Robustness to adversarial and decentralized federated settings

Establish the robustness of FIRM to manipulated Fisher statistics, poisoned task vectors, client dropout, and fully decentralized aggregation.

Background

FIRM assumes a trusted central server and honest-but-curious clients. The authors explicitly identify unresolved robustness questions involving malicious or unreliable client behavior, incomplete participation, and removal of the trusted centralized server.

References

Robustness to manipulated Fisher statistics, poisoned task vectors, client dropout, and fully decentralized aggregation remains open.

— FIRM: Fisher-Informed Retention Merging for Federated LLM Unlearning  (2610.21720 - Shah et al., 15 Oct 2026) in Limitations section