Sparse MoE adapters with per-client expert selection and a thresholded load-balancing loss improve federated fine-tuning accuracy under non-IID data compared with LoRA baselines.
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
-
FFT-MoE: Efficient Federated Fine-Tuning for Foundation Models via Large-scale Sparse MoE under Heterogeneous Edge
Sparse MoE adapters with per-client expert selection and a thresholded load-balancing loss improve federated fine-tuning accuracy under non-IID data compared with LoRA baselines.