Research map / Image generation / Text-to-image diffusion
A Unified Framework for Diffusion Model Unlearning with f-Divergence
Nicola Novello, Federico Fontana, Luigi Cinque, Deniz Gunduz, Andrea M Tonello
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace MSE (KL) loss with closed-form Hellinger loss (ablation) | diffusion unlearning (CAbl-style) · object and style erasure, SD 1.4/2.1 | prior preservation and KID: KID 0.250 -> 0.005 (Grumpy Cat, erasing Dali) | Improved |
| replace MSE with Hellinger loss (ablation) | diffusion unlearning · nudity erasure I2P, SD 1.5 | nudity generation rate: 0.120 -> 0.063 | Improved |
| replace MSE with Hellinger loss (ablation) | diffusion unlearning · Ring-A-Bell adversarial prompts, SD 1.5 | nudity generation rate: 0.063 -> 0.157 | Hurt |
| replace MSE with chi2 loss (ablation) | diffusion unlearning · object and style erasure | erasure strength vs preservation | Mixed |
| replace MSE with chi2 loss (ablation) | diffusion unlearning · nudity erasure I2P, SD 1.5 | nudity generation rate: 0.120 -> 0.204 | Hurt |
| use Hellinger instead of MSE for sequential erasure of 10 styles (ablation) | diffusion unlearning · 10 artistic styles, SD 1.4 | prior preservation and erasure | Improved |
| use Hellinger loss with bounded gradients (ablation) | diffusion unlearning · fine-tuning stability | generation stability | Improved |
| use variational min-max loss instead of closed-form (ablation) | Hellinger f-DMU · Van Gogh erasure, SD 2.1 | KID (preserved): 0.027 -> 0.051 | Hurt |
| use variational loss instead of closed-form (ablation) | f-DMU · unlearning speed | iterations to low CS and CA | Improved |
| use Hellinger variational loss instead of DoCo | variational f-DMU · Van Gogh erasure, SD 2.1 | preserved CLIP accuracy and erasure | Improved |
| use Hellinger variational loss instead of Hellinger closed-form for nudity (ablation) | f-DMU · I2P, SD 1.5 | nudity generation rate: 0.063 -> 0.430 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAFREE — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models (NeurIPS 2026 · similarity 0.91)
- Robust Diffusion Models via Divergence-Induced Weighted Denoising (NeurIPS 2026 · similarity 0.91)
- Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient (AAAI 2025 · similarity 0.90)
- The Illusion of Unlearning: The Unstable Nature of Machine Unlearning in Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.89)
- Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models (CVPR 2025 · similarity 0.89)
- CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models (NeurIPS 2025 · similarity 0.88)
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.
