MANCE: Manifold-Aware Concept Erasure
yanaiela · x · 2026-07-14
The paper introduces MANCE (Manifold-aware Concept Erasure). The core idea is to first estimate the data manifold using natural input representations, then constrain model modifications to remain on this manifold, thereby erasing concepts without damaging others.
The authors note the method works well in practice, usable either standalone or as a plugin for existing erasure techniques like INLP and LEASE. Experiments span 119 settings across text, vision, multiple concepts, and LLM/VLM. Results show that concepts like sycophancy, gender, 安全相关概念, and various visual concepts can be "surgically" removed from model activations with minimal collateral damage.
Related event: MANCE: A Novel Manifold-Aware Concept Erasure Method(3 posts)→
More from Research
- Structural ensembles beat single predictions in TCR:pMHC generalization study — quaidmorris · 2026-07-22
- Structural ensembles, not single predictions, drive robust TCR:pMHC generalization — quaidmorris · 2026-07-22
- enFoldX turns AlphaFold3 ensemble noise into a TCR–peptide–MHC predictor — quaidmorris · 2026-07-22
- enFoldX tops 8 neoantigen scans and an unseen-peptide benchmark — quaidmorris · 2026-07-22
- enFoldX reaches AUC 0.82 on human VDJdb and transfers to mouse at 0.76 — quaidmorris · 2026-07-22
- enFoldX extracts 106 interface and confidence features from AF3 ensembles — quaidmorris · 2026-07-22