Research map / Federated learning and privacy / Personalized and heterogeneous FL
UB-SMoE: Universally Balanced Sparse Mixture-of-Experts for Resource-adaptive Federated Fine-tuning of Foundation Models
Van-Tuan Tran, Hong-Hanh Nguyen-Le, Marco Ruffini, Merim Dzaferagic
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add pseudo-gradients (ablation) | UB-SMoE · Commonsense-15K avg | accuracy: 0.1701 -> 0.2659 | Improved |
| add phi regularization and utilization-aware update (DMR) on top of PG (ablation) | UB-SMoE · Commonsense-15K avg | accuracy: 0.2659 -> 0.4267 | Improved |
| add phi regularization to PG (ablation) | UB-SMoE · Commonsense-15K avg | accuracy: 0.2659 -> 0.2839 | Improved |
| add utilization-aware update to PG (ablation) | UB-SMoE · Commonsense-15K avg | accuracy: 0.2659 -> 0.3591 | Improved |
| add pseudo-gradients to DMR (ablation) | UB-SMoE · Commonsense-15K avg | accuracy: 0.4009 -> 0.4267 | Improved |
| use sparse experts instead of heterogeneous LoRA rank at tight budget | UB-SMoE vs HetLoRA · Commonsense-15K beta1 | accuracy: 0.0079 -> 0.3936 | Improved |
| use UB-SMoE instead of A3SMoE | UB-SMoE · Commonsense-15K OLMoE avg | accuracy: 0.3861 -> 0.4267 | Improved |
| use UB-SMoE instead of A3SMoE at budget beta2 | UB-SMoE · Commonsense-15K beta2 | accuracy: 0.4310 -> 0.3359 | Hurt |
| use UB-SMoE instead of A3SMoE under non-IID | UB-SMoE · TeleQuAD non-IID avg | BERTScore F1: 40.70 -> 60.58 | Improved |
| use UB-SMoE instead of A3SMoE under IID | UB-SMoE · TeleQuAD IID avg | BERTScore F1: 60.62 -> 59.63 | No clear effect |
| use UB-SMoE instead of FlexLoRA on dense OLMo-1B | UB-SMoE · Commonsense-15K OLMo-1B | accuracy: 0.3453 -> 0.5636 | Improved |
| use UB-SMoE instead of A3SMoE with 32 long-tail clients | UB-SMoE · Commonsense-15K 32 clients long-tail | accuracy: 0.2988 -> 0.3047 | Improved |
Closest papers by meaning
- Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.88)
- Cluster Based Heterogeneous Federated Foundation Model Adaptation and Fine-Tuning (AAAI 2025 · similarity 0.87)
- SparsyFed: Sparse Adaptive Federated Learning (ICLR 2025 · similarity 0.86)
- Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA (NeurIPS 2025 · similarity 0.86)
- Merge, Then Compress: Demystify Efficient SMoE with Hints from Its Routing Policy (ICLR 2024 · similarity 0.86)
- Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
