TY - GEN
T1 - Distributional Shift Adaptation using Domain-Specific Features
AU - Tahir, Anique
AU - Cheng, Lu
AU - Guo, Ruocheng
AU - Liu, Huan
N1 - Funding Information:
ACKNOWLEDGEMENTS We thank Paras Sheth for his valuable help and feedback. This work was supported by the Office of Naval Research under Award No. N00014-21-1-4002 and the National Science Foundation (NSF) grant 2036127. Opinions, interpretations, conclusions, and recommendations are those of the authors.
Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - Machine learning algorithms typically assume that the training and test samples come from the same distributions, i.e., in-distribution. However, in open-world scenarios, streaming big data can be Out-Of-Distribution (OOD), rendering these algorithms ineffective. Prior solutions to the OOD challenge seek to identify invariant features across different training domains. The underlying assumption is that these invariant features should also work reasonably well in the unlabeled target domain. By contrast, this work is interested in the domain-specific features that include both invariant features and features unique to the target domain. We propose a simple yet effective approach that relies on correlations in general regardless of whether the features are invariant or not. Our approach uses the most confidently predicted samples identified by an OOD base model (teacher model) to train a new model (student model) that effectively adapts to the target domain. Empirical evaluations on benchmark datasets show that the performance is improved over the SOTA by ∼10-20%.
AB - Machine learning algorithms typically assume that the training and test samples come from the same distributions, i.e., in-distribution. However, in open-world scenarios, streaming big data can be Out-Of-Distribution (OOD), rendering these algorithms ineffective. Prior solutions to the OOD challenge seek to identify invariant features across different training domains. The underlying assumption is that these invariant features should also work reasonably well in the unlabeled target domain. By contrast, this work is interested in the domain-specific features that include both invariant features and features unique to the target domain. We propose a simple yet effective approach that relies on correlations in general regardless of whether the features are invariant or not. Our approach uses the most confidently predicted samples identified by an OOD base model (teacher model) to train a new model (student model) that effectively adapts to the target domain. Empirical evaluations on benchmark datasets show that the performance is improved over the SOTA by ∼10-20%.
UR - http://www.scopus.com/inward/record.url?scp=85147963336&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85147963336&partnerID=8YFLogxK
U2 - 10.1109/BigData55660.2022.10020444
DO - 10.1109/BigData55660.2022.10020444
M3 - Conference contribution
AN - SCOPUS:85147963336
T3 - Proceedings - 2022 IEEE International Conference on Big Data, Big Data 2022
SP - 5593
EP - 5597
BT - Proceedings - 2022 IEEE International Conference on Big Data, Big Data 2022
A2 - Tsumoto, Shusaku
A2 - Ohsawa, Yukio
A2 - Chen, Lei
A2 - Van den Poel, Dirk
A2 - Hu, Xiaohua
A2 - Motomura, Yoichi
A2 - Takagi, Takuya
A2 - Wu, Lingfei
A2 - Xie, Ying
A2 - Abe, Akihiro
A2 - Raghavan, Vijay
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 IEEE International Conference on Big Data, Big Data 2022
Y2 - 17 December 2022 through 20 December 2022
ER -