Research map / Image generation / Text-to-image diffusion
To Erase, or Not to Erase: Robust Training-Free Concept Erasure with Preservation aware Adaptive Ranked Subspace Expansion
Shaswati Saha, Rajasekhar Anguluri, Manas Gaur
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 |
|---|---|---|---|
| add adaptive subspace expansion (ASE) (ablation) | PARSE · Nudity erasure | CCE ASR: 24.00 -> 3.41 | Improved |
| add preservation-aware projection (PAP) (ablation) | PARSE · Nudity erasure | CCE ASR: 4.20 -> 3.41 | Improved |
| add preservation-aware projection (PAP) (ablation) | PARSE · Nudity erasure | FID: 15.20 -> 14.24 | Improved |
| add knowledge search stage 1 and 2 (ablation) | PARSE · Nudity erasure | CCE ASR: 62.50 -> 3.41 | Improved |
| use preservation-aware projection instead of AdaVD value projection | PARSE · Nudity erasure | CCE ASR: 86.28 -> 3.41 | Improved |
| discover erase concepts via diffusion CFG rather than CLIP proxy space | PARSE · Concept erasure | ASR | Improved |
| add PARSE over STEREO on nudity FID | PARSE · Nudity erasure | FID: 16.71 -> 14.24 | Improved |
| add PARSE against SD 1.4 on nudity target ASR | SD 1.4 · Nudity erasure | Target ASR: 65.23 -> 2.09 | Improved |
| add PARSE to erase Van Gogh style | SD 1.4 · Van Gogh style erasure | CCE ASR: 65.50 -> 6.02 | Improved |
| add PARSE for multi-concept erasure | PARSE · multi-concept erasure | BEUS | Improved |
Closest papers by meaning
- Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters (CVPR 2025 · similarity 0.93)
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.93)
- Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptation (CVPR 2025 · similarity 0.93)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.92)
- All but One: Surgical Concept Erasing with Model Preservation in Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.92)
- Consistency-Preserving Concept Erasure via Unsafe–Safe Pairing and Directional Fisher-weighted Adaptation (NeurIPS 2026 · similarity 0.91)
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.
