TY - GEN
T1 - JuryGCN
T2 - 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD 2022
AU - Kang, Jian
AU - Zhou, Qinghai
AU - Tong, Hanghang
N1 - In this paper, we study the problem of jackknife uncertainty quantification on Graph Convolutional Network (GCN) from the fre-quentist perspective. We formally define the jackknife uncertainty of a node as the width of confidence interval by a jackknife (leave-one-out) estimator. To scale up the computation, we rely on the influence functions for efficient estimation of the leave-one-out parameters without re-training. The proposed JuryGCN framework is applied to both active learning, where the most uncertain nodes are selected to query the oracle, and semi-supervised node classification, where the jackknife uncertainty serves as the importance of loss to focus on nodes with high uncertainty. Extensive evaluations on real-world datasets demonstrate the efficacy of JuryGCN in both active learning and semi-supervised node classification. Our proposed JuryGCN is able to generalize on other learning tasks beyond GCN, which is the future direction we would like to investigate. ACKNOWLEDGEMENT This work is supported by National Science Foundation under grant No. 1947135, and 2134079 by the NSF Program on Fairness in AI in collaboration with Amazon under award No. 1939725, by DARPA HR001121C0165, by NIFA award 2020-67021-32799, and Army Research Office (W911NF2110088). The content of the information in this document does not necessarily reflect the position or the policy of the Government or Amazon, and no official endorsement should be inferred. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation here on. REFERENCES
PY - 2022/8/14
Y1 - 2022/8/14
N2 - Graph Convolutional Network (GCN) has exhibited strong empirical performance in many real-world applications. The vast majority of existing works on GCN primarily focus on the accuracy while ignoring how confident or uncertain a GCN is with respect to its predictions. Despite being a cornerstone of trustworthy graph mining, uncertainty quantification on GCN has not been well studied and the scarce existing efforts either fail to provide deterministic quantification or have to change the training procedure of GCN by introducing additional parameters or architectures. In this paper, we propose the first frequentist-based approach named JuryGCN in quantifying the uncertainty of GCN, where the key idea is to quantify the uncertainty of a node as the width of confidence interval by a jackknife estimator. Moreover, we leverage the influence functions to estimate the change in GCN parameters without re-training to scale up the computation. The proposed JuryGCN is capable of quantifying uncertainty deterministically without modifying the GCN architecture or introducing additional parameters. We perform extensive experimental evaluation on real-world datasets in the tasks of both active learning and semi-supervised node classification, which demonstrate the efficacy of the proposed method.
AB - Graph Convolutional Network (GCN) has exhibited strong empirical performance in many real-world applications. The vast majority of existing works on GCN primarily focus on the accuracy while ignoring how confident or uncertain a GCN is with respect to its predictions. Despite being a cornerstone of trustworthy graph mining, uncertainty quantification on GCN has not been well studied and the scarce existing efforts either fail to provide deterministic quantification or have to change the training procedure of GCN by introducing additional parameters or architectures. In this paper, we propose the first frequentist-based approach named JuryGCN in quantifying the uncertainty of GCN, where the key idea is to quantify the uncertainty of a node as the width of confidence interval by a jackknife estimator. Moreover, we leverage the influence functions to estimate the change in GCN parameters without re-training to scale up the computation. The proposed JuryGCN is capable of quantifying uncertainty deterministically without modifying the GCN architecture or introducing additional parameters. We perform extensive experimental evaluation on real-world datasets in the tasks of both active learning and semi-supervised node classification, which demonstrate the efficacy of the proposed method.
KW - graph neural networks
KW - jackknife
KW - uncertainty quantification
UR - http://www.scopus.com/inward/record.url?scp=85137145457&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85137145457&partnerID=8YFLogxK
U2 - 10.1145/3534678.3539286
DO - 10.1145/3534678.3539286
M3 - Conference contribution
AN - SCOPUS:85137145457
T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
SP - 742
EP - 752
BT - KDD 2022 - Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
PB - Association for Computing Machinery
Y2 - 14 August 2022 through 18 August 2022
ER -