Research map / Image generation / Text-to-image diffusion
Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron Masking
Kaiyuan Deng, Bo Hui, Gen Li, Jie Ji, Minghai Qin, Geng Yuan, Xiaolong Ma
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 |
|---|---|---|---|
| use FIA masking instead of fine-tuning-based unlearning (ESD, SalUn, SPM, MACE) | SD v1.5 · Imagenette 10-class unlearning | avg forgetting accuracy (%): 1.9 vs 38.58 (ESD), 23.17 (SalUn), 78.22 (MACE) | Improved |
| use FIA instead of fine-tuning methods in terms of image quality | SD v1.5 · MS COCO-30K CLIP after 10-class unlearning | CLIP score: 29.56 vs 31.05 (MACE), 31.42 (SD v1.5) | Hurt |
| use FIA instead of training-free CP | SD v1.5 · Imagenette 10-class unlearning | CLIP score: 27.93 -> 29.56 | Improved |
| use FIA instead of CP for preservation | SD v1.5 · Imagenette forget 5 / preserve 5 | preserving accuracy (%): 52.4 -> 76.7 | Improved |
| use FIA instead of UCE | SD v1.5 · Imagenette forget 5 / preserve 5 | overall score (%): 81.7 -> 86.0 | Improved |
| increase number of forgotten classes (forget X / preserve 10-X) (ablation) | FIA · Imagenette | forgetting accuracy | No clear effect |
| apply FIA to SD v1.4 for explicit content | SD v1.4 · I2P NudeNet | total detections: 743 -> 32 | Improved |
| apply FIA to SD v1.4 | SD v1.4 · MS COCO-30K | FID: 14.04 -> 14.02 | No clear effect |
| apply FIA to forget five artist styles | SD v1.5 · Five artist styles | CLIPa: 42.10 -> 27.45 | Improved |
| apply FIA to forget five artist styles | SD v1.5 · MS COCO-30K | FID: 14.5 -> 16.7 | 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.
- SalUn — SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (ICLR 2024) · better on 2 of 2 shared results
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
- Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient (AAAI 2025 · similarity 0.91)
- ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning (ICLR 2025 · similarity 0.90)
- Machine Unlearning for Image-to-Image Generative Models (ICLR 2024 · similarity 0.89)
- Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models (NeurIPS 2024 · similarity 0.89)
- Minimalist Concept Erasure in Generative Models (ICML 2025 · similarity 0.89)
- You Can’t Have It Both Ways: Concept Entanglement Limits Diffusion Model Unlearning (NeurIPS 2026 · similarity 0.89)
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.
