AMSFL adaptively allocates local training steps in federated learning using gradient-difference drift estimates, but the stated convergence and optimality proofs contain significant errors.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
AMSFL: Adaptive Multi-Step Federated Learning via Gradient Difference-Based Error Modeling
AMSFL adaptively allocates local training steps in federated learning using gradient-difference drift estimates, but the stated convergence and optimality proofs contain significant errors.