Multi-agent privacy amplification for collaborative GNNs

Develop privacy-amplification mechanisms for collaborative graph settings that exploit sequential multi-agent computation chains to strengthen privacy while avoiding the information loss caused by local differential privacy.

Background

Most existing graph-learning privacy mechanisms apply local differential privacy independently to training or inference data, which can substantially reduce learning effectiveness. The survey proposes privacy amplification across collaborative computation chains as a promising but unestablished alternative for sequential MPNN operations.

References

This approach fits naturally with sequential MPNN operations and could limit information loss while maintaining privacy guarantees, but is still unexplored for collaborative graph settings.

— From Euclidean to Graph-Structured Data: A Survey of Collaborative Learning  (2609.02984 - Bourgerie et al., 2 Sep 2026) in Section 7, paragraph “Open challenges”