FSL-SAGE lets memory-constrained clients train large federated models in parallel by using periodically aligned auxiliary models to estimate server-side gradient feedback, with claimed O(1/√T) convergence.
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
CONDITIONAL 1representative citing papers
citing papers explorer
-
FSL-SAGE: Accelerating Federated Split Learning via Smashed Activation Gradient Estimation
FSL-SAGE lets memory-constrained clients train large federated models in parallel by using periodically aligned auxiliary models to estimate server-side gradient feedback, with claimed O(1/√T) convergence.