analiza aiqbench
This research addresses the critical gap where standard unlearning techniques fail when Large Language Models are quantized for deployment, proposing a framework that maintains data privacy without sacrificing model utility. By leveraging loss landscape curvature analysis, the authors introduce sensitivity-guided regularization and forget-critical optimization to ensure robust forgetting across different compression levels. The work provides a necessary advancement for practical AI compliance, ensuring that unlearning remains effective in real-world scenarios where models are inevitably compressed.
Quantization-Robust Unlearning through the Lens of Retain-Forget Loss Landscapes Interaction
arXiv:2609.27355v1 Announce Type: cross Abstract: Unlearning ensures LLM compliance by removing the influence of private or copyrighted training data. However, since LLM models typically undergo post-training compression, like quantization, in practical deployment, it has been observed that the unlearning effect can be substantially weakened, with the forgetting behavior degrading more severely than that of model utility. This paper proposes a quantization-robust unlearning framework that makes forgetting robust to quantization while maintaining overall model utility. We analyze this gap through the lens of loss landscape. Specifically, our analysis reveals a curvature-based criteria that pinpoints sensitive weights in the unlearned model that leads to both non-robust forgetting and reduced utility. We therefore propose sensitivity-guided noisy regularization, which is applied on the sensitive parameters to steer the model convergence towards a smoother minima of uniformly low forget and retain losses. Balancing unlearning and utility, we further propose forget-critical optimization, which updates only forget-critical layers, preserving most of the network to retain useful knowledge. Extensive experiments on the MUSE and TOFU benchmarks across multiple LLM unlearning algorithms show that our approach achieves substantially more quantization-resilient forgetting while maintaining utility.
Abstract:Unlearning ensures LLM compliance by removing the influence of private or copyrighted training data. However, since LLM models typically undergo post-training compression, like quantization, in practical deployment, it has been observed that the unlearning effect can be substantially weakened, with the forgetting behavior degrading more severely than that of model utility. This paper proposes a quantization-robust unlearning framework that makes forgetting robust to quantization while maintaining overall model utility. We analyze this gap through the lens of loss landscape. Specifically, our analysis reveals a curvature-based criteria that pinpoints sensitive weights in the unlearned model that leads to both non-robust forgetting and reduced utility. We therefore propose sensitivity-guided noisy regularization, which is applied on the sensitive parameters to steer the model convergence towards a smoother minima of uniformly low forget and retain losses. Balancing unlearning and utility, we further propose forget-critical optimization, which updates only forget-critical layers, preserving most of the network to retain useful knowledge. Extensive experiments on the MUSE and TOFU benchmarks across multiple LLM unlearning algorithms show that our approach achieves substantially more quantization-resilient forgetting while maintaining utility.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2609.27355 [cs.LG]
(or arXiv:2609.27355v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.27355 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 277 cuvinte
Subiecte conexe
Reflections on unlearning and inoculation
LessWrong · 20 sept. 2026
TULIP: Targeted LLM Unlearning at Layers Identified Per-Input
arXiv · 29 sept. 2026
Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMs
arXiv · 11 sept. 2026
Red-teaming LLM unlearning: LUNAR's "forgotten" knowledge is still recoverable
LessWrong · 23 iul. 2026