f-Domain Adversarial Learning: Theory and Algorithms

David Acuna, Guojun Zhang, Marc T. Law, Sanja Fidler
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:66-75, 2021.

Abstract

Unsupervised domain adaptation is used in many machine learning applications where, during training, a model has access to unlabeled data in the target domain, and a related labeled dataset. In this paper, we introduce a novel and general domain-adversarial framework. Specifically, we derive a novel generalization bound for domain adaptation that exploits a new measure of discrepancy between distributions based on a variational characterization of f-divergences. It recovers the theoretical results from Ben-David et al. (2010a) as a special case and supports divergences used in practice. Based on this bound, we derive a new algorithmic framework that introduces a key correction in the original adversarial training method of Ganin et al. (2016). We show that many regularizers and ad-hoc objectives introduced over the last years in this framework are then not required to achieve performance comparable to (if not better than) state-of-the-art domain-adversarial methods. Experimental analysis conducted on real-world natural language and computer vision datasets show that our framework outperforms existing baselines, and obtains the best results for f-divergences that were not considered previously in domain-adversarial learning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-acuna21a, title = {f-Domain Adversarial Learning: Theory and Algorithms}, author = {Acuna, David and Zhang, Guojun and Law, Marc T. and Fidler, Sanja}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {66--75}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/acuna21a/acuna21a.pdf}, url = {https://proceedings.mlr.press/v139/acuna21a.html}, abstract = {Unsupervised domain adaptation is used in many machine learning applications where, during training, a model has access to unlabeled data in the target domain, and a related labeled dataset. In this paper, we introduce a novel and general domain-adversarial framework. Specifically, we derive a novel generalization bound for domain adaptation that exploits a new measure of discrepancy between distributions based on a variational characterization of f-divergences. It recovers the theoretical results from Ben-David et al. (2010a) as a special case and supports divergences used in practice. Based on this bound, we derive a new algorithmic framework that introduces a key correction in the original adversarial training method of Ganin et al. (2016). We show that many regularizers and ad-hoc objectives introduced over the last years in this framework are then not required to achieve performance comparable to (if not better than) state-of-the-art domain-adversarial methods. Experimental analysis conducted on real-world natural language and computer vision datasets show that our framework outperforms existing baselines, and obtains the best results for f-divergences that were not considered previously in domain-adversarial learning.} }
Endnote
%0 Conference Paper %T f-Domain Adversarial Learning: Theory and Algorithms %A David Acuna %A Guojun Zhang %A Marc T. Law %A Sanja Fidler %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-acuna21a %I PMLR %P 66--75 %U https://proceedings.mlr.press/v139/acuna21a.html %V 139 %X Unsupervised domain adaptation is used in many machine learning applications where, during training, a model has access to unlabeled data in the target domain, and a related labeled dataset. In this paper, we introduce a novel and general domain-adversarial framework. Specifically, we derive a novel generalization bound for domain adaptation that exploits a new measure of discrepancy between distributions based on a variational characterization of f-divergences. It recovers the theoretical results from Ben-David et al. (2010a) as a special case and supports divergences used in practice. Based on this bound, we derive a new algorithmic framework that introduces a key correction in the original adversarial training method of Ganin et al. (2016). We show that many regularizers and ad-hoc objectives introduced over the last years in this framework are then not required to achieve performance comparable to (if not better than) state-of-the-art domain-adversarial methods. Experimental analysis conducted on real-world natural language and computer vision datasets show that our framework outperforms existing baselines, and obtains the best results for f-divergences that were not considered previously in domain-adversarial learning.
APA
Acuna, D., Zhang, G., Law, M.T. & Fidler, S.. (2021). f-Domain Adversarial Learning: Theory and Algorithms. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:66-75 Available from https://proceedings.mlr.press/v139/acuna21a.html.

Related Material