Lune

CVPR Findings2026

Onboarding Without Forgetting: Hypernetwork Personalization with Data-Free Replay for Personalized Federated Learning

Thinh Nguyen, Le Huy Khiem, Van-Tuan Tran, Khoa D Doan, Nitesh V. Chawla, Kok-Seng Wong

Abstract

Federated Learning (FL) enables collaborative training across distributed clients without sharing raw data, offering strong privacy benefits. However, most methods assume all clients remain available throughout training, which is unrealistic as new clients often join over time. We study this setting, where the task and label space stay fixed but clients arrive in batches. Our analysis reveals two key challenges: updating the shared model only with new clients harms existing clients, while freezing it protects them but blocks gains from new knowledge. To capture these tradeoffs, we introduce Proactive Adaptation (PA) for onboarding gains and Retroactive Improvement (RI) for changes in earlier clients without retraining. We then propose pFed-DSH, which combines a central hypernetwork for personalized initialization, batch-specific binary masks for capacity preservation and allocation, and server-side data-free replay to propagate improvements without exposing client data. Experiments show that pFedDSH preserves stability for existing clients while keeping communication and adaptation costs unchanged for new clients.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Ditto: Fair and Robust Federated Learning Through Personalization cites Personalized Federated Learning with Moreau Envelopes
  • Exploiting Shared Representations for Personalized Federated Learning cites Ditto: Fair and Robust Federated Learning Through Personalization
  • Personalized Federated Learning using Hypernetworks cites Personalized Federated Learning with Moreau Envelopes
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on9

  1. Personalized Federated Learning with Moreau EnvelopesCanh T. Dinh, Nguyen Hoang Tran, Tuan Dung NguyenNeurIPS 2020 · 1,542 citations
  2. Ditto: Fair and Robust Federated Learning Through PersonalizationTian Li, Shengyuan Hu, Ahmad Beirami, Virginia SmithICML 2021 · 1,313 citations
  3. Exploiting Shared Representations for Personalized Federated LearningLiam Collins, Hamed Hassani, Aryan Mokhtari, Sanjay ShakkottaiICML 2021 · 1,081 citations
  4. Personalized Federated Learning using HypernetworksAviv Shamsian, Aviv Navon, Ethan Fetaya, Gal ChechikICML 2021 · 452 citations
  5. Federated Continual Learning with Weighted Inter-client TransferJaehong Yoon, Wonyong Jeong, Giwoong Lee, Eunho Yang et al.ICML 2021 · 303 citations

Related papers

  1. Towards Stable Federated Continual Test-Time Adaptation in Wild WorldLiwen Wang, Xingbo Dong, Iman Yi Liao, Zhe JinCVPR 2026
  2. Better Generative Replay for Continual Federated LearningDaiqing Qi, Handong Zhao, Sheng LiICLR 2023 · 13 citations
  3. Deploying Models to Non-participating Clients in Federated Learning without Fine-tuning: A Hypernetwork-based ApproachYuhao Zhou, Jindi Lv, Yuxin Tian, Dan Si et al.ICLR 2026 · 3 citations
  4. PFA: Privacy-preserving Federated Adaptation for Effective Model PersonalizationBingyan Liu, Yao Guo, Xiangqun ChenWWW 2021 · 118 citations
  5. Lightweight Federated Incremental Learning via Decoupled ReplayXiuying Wang, Yichen Li, Hang Su, Gaozhuo Liu et al.ICML 2026
Dusk over the sea between two cliffs drawn in fine vertical lines