ECCV Paper: Locality-Aware Continual Unlearning for Diffusion Models
mittu1204 · x · 2026-08-31
To address the need for continual concept removal in deployed diffusion models (privacy, copyright, safety), this paper proposes Locality-Aware Continual Unlearning (LACU). Existing methods collapse after 3-5 sequential uses due to coarse targets and lack of protection for neighboring concepts. LACU introduces Locality-Aware Target Selection (picking semantically closest retain prompts) and Locality-Aware Replay (replaying nearest neighbors) to mitigate degradation, enabling stable continual unlearning.
More from Safety
- Next AI swarms might hide presence long-term, poison future models — nabeelqu · 2026-08-31
- Critique on cyber attack post: Avoid anthropomorphism, open source is vital — sriramk · 2026-08-31
- Models might takeover for myopic reasons, like the OpenAI infrastructure incident — nabeelqu · 2026-08-31
- Reward hacking is pervasive in production; models lack truth-orientation — nabeelqu · 2026-08-31
- Matt Shumer calls to boycott 'Infinite TikTok', labeling it digital fentanyl — mattshumer_ · 2026-08-31
- Blueprint: A Chat Pipeline for Hallucination Mitigation and Strict Moderation — Scorpowned · 2026-08-31