Robustness under node churn

Evaluate quantitatively the robustness of Tram-FL under node churn, including failures and changing node availability during training.

Background

Tram-FL carries the complete training state in the circulating model and momentum, allowing a preceding node to reroute the state after a transmission timeout and redistribute the failed node’s allocation in a subsequent cycle. Although this suggests that node failures need not destroy the global training state, the paper does not experimentally or theoretically quantify performance under node churn.

References

A quantitative evaluation of robustness under node churn is left as future work.

— Tram-FL: Reducing Communication and Computation Costs through Sequential Model Circulation in Decentralized Federated Learning  (2610.07859 - Maejima et al., 6 Oct 2026) in Section 3.1, System Model