ArKrum combines median-based outlier filtering with multi-update averaging to make Krum-style federated aggregation parameter-free and more stable, matching or beating Krum and mKrum on benchmark attacks, but failing on label flipping.
In: Proceedings of the Int’l ACM Symposium on Mobility Management and Wireless Access
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Secure and Private Federated Learning: Achieving Adversarial Resilience through Robust Aggregation
ArKrum combines median-based outlier filtering with multi-update averaging to make Krum-style federated aggregation parameter-free and more stable, matching or beating Krum and mKrum on benchmark attacks, but failing on label flipping.