AI Paper Atlas

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

Federated Variational Preference Alignment with Gumbel-Softmax Prior for Personalized User Preferences

Jabin Koo, Hoyoung Kim, Minwoo Jang, Jungseul Ok

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Sparse and collaboration-graph personalization · 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
Use FedVPA-GP instead of FedBiscuitQwen-2 0.5B, 10 clients · HH-RLHFHelpful win-rate: 48.85 -> 66.45Improved
Use FedVPA-GP instead of FedBiscuitGemma-2B, 10 clients · HH-RLHFHarmless win-rate: 82.45 -> 96.34Improved
Add Federated Mixture Prior and orthogonal loss to FedVPLQwen-2 0.5B, 100 clients · HH-RLHFHelpful win-rate: 53.05 -> 55.18Improved
Add Federated Mixture Prior and orthogonal loss to FedVPLGemma-2B, 10 clients · HH-RLHFHelpful win-rate: 66.82 -> 73.21Improved
Increase clients from 10 to 100FedVPA-GP Qwen-2 0.5B · HH-RLHFHelpful win-rate: 66.45 -> 55.18Hurt
Increase clients from 10 to 100FedVPL Gemma-2B · HH-RLHFHarmless win-rate: 89.15 -> 80.42Hurt
Evaluate on unseen clients instead of seen clientsFedVPL · HH-RLHF unseen clientsHarmless win-rate: 83.82 -> 75.21Hurt
Evaluate on unseen clients instead of seen clientsFedVPA-GP · HH-RLHF unseen clientsHarmless win-rate: 94.25 -> 91.23Hurt
Evaluate on unseen clients instead of seen clientsFedBiscuit · HH-RLHF unseen clientsHelpful win-rate: 47.32 -> 47.62No clear effect
Use FedVPA-GP instead of FedBiscuit on unseen clientsunseen clients · HH-RLHF unseen clientsHelpful win-rate: 47.62 -> 63.16Improved
Use FedVPA-GP instead of FedBiscuit at 70/30 client splitQwen-2 0.5B, 10 clients · HH-RLHF client ratioHelpful win-rate: 49.12 -> 68.12Improved
Use FedVPA-GP instead of FedBiscuit at 20/80 client splitQwen-2 0.5B, 10 clients · HH-RLHF client ratioHarmless win-rate: 78.21 -> 90.32Improved

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