AI Papers Daily

#Circuit Attribution

1 篇

Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution

量化永久遗忘：通过电路归因实现抗压缩的机器遗忘

🧭 LLM ★ 深度解读

Lexsi Labs

提出 MANSU 框架，利用电路归因与量级底线保证 4-bit 量化后遗忘不逆转

LLM | 2026-05-14 | cs.LG | score 8/10

Machine UnlearningQuantizationCircuit AttributionLLMSafety