迁移学习算法研究-庄福振NewPPT格式课件下载.ppt

1、通常不能满足！训练集测试集分类器训练集测试集分类器INSTITUTE OF COMPUTING TECHNOLOGY迁移学习迁移学习42022/11/10l实际应用学习场景HP 新闻新闻Lenovo 新闻新闻不同源、分布不一致不同源、分布不一致人工标记训练样本，费人工标记训练样本，费时耗力时耗力迁移迁移学习学习 运用已有的知识对运用已有的知识对不同但相关领域不同但相关领域问题问题进行求解的一种新的机器学习方法进行求解的一种新的机器学习方法 放宽了传统机器学习的两个基本假设放宽了传统机器学习的两个基本假设INSTITUTE OF COMPUTING TECHNOLOGY迁移学习场景迁移学习场景（

2、1/4）（1/4）52022/11/10l迁移学习场景无处不在迁移迁移知识知识迁移迁移知识知识图像分类图像分类HP 新闻新闻Lenovo 新闻新闻新闻网页分类新闻网页分类INSTITUTE OF COMPUTING TECHNOLOGY异构特征空间6The apple is the pomaceous fruit of the apple tree,species Malus domestica in the rose family Rosaceae.Banana is the common name for a type of fruit and also the herbaceous pl

3、ants of the genus Musa which produce this commonly eaten fruit.Training:TextFuture:ImagesApplesBananas迁移学习场景迁移学习场景（2/4）（2/4）2022/11/10from Prof.Qiang YangXin Jin,Fuzhen Zhuang,Sinno Jialin Pan,Changying Du,Ping Luo,Qing He:Heterogeneous Multi-task Semantic Feature Learning for Classification.CIKM 20

4、15:1847-1850.INSTITUTE OF COMPUTING TECHNOLOGY Test Test Training TrainingClassifierClassifier72.65%DVDElectronicsElectronics84.60%ElectronicsDrop!迁移学习场景迁移学习场景（3/4）（3/4）72022/11/10from Prof.Qiang YangINSTITUTE OF COMPUTING TECHNOLOGY8DVDElectronicsBookKitchenClothesVideo gameFruitHotelTeaImpractical

5、!迁移学习场景迁移学习场景（4/4）（4/4）2022/11/10from Prof.Qiang YangINSTITUTE OF COMPUTING TECHNOLOGYOutlinepConcept Learning for Transfer Learning Concept Learning based on Non-negative Matrix Tri-factorization for Transfer Learning Concept Learning based on Probabilistic Latent Semantic Analysis for Transfer Lea

6、rningpTransfer Learning using Auto-encodersTransfer Learning from Multiple Sources with Autoencoder RegularizationSupervised Representation Learning:Transfer Learning with Deep Auto-encoders92022/11/10INSTITUTE OF COMPUTING TECHNOLOGYConcept Learning based on Non-negative Matrix Tri-factorization fo

7、r Transfer LearningConcept Learning for Transfer Learning102022/11/10INSTITUTE OF COMPUTING TECHNOLOGYIntroductionMany traditional learning techniques work well only under the assumption:Training and test data follow the same distribution Training（labeled）ClassifierTest（unlabeled）From different comp

8、aniesEnterpriseNewsClassification:includingtheclasses“ProductAnnouncement”,“Businessscandal”,“Acquisition”,Product announcement:HPs just-released LaserJet Pro P1100 printer and the LaserJet Pro M1130 and M1210 multifunction printers,price performance.Announcement for Lenovo ThinkPad ThinkCentre pric

9、e$150 off Lenovo K300 desktop using coupon code.Lenovo ThinkPad ThinkCentre price$200 off Lenovo IdeaPad U450p laptop using.their performanceHP newsLenovo newsDifferent distributionFail!11Concept Learning for Transfer Learning2022/11/10INSTITUTE OF COMPUTING TECHNOLOGYMotivation（1/3）Example Analysis

10、 Product announcement:HPs just-released LaserJet Pro P1100 printer and the LaserJet Pro M1130 and M1210 multifunction printers,price performance.Announcement for Lenovo ThinkPad ThinkCentre price$150 off Lenovo K300 desktop using coupon code.Lenovo ThinkPad ThinkCentre price$200 off Lenovo IdeaPad U

11、450p laptop using.their performanceHP newsLenovo newsProductword conceptLaserJet,printer,price,performance ThinkPad,ThinkCentre,price,performance RelatedProductannouncementdocument class:12Share some common words:announcement,price,performance indicateConcept Learning for Transfer Learning2022/11/10

12、INSTITUTE OF COMPUTING TECHNOLOGYMotivation（2/3）Example Analysis:HPLaserJet,printer,price,performance et al.LenovoThinkpad,Thinkcentre,price,performance et al.The words expressing the same word concept are domain-dependent 13ProductProductannouncementword conceptindicatesThe association between word

13、 concepts and document classes is domain-independent Concept Learning for Transfer Learning2022/11/10INSTITUTE OF COMPUTING TECHNOLOGYMotivation（3/3）14Further observations:Different domains may use same key words to express the same concept（denoted as identical concept）Different domains may also use

14、 different key words to express the same concept（denoted as alike concept）Different domains may also have their own distinct concepts（denoted as distinct concept）The identical and alike concepts are used as the shared concepts for knowledge transferWe try to model these three kinds of concepts simul

15、taneously for transfer learning text classificationConcept Learning for Transfer Learning2022/11/10INSTITUTE OF COMPUTING TECHNOLOGYPreliminary KnowledgeBasic formula of matrix tri-factorization:where the input X is the word-document co-occurrence matrix denotes concept information,may vary in different domainsF denotes the document classification information indeed is the associ