Research map / Image generation / Text-to-image diffusion
OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasure
Chuancheng Shi, Wenhua Wu, Fei Shen, Xiaogang Zhu, Kun Hu, Zhiyong Wang
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 |
|---|---|---|---|
| select the global optimal layer 10 instead of layer 9 (ablation) | OrthoEraser · I2P nudity | total detections: 17 -> 5 | Improved |
| select the global optimal layer instead of first sensitivity peak layer 3 (ablation) | OrthoEraser · I2P nudity | total detections: 24 -> 5 | Improved |
| suppress identified sensitive neurons instead of random neurons of same size (ablation) | OrthoEraser · I2P nudity | total detections: 598 -> 5 | Improved |
| add coupled-neuron orthogonal projection to sensitive suppression (ablation) | OrthoEraser · I2P nudity | total detections: 17 -> 5 | Improved |
| add coupled-neuron orthogonal projection to sensitive suppression (ablation) | OrthoEraser · MS COCO-30K | FID: 16.64 -> 1.15 | Improved |
| add coupled-neuron orthogonal projection to sensitive suppression (ablation) | OrthoEraser · MS COCO-30K | CLIP score: 30.87 -> 31.33 | Improved |
| add sensitive-neuron suppression to coupled-only suppression (ablation) | OrthoEraser · I2P nudity | total detections: 604 -> 5 | Improved |
| use orthogonal projection instead of coupled-aligned suppression (ablation) | OrthoEraser · MS COCO-30K | FID: 23.95 -> 1.15 | Improved |
| use orthogonal projection instead of coupled-aligned suppression (ablation) | OrthoEraser · MS COCO-30K | CLIP score: 26.31 -> 31.33 | Improved |
| apply OrthoEraser vs SNCE on nudity | SNCE baseline · I2P nudity | total detections: 17 -> 5 | Improved |
| extend to violence concept | OrthoEraser · I2P Violence | detection rate: 40.1 -> 15.6 | Improved |
| apply to adversarial prompts Ring-A-Bell | OrthoEraser · Ring-A-Bell | attack success rate: 98.7 -> 2.7 | Improved |
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.
- DuMo — DuMo: Dual Encoder Modulation Network for Precise Concept Erasure (AAAI 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
- Minimalist Concept Erasure in Generative Models (ICML 2025 · similarity 0.86)
- EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignment (NeurIPS 2025 · similarity 0.84)
- STEREO: A Two-Stage Framework for Adversarially Robust Concept Erasing from Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.83)
- ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning (ICLR 2025 · similarity 0.83)
- CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models (NeurIPS 2025 · similarity 0.83)
- Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models (NeurIPS 2024 · similarity 0.83)
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.
