Research map / Label-efficient and robust learning / Recognition and long-tail
Robust Concept Erasure via Kernelized Rate-Distortion Maximization
Somnath Basu Roy Chowdhury, Nicholas Monath, Kumar Avinava Dubey, Amr Ahmed, Snigdha Chaturvedi
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 |
|---|---|---|---|
| Apply KRaM concept erasure | BERT representations · DIAL race probing | Acc. (a): 87.7 -> 54.0 | Improved |
| Apply KRaM concept erasure | GloVe embeddings · gender probing | Acc. (a): 100.0 -> 52.6 | Improved |
| Apply KRaM concept erasure | BERT representations · DIAL sentiment task | Acc. (y): 75.5 -> 72.4 | Hurt |
| Apply KRaM concept erasure | UCI Crimes representations · UCI Crimes | delta GDP: 0.058 -> 0.001 | Improved |
| Use a linear erasure function instead of non-linear (ablation) | KRaM · Synthetic | alignment Ak: 0.67 -> 0.75 | Improved |
| Use a linear erasure function instead of non-linear (ablation) | KRaM · GloVe gender | Acc. (a): 52.6 -> 67.0 | Hurt |
| Swap kernel from Gaussian to Laplace (ablation) | KRaM · Synthetic | MSE (a): 0.109 -> 0.083 | Hurt |
| Swap kernel from Gaussian to Cauchy (ablation) | KRaM · Synthetic | MSE (a): 0.109 -> 0.092 | Hurt |
| Increase lambda penalty on |R(Z)-b| (ablation) | KRaM · GloVe gender | gender accuracy | Hurt |
| Increase scaling factor r in R(Z)=rb (ablation) | KRaM · GloVe gender | Ak and gender accuracy | Mixed |
| Vary Gaussian kernel sigma (ablation) | KRaM · Synthetic | MSE (a) | Mixed |
| Replace rate-distortion objective with MI estimators (InfoNCE, MINE, CLUB, KNIFE) (ablation) | KRaM · GloVe gender | Ak: 0.65 -> 0.53 | Hurt |
Closest papers by meaning
- Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptation (CVPR 2025 · similarity 0.85)
- Consistency-Preserving Concept Erasure via Unsafe–Safe Pairing and Directional Fisher-weighted Adaptation (NeurIPS 2026 · similarity 0.85)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.84)
- All but One: Surgical Concept Erasing with Model Preservation in Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.84)
- CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models (NeurIPS 2025 · similarity 0.84)
- Obliviator Reveals the Cost of Nonlinear Guardedness in Concept Erasure (NeurIPS 2025 · similarity 0.84)
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.
