Trust-Aware Topology Learning for Dynamic Decentralized Federated Learning under Adversaries

This work shows that in decentralized learning, adversaries can attack the collaboration topology itself, not just model updates, and introduces DMTT, a trust-screened aggregation protocol that provably bounds Byzantine influence while empirically matching no-attack accuracy across mobile sensing benchmarks where existing robust aggregators collapse.

August 2026 · Shubham Vaishnav, Murtaza Rangwala, Ali Beikmohammadi, Sindri Magnússon, Rajkumar Buyya