OblivSage: Oblivious Graph Sampling for Privacy-Preserving GNN
Published in Australasian Conference on Information Security and Privacy (ACISP 2026), 2026
Recommended citation: Zhibo Xu, Yiming Qin, Shangqi Lai, Xiaoning Liu, Alsharif Abuadbba, Tsz Hon Yuen, Joseph K. Liu, and Xingliang Yuan. “OblivSage: Oblivious Graph Sampling for Privacy-Preserving GNN.” In Information Security and Privacy (ACISP 2026), Lecture Notes in Computer Science, vol. 16793, pp. 431–450, 2026. https://doi.org/10.1007/978-981-92-3015-0_20
OblivSage enables privacy-preserving collaborative training of graph neural networks by using function secret sharing for dynamic, oblivious subgraph sampling. It reduces the computational and communication cost of secure GNN training while protecting graph topology and sampling access patterns.