworks
LukeMetz,BenPoole*,DavidPfau,JaschaSohl-Dickstein
CategoricalReparameterizationwithGumbel-Softmax
EricJang,Shixiang(Shane)Gu*,BenPoole*
DecomposingMotionandContentforNaturalVideoSequencePrediction
RubenVillegas,JimeiYang,SeunghoonHong,XunyuLin,HonglakLee
DensityEstimationUsingRealNVP
LaurentDinh*,JaschaSohl-Dickstein,SamyBengio
LatentSequenceDecompositions
WilliamChan*,YuZhang*,QuocLe,NavdeepJaitly*
LearningaNaturalLanguageInterfacewithNeuralProgrammer
ArvindNeelakantan*,QuocV.Le,MartínAbadi,AndrewMcCallum*,DarioAmodei*
DeepInformationPropagation
SamuelSchoenholz,JustinGilmer,SuryaGanguli,JaschaSohl-Dickstein
IdentityMattersinDeepLearning
MoritzHardt,TengyuMa
ALearnedRepresentationForArtisticStyle
VincentDumoulin*,JonathonShlens,ManjunathKudlur
AdversarialTrainingMethodsforSemi-SupervisedTextClassification
TakeruMiyato,AndrewM.Dai,IanGoodfellow
HyperNetworks
DavidHa,AndrewDai,QuocV.Le
LearningtoRememberRareEvents
LukaszKaiser,OfirNachum,AurkoRoy*,SamyBengio
WorkshopTrack
ParticleValueFunctions
ChrisJ.Maddison,DieterichLawson,GeorgeTucker,NicolasHeess,ArnaudDoucet,AndriyMnih,YeeWhyeTeh
NeuralCombinatorialOptimizationwithReinforcementLearning
IrwanBello,HieuPham,QuocV.Le,MohammadNorouzi,SamyBengio
ShortandDeep:SketchingandNeuralNetworks
AmitDaniely,NevenaLazic,YoramSinger,KunalTalwar
ExplainingtheLearningDynamicsofDirectFeedbackAlignment
JustinGilmer,ColinRaffel,SamuelS.Schoenholz,MaithraRaghu,JaschaSohl-Dickstein
TrainingaSubsamplingMechanisminExpectation
ColinRaffel,DieterichLawson
TuningRecurrentNeuralNetworkswithReinforcementLearning
NatashaJaques*,Shixiang(Shane)Gu*,RichardE.Turner,DouglasEck
REBAR:Low-Variance,UnbiasedGradientEstimatesforDiscreteLatentVariableModels
GeorgeTucker,AndriyMnih,ChrisJ.Maddison,JaschaSohl-Dickstein
AdversarialExamplesinthePhysicalWorld
AlexeyKurakin,IanGoodfellow,SamyBengio
RegularizingNeuralNetworksbyPenalizingConfidentOutputDistributions
GabrielPereyra,GeorgeTucker,JanChorowski,LukaszKaiser,GeoffreyHinton
UnsupervisedPerceptualRewardsforImitationLearning
PierreSermanet,KelvinXu,SergeyLevine
ChangingModelBehavioratTest-timeUsingReinforcementLearning
AugustusOdena,DieterichLawson,ChristopherOlah
*工作內容在谷歌就職時完成
工作內容在OpenAI任職時完成
詳細信息可訪問了解，雷鋒網編譯。research.googleblog