AI Paper Atlas

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

Generalized and Personalized Federated Learning with Black-Box Foundation Models via Orthogonal Transformations

Eun Gyung Kong, Jewon Yeom, Yonghoon Jeon, Taesup Kim

CVPR 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated learning with pretrained models · 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 orthogonality constraint on local transform (vs FedLT) (ablation)FedOT · FEMNISTaverage generalization accuracy (%): 88.74 -> 94.93Improved
add orthogonality constraint on local transform (vs FedLT) (ablation)FedOT · VLCSpseudo-gradient cosine similarity: -0.084 -> 0.002Improved
use orthogonal linear instead of nonlinear MLP adapter (ablation)FedOT · VLCSpseudo-gradient cosine similarity: -0.307 -> 0.002Improved
add orthogonality constraint on local transform (vs FedLT) (ablation)FedOT · 5-dataset averagecomprehensive accuracy (C, %): 84.24 -> 83.65Hurt
add orthogonality constraint on local transform (vs FedLT) (ablation)FedOT · TerraIncognitaaverage generalization accuracy (%): 29.87 -> 29.58No clear effect
add block-diagonal orthogonal transformFedOT · 5-dataset averagecomprehensive accuracy (C, %): 83.65 -> 86.10Improved
add client-specific local orthogonal transforms to all-global model (ablation)FedOT · 5-dataset averagepersonalization accuracy (P, %): 84.09 -> 86.21Improved
add global classifier sharing to purely local training (ablation)FedOT(+B) · 5-dataset averagepersonalization accuracy (P, %): 84.54 -> 88.58Improved
random classifier init instead of text-encoder init (ablation)FedOT · FEMNISTcomprehensive accuracy (C, %): 95.84 -> 96.51Improved
random classifier init instead of text-encoder init (ablation)FedOT · VLCScomprehensive accuracy (C, %): 77.13 -> 74.23Hurt
scale participating clients from 1 to 75FedOT · FEMNIST cross-devicegeneralization accuracy (%): 44.00 -> 71.18Improved
increase participating clients from 40 to 75PromptFL · FEMNIST cross-devicegeneralization accuracy (%): 73.54 -> 65.30Hurt

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