Continual Concept Erasure in Diffusion Models by Suppressing Cross-Edit Interference

2026-10-01

AI Summary

This study addresses the problem of accumulated editing interference in continual concept erasure for diffusion models, which leads to the failure of previously erased concepts and degradation of generation quality. To this end, we propose CEASE, a method that introduces a shared replacement invariance matrix and an orthogonal projection mechanism to theoretically decompose and suppress cross-edit interference arising from repeated activations and directional overlaps. By imposing subspace constraints via a closed-form solver, CEASE achieves training-free continual concept erasure. Experimental results demonstrate that CEASE attains state-of-the-art erasure-preservation trade-offs across sequential erasure tasks involving celebrities and artistic styles, significantly outperforming existing methods.

Abstract

Concept erasure removes copyright-protected, privacy-sensitive, or otherwise undesirable concepts from pretrained text-to-image diffusion models to support content governance and compliance. As erasure requests arrive over time, models must remove new targets without undoing prior erasures. Existing methods do not constrain interference across edits: residual perturbations outside the retain set interact and accumulate, degrading unrelated generations and sometimes collapsing previously erased targets into noise. We propose CEASE (Continual Erasure via Adaptive Subspace Editing), a training-free method that imposes two subspace constraints on a closed-form solver. CEASE adds the token representation of the shared replacement to the solver's invariance matrix and, when interference is detected, projects the current update onto the orthogonal complement of dominant output directions extracted from cumulative past updates. A closed-form decomposition attributes the accumulated interference to repeated activation of the shared replacement and overlap between successive update directions, showing that the two constraints suppress these respective sources. Across continual erasure of celebrities, artistic styles, and instances, CEASE achieves the most consistent erase-preserve trade-off, while existing methods either degrade general generation or insufficiently erase targets.

Problem

• Continual Concept Erasure
• Diffusion Models
• Cross-Edit Interference
• Content Governance

Innovation

• Continual Concept Erasure
• Diffusion Models
• Training-free
• Subspace Editing
• Cross-Edit Interference