结合域混淆与MK-MMD的深度适应网络
|
摘要
在深度学习的应用场景中,常会遇到缺乏大量标记数据的情况,域适应作为利用相关源域标记数据信息来对目标域数据进行信息补充的一种迁移学习方法,解决此类问题是非常有效的.在域适应方法中,基于最大均值误差(MMD)度量来缩小源域目标域差异的方法被广泛应用,深度适应网络(DAN)是其中经典方法之一.但是结合多核最大均值误差(MK-MMD)思想的DAN方法在特征迁移层面仍有提升空间,且该方法在不同迁移场景下的适用效果有差异.本文针对这两个问题,结合域混淆思想,进一步提升域适应效果.同时,从实验与理论两方面探究MK-MMD度量在不同场景下的适用权重以及MK-MMD与域混淆的最佳组合方式.
In deep learning based applications,the lack of labeled data is often encountered. Domain adaptation is an effective transfer learning method which uses the labeled data of source domain to supplement useful information for target domain,in which case the source domain and target domain are related. The method based on Maximum mean discrepancies( MMD) is widely applied in domain adaptation,which is used to reduce the difference between source and target domain. Deep Adaptation Network( DAN) is one of the classical methods,which utilizes Multi-Kernel MMD( MK-MMD). However,DAN can still be improved in feature level transfer and it has different effects in different adaptation scenarios. To solve these two problems,we first combine Domain Confusion with MKMMD to further improve the adaptability of models. At the same time,we explore the suitable weights of MK-MMD in different adaptation scenarios and the best combination of MK-MMD and domain confusion from both experimental and theoretical aspects.
In deep learning based applications,the lack of labeled data is often encountered. Domain adaptation is an effective transfer learning method which uses the labeled data of source domain to supplement useful information for target domain,in which case the source domain and target domain are related. The method based on Maximum mean discrepancies( MMD) is widely applied in domain adaptation,which is used to reduce the difference between source and target domain. Deep Adaptation Network( DAN) is one of the classical methods,which utilizes Multi-Kernel MMD( MK-MMD). However,DAN can still be improved in feature level transfer and it has different effects in different adaptation scenarios. To solve these two problems,we first combine Domain Confusion with MKMMD to further improve the adaptability of models. At the same time,we explore the suitable weights of MK-MMD in different adaptation scenarios and the best combination of MK-MMD and domain confusion from both experimental and theoretical aspects.
引文
[1]Russakovsky O,Deng J,Su H,et al.Imagenet large scale visual recognition challenge[J].International Journal of Computer Vision,2015,115(3):211-252.
[2]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778.
[3]Everingham M,Eslami S M A,Van Gool L,et al.The pascal visual object classes challenge:a retrospective[J].International Journal of Computer Vision,2015,111(1):98-136.
[4]Girshick R.Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision,2015:1440-1448.
[5]Krause J,Stark M,Deng J,et al.3d object representations for finegrained categorization[C]//Proceedings of the IEEE International Conference on Computer Vision Workshops,2013:554-561.
[6]Gebru T,Hoffman J,Fei-Fei L.Fine-grained recognition in the wild:A multi-task domain adaptation approach[C]//Computer Vision(ICCV),2017 IEEE International Conference on,2017:1358-1367.
[7]Cordts M,Omran M,Ramos S,et al.The cityscapes dataset for semantic urban scene understanding[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition,2016:3213-3223.
[8]Zhao H,Shi J,Qi X,et al.Pyramid scene parsing network[C]//IEEE Conf.on Computer Vision and Pattern Recognition(CVPR),2017:2881-2890.
[9]Patel V M,Gopalan R,Li R,et al.Visual domain adaptation:a survey of recent advances[J].IEEE Signal Processing Magazine,2015,32(3):53-69.
[10]Borgwardt K M,Gretton A,Rasch M J,et al.Integrating structured biological data by kernel maximum mean discrepancy[J].Bioinformatics,2006,22(14):49-57.
[11]Long M,Cao Y,Wang J,et al.Learning transferable features with deep adaptation networks[C]//Proceedings of the 32nd International Conference on International Conference on Machine LearningVolume 37.JMLR.org,2015:97-105.
[12]Tzeng E,Hoffman J,Darrell T,et al.Simultaneous deep transfer across domains and tasks[C]//Proceedings of the IEEE International Conference on Computer Vision,2015:4068-4076.
[13]Ghifary M,Kleijn W B,Zhang M.Domain adaptive neural networks for object recognition[C]//Pacific Rim International Conference on Artificial Intelligence,Springer,Cham,2014:898-904.
[14]Tzeng E,Hoffman J,Zhang N,et al.Deep domain confusion:Maximizing for domain invariance[J].arXiv preprint arXiv:1412.3474,2014.
[15]Long M,Zhu H,Wang J,et al.Deep transfer learning with joint adaptation networks[C]//International Conference on Machine Learning,2017:2208-2217.
[16]Goodfellow I J,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[C]//International Conference on Neural Information Processing Systems,MIT Press,2014:2672-2680.
[17]Ganin Y,Ustinova E,Ajakan H,et al.Domain-adversarial training of neural networks[J].Journal of Machine Learning Research,2016,17(59):1-35.
[18]Bousmalis K,Silberman N,Dohan D,et al.Unsupervised pixellevel domain adaptation with generative adversarial networks[C]//IEEE Conference on Computer Vision and Pattern Recognition,2017:95-104.
[19]Gretton A,Sejdinovic D,Strathmann H,et al.Optimal kernel choice for large-scale two-sample tests[C]//Advances in Neural Information Processing Systems,2012:1205-1213.
[2]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778.
[3]Everingham M,Eslami S M A,Van Gool L,et al.The pascal visual object classes challenge:a retrospective[J].International Journal of Computer Vision,2015,111(1):98-136.
[4]Girshick R.Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision,2015:1440-1448.
[5]Krause J,Stark M,Deng J,et al.3d object representations for finegrained categorization[C]//Proceedings of the IEEE International Conference on Computer Vision Workshops,2013:554-561.
[6]Gebru T,Hoffman J,Fei-Fei L.Fine-grained recognition in the wild:A multi-task domain adaptation approach[C]//Computer Vision(ICCV),2017 IEEE International Conference on,2017:1358-1367.
[7]Cordts M,Omran M,Ramos S,et al.The cityscapes dataset for semantic urban scene understanding[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition,2016:3213-3223.
[8]Zhao H,Shi J,Qi X,et al.Pyramid scene parsing network[C]//IEEE Conf.on Computer Vision and Pattern Recognition(CVPR),2017:2881-2890.
[9]Patel V M,Gopalan R,Li R,et al.Visual domain adaptation:a survey of recent advances[J].IEEE Signal Processing Magazine,2015,32(3):53-69.
[10]Borgwardt K M,Gretton A,Rasch M J,et al.Integrating structured biological data by kernel maximum mean discrepancy[J].Bioinformatics,2006,22(14):49-57.
[11]Long M,Cao Y,Wang J,et al.Learning transferable features with deep adaptation networks[C]//Proceedings of the 32nd International Conference on International Conference on Machine LearningVolume 37.JMLR.org,2015:97-105.
[12]Tzeng E,Hoffman J,Darrell T,et al.Simultaneous deep transfer across domains and tasks[C]//Proceedings of the IEEE International Conference on Computer Vision,2015:4068-4076.
[13]Ghifary M,Kleijn W B,Zhang M.Domain adaptive neural networks for object recognition[C]//Pacific Rim International Conference on Artificial Intelligence,Springer,Cham,2014:898-904.
[14]Tzeng E,Hoffman J,Zhang N,et al.Deep domain confusion:Maximizing for domain invariance[J].arXiv preprint arXiv:1412.3474,2014.
[15]Long M,Zhu H,Wang J,et al.Deep transfer learning with joint adaptation networks[C]//International Conference on Machine Learning,2017:2208-2217.
[16]Goodfellow I J,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[C]//International Conference on Neural Information Processing Systems,MIT Press,2014:2672-2680.
[17]Ganin Y,Ustinova E,Ajakan H,et al.Domain-adversarial training of neural networks[J].Journal of Machine Learning Research,2016,17(59):1-35.
[18]Bousmalis K,Silberman N,Dohan D,et al.Unsupervised pixellevel domain adaptation with generative adversarial networks[C]//IEEE Conference on Computer Vision and Pattern Recognition,2017:95-104.
[19]Gretton A,Sejdinovic D,Strathmann H,et al.Optimal kernel choice for large-scale two-sample tests[C]//Advances in Neural Information Processing Systems,2012:1205-1213.