AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

Decoupled Low-Rank Adaptation for Robust Federated Fine-Tuning

Xiuwen Fang, Xuliang Yang, Mang Ye

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated prompt and LoRA tuning

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 decoupled LoRA fine-tuning (dual-branch sample partition) after warm-up (ablation)FedDR-LoRA · CIFAR-100, Pairflip 0.2accuracy (%): 88.08 -> 90.54Improved
add decoupled LoRA fine-tuning after warm-up (ablation)FedDR-LoRA · CIFAR-100, Pairflip 0.4accuracy (%): 85.41 -> 87.37Improved
add noisy branch negative learning on top of DLF (ablation)FedDR-LoRA · CIFAR-100, Pairflip 0.4accuracy (%): 87.37 -> 88.30Improved
add noisy branch negative learning on top of DLF (ablation)FedDR-LoRA · CIFAR-100, Symflip 0.2accuracy (%): 90.66 -> 91.37Improved
add DivideMix to FedLRFedLR · CIFAR-100, Pairflip 0.4accuracy (%): 60.11 -> 70.12Improved
add Relabel to FedLRFedLR · CIFAR-100, Symflip 0.2accuracy (%): 85.42 -> 75.31Hurt
add SCE loss to FedLRFedLR · CIFAR-100, Pairflip 0.4accuracy (%): 60.11 -> 66.79Improved
aggregate only B (clean and noisy separately) and keep A local, as in FedDR-LoRA, versus aggregating A and BFedDR-LoRA · CIFAR-100accuracy (%)Improved
use FedDR-LoRA instead of baselines under instance-dependent noise 0.4FedDR-LoRA · CIFAR-100, instance-dependent noiseaccuracy (%): 72.44 -> 88.23Improved
use FedDR-LoRA under heterogeneous per-client noise ratesFedDR-LoRA · CIFAR-100, Pairflip heterogeneousaccuracy (%): 77.07 -> 88.33Improved
FedDR-LoRA on a text task (RoBERTa-base) with low noise Symflip 0.2FedDR-LoRA · 20 Newsgroups, Symflip 0.2accuracy (%): 66.42 -> 66.53Mixed

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