Lune

CVPR Findings2026

FedErase: Personalized Federated Unlearning for Text-to-Image Diffusion Models

Tianyu Geng, Wenfei Liang, Sijie Wang, Rui She, Wee Peng Tay

Abstract

Text-to-image diffusion models have achieved remarkable success in generating high-quality images from natural language descriptions. However, their large-scale training data often include sensitive, copyrighted, or biased content, raising serious privacy and ethical concerns. Privacy regulations such as the GDPR further mandate the ability to erase user-specific data from trained models, motivating the need for effective unlearning mechanisms. While existing machine unlearning approaches primarily focus on centralized classification models, federated unlearning (FU) for generative tasks remains largely unexplored. In this work, we present FedErase, a family of personalized federated unlearning (PFU) approaches for text-to-image diffusion. FedErase instantiates several design paradigms: (i) representation-head decoupling for personalization (FedErase-R), (ii) hypernetwork-driven parameter generation (FedErase-H), (iii) gradient-based unlearning compatible with standard federated aggregation (FedErase-G), and (iv) sparse parameter selection for precise forgetting (FedErase-S). Extensive experiments across diverse concepts under non-IID client distributions verify that FedErase enables client-specific forgetting while maintaining image quality, and offers new insights into the trade-offs between personalization, stability, and efficiency in PFU for diffusion models.

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

  • High-Resolution Image Synthesis with Latent Diffusion Models cites Denoising Diffusion Probabilistic Models
  • Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding cites High-Resolution Image Synthesis with Latent Diffusion Models
  • Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding cites Denoising Diffusion Probabilistic Models
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on20

  1. Denoising Diffusion Probabilistic ModelsJonathan Ho, Ajay Jain, Pieter AbbeelNeurIPS 2020 · 35,902 citations
  2. High-Resolution Image Synthesis with Latent Diffusion ModelsRobin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser et al.CVPR 2022 · 13,123 citations
  3. Photorealistic Text-to-Image Diffusion Models with Deep Language UnderstandingChitwan Saharia, William Chan, Saurabh Saxena, Lala Li et al.NeurIPS 2022 · 8,965 citations
  4. An Efficient Framework for Clustered Federated LearningAvishek Ghosh, Jichan Chung, Dong Yin, Kannan RamchandranNeurIPS 2020 · 1,329 citations
  5. Exploiting Shared Representations for Personalized Federated LearningLiam Collins, Hamed Hassani, Aryan Mokhtari, Sanjay ShakkottaiICML 2021 · 1,081 citations

Related papers

  1. FORGET ME: Federated Unlearning for Face Generation ModelsFan Qi, Ao Liu, Zixin Zhang, Changsheng XuACM MM 2025 · 1 citation
  2. Incentive and Dynamic Client Selection for Federated UnlearningYijing Lin, Zhipeng Gao, Hongyang Du, Dusit Niyato et al.WWW 2024 · 28 citations
  3. Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron MaskingKaiyuan Deng, Bo Hui, Gen Li, Jie Ji et al.ICML 2026 · 1 citation
  4. GenErase: Generalizable and Semantically-Aware Concept Erasure in Diffusion ModelsKorada Sri Vardhana, Soma BiswasCVPR 2026
  5. Boosting Alignment for Post-Unlearning Text-to-Image Generative ModelsMyeongseob Ko, Henry Li, Zhun Wang, Jonathan Patsenker et al.NeurIPS 2024 · 22 citations
Dusk over the sea between two cliffs drawn in fine vertical lines