AI Paper Atlas

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

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Non-IID federated learning · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add pseudo-gradients (ablation)UB-SMoE · Commonsense-15K avgaccuracy: 0.1701 -> 0.2659Improved
add phi regularization and utilization-aware update (DMR) on top of PG (ablation)UB-SMoE · Commonsense-15K avgaccuracy: 0.2659 -> 0.4267Improved
add phi regularization to PG (ablation)UB-SMoE · Commonsense-15K avgaccuracy: 0.2659 -> 0.2839Improved
add utilization-aware update to PG (ablation)UB-SMoE · Commonsense-15K avgaccuracy: 0.2659 -> 0.3591Improved
add pseudo-gradients to DMR (ablation)UB-SMoE · Commonsense-15K avgaccuracy: 0.4009 -> 0.4267Improved
use sparse experts instead of heterogeneous LoRA rank at tight budgetUB-SMoE vs HetLoRA · Commonsense-15K beta1accuracy: 0.0079 -> 0.3936Improved
use UB-SMoE instead of A3SMoEUB-SMoE · Commonsense-15K OLMoE avgaccuracy: 0.3861 -> 0.4267Improved
use UB-SMoE instead of A3SMoE at budget beta2UB-SMoE · Commonsense-15K beta2accuracy: 0.4310 -> 0.3359Hurt
use UB-SMoE instead of A3SMoE under non-IIDUB-SMoE · TeleQuAD non-IID avgBERTScore F1: 40.70 -> 60.58Improved
use UB-SMoE instead of A3SMoE under IIDUB-SMoE · TeleQuAD IID avgBERTScore F1: 60.62 -> 59.63No clear effect
use UB-SMoE instead of FlexLoRA on dense OLMo-1BUB-SMoE · Commonsense-15K OLMo-1Baccuracy: 0.3453 -> 0.5636Improved
use UB-SMoE instead of A3SMoE with 32 long-tail clientsUB-SMoE · Commonsense-15K 32 clients long-tailaccuracy: 0.2988 -> 0.3047Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy