Rumor Source Detection With Multiple Observations Under Adaptive Diffusions

Recent work, motivated by anonymous messaging platforms, has introduced adaptive diffusion protocols which can obfuscate the source of a rumor: “snapshot adversary” with access to subgraph “infected” nodes do no better than randomly guessing entity node. What happens if adversary multiple independent snapshots? We study this question when underlying graph is infinite d-regular tree. show that (1) weak form obfuscation still possible in case two snapshots, but (2) already three observations there simple algorithm finds rumor constant probability, regardless protocol. also characterize tradeoff between local spreading and for (under single snapshot). These results raise questions about robustness anonymity guarantees information social networks.