Topology-Based Reconstruction Prevention for Decentralised Learning
This addresses security vulnerabilities in decentralized learning systems for privacy-sensitive applications, offering a novel defense but is incremental as it builds on existing privacy mechanisms.
The paper tackles the problem of reconstruction attacks in decentralized learning, where adversaries can infer private data after multiple privacy-preserving summations, achieving an 11.0% success rate with three adversaries in subgraphs of 18 users. It proposes the first topology-based defense, showing that reconstruction requires adversaries linear in the shortest cycle length, making attacks impossible in acyclic networks.
Decentralised learning has recently gained traction as an alternative to federated learning in which both data and coordination are distributed. To preserve the confidentiality of users' data, decentralised learning relies on differential privacy, multi-party computation, or both. However, running multiple privacy-preserving summations in sequence may allow adversaries to perform reconstruction attacks. Current reconstruction countermeasures either cannot trivially be adapted to the distributed setting, or add excessive amounts of noise. In this work, we first show that passive honest-but-curious adversaries can infer other users' private data after several privacy-preserving summations. For example, in subgraphs with 18 users, we show that only three passive honest-but-curious adversaries succeed at reconstructing private data 11.0% of the time, requiring an average of 8.8 summations per adversary. The success rate depends only on the adversaries' direct neighbourhood, and is independent of the size of the full network. We consider weak adversaries that do not control the graph topology, cannot exploit the summation's inner workings, and do not have auxiliary knowledge; and show that these adversaries can still infer private data. We analyse how reconstruction relates to topology and propose the first topology-based decentralised defence against reconstruction attacks. We show that reconstruction requires a number of adversaries linear in the length of the network's shortest cycle. Consequently, exact attacks over privacy-preserving summations are impossible in acyclic networks. Our work is a stepping stone for a formal theory of topology-based decentralised reconstruction defences. Such a theory would generalise our countermeasure beyond summation, define confidentiality in terms of entropy, and describe the interactions with (topology-aware) differential privacy.