Structural Re-weighting Improves Graph Domain Adaptation

Shikun Liu, Tianchun Li, Yongbin Feng, Nhan Tran, Han Zhao, Qiang Qiu, Pan Li
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:21778-21793, 2023.

Abstract

In many real-world applications, graph-structured data used for training and testing have differences in distribution, such as in high energy physics (HEP) where simulation data used for training may not match real experiments. Graph domain adaptation (GDA) is a method used to address these differences. However, current GDA primarily works by aligning the distributions of node representations output by a single graph neural network encoder shared across the training and testing domains, which may often yield sub-optimal solutions. This work examines different impacts of distribution shifts caused by either graph structure or node attributes and identifies a new type of shift, named conditional structure shift (CSS), which current GDA approaches are provably sub-optimal to deal with. A novel approach, called structural reweighting (StruRW), is proposed to address this issue and is tested on synthetic graphs, four benchmark datasets, and a new application in HEP. StruRW has shown significant performance improvement over the baselines in the settings with large graph structure shifts, and reasonable performance improvement when node attribute shift dominates.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-liu23u, title = {Structural Re-weighting Improves Graph Domain Adaptation}, author = {Liu, Shikun and Li, Tianchun and Feng, Yongbin and Tran, Nhan and Zhao, Han and Qiu, Qiang and Li, Pan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {21778--21793}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/liu23u/liu23u.pdf}, url = {https://proceedings.mlr.press/v202/liu23u.html}, abstract = {In many real-world applications, graph-structured data used for training and testing have differences in distribution, such as in high energy physics (HEP) where simulation data used for training may not match real experiments. Graph domain adaptation (GDA) is a method used to address these differences. However, current GDA primarily works by aligning the distributions of node representations output by a single graph neural network encoder shared across the training and testing domains, which may often yield sub-optimal solutions. This work examines different impacts of distribution shifts caused by either graph structure or node attributes and identifies a new type of shift, named conditional structure shift (CSS), which current GDA approaches are provably sub-optimal to deal with. A novel approach, called structural reweighting (StruRW), is proposed to address this issue and is tested on synthetic graphs, four benchmark datasets, and a new application in HEP. StruRW has shown significant performance improvement over the baselines in the settings with large graph structure shifts, and reasonable performance improvement when node attribute shift dominates.} }
Endnote
%0 Conference Paper %T Structural Re-weighting Improves Graph Domain Adaptation %A Shikun Liu %A Tianchun Li %A Yongbin Feng %A Nhan Tran %A Han Zhao %A Qiang Qiu %A Pan Li %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-liu23u %I PMLR %P 21778--21793 %U https://proceedings.mlr.press/v202/liu23u.html %V 202 %X In many real-world applications, graph-structured data used for training and testing have differences in distribution, such as in high energy physics (HEP) where simulation data used for training may not match real experiments. Graph domain adaptation (GDA) is a method used to address these differences. However, current GDA primarily works by aligning the distributions of node representations output by a single graph neural network encoder shared across the training and testing domains, which may often yield sub-optimal solutions. This work examines different impacts of distribution shifts caused by either graph structure or node attributes and identifies a new type of shift, named conditional structure shift (CSS), which current GDA approaches are provably sub-optimal to deal with. A novel approach, called structural reweighting (StruRW), is proposed to address this issue and is tested on synthetic graphs, four benchmark datasets, and a new application in HEP. StruRW has shown significant performance improvement over the baselines in the settings with large graph structure shifts, and reasonable performance improvement when node attribute shift dominates.
APA
Liu, S., Li, T., Feng, Y., Tran, N., Zhao, H., Qiu, Q. & Li, P.. (2023). Structural Re-weighting Improves Graph Domain Adaptation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:21778-21793 Available from https://proceedings.mlr.press/v202/liu23u.html.

Related Material