analiza aiqbench
This paper introduces GLIME, a novel framework for lifelong knowledge editing in large language models that addresses the common issue of overfitting during continual updates. By integrating preference optimization with replay-based techniques and gradient constraints, it successfully preserves both newly edited facts and the model's general capabilities without full retraining. The approach represents a significant step toward maintaining robust performance in dynamic environments where models require frequent factual corrections.
Generalizable Lifelong Model Editing via Preference Optimization
arXiv:2609.36748v1 Announce Type: new Abstract: Knowledge editing enables rapid updates of specific factual knowledge in large language models (LLMs) without full retraining. However, more realistic scenarios call for a lifelong framework that handles continual updates rather than one-off modifications. In such settings, existing editing methods often overfit to target prompts, significantly degrading both the generalization of the edited knowledge and the model's general capabilities. To address this issue, we propose GLIME (Generalizable Lifelong Model Editing), which combines knowledge editing with preference optimization over generation behavior. GLIME further incorporates replay-based editing and a gradient constraint to preserve previously edited knowledge. Experimental results show that GLIME significantly improves knowledge generalization in lifelong editing settings while maintaining both editing performance and general capabilities.
Abstract:Knowledge editing enables rapid updates of specific factual knowledge in large language models (LLMs) without full retraining. However, more realistic scenarios call for a lifelong framework that handles continual updates rather than one-off modifications. In such settings, existing editing methods often overfit to target prompts, significantly degrading both the generalization of the edited knowledge and the model's general capabilities. To address this issue, we propose GLIME (Generalizable Lifelong Model Editing), which combines knowledge editing with preference optimization over generation behavior. GLIME further incorporates replay-based editing and a gradient constraint to preserve previously edited knowledge. Experimental results show that GLIME significantly improves knowledge generalization in lifelong editing settings while maintaining both editing performance and general capabilities.
Artificial Intelligence (cs.AI)
arXiv:2609.36748 [cs.AI]
(or arXiv:2609.36748v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.36748 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 · 209 cuvinte
Subiecte conexe
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
LessWrong · 15 oct. 2026
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · 12 iun. 2026
Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs
arXiv · 15 oct. 2026
Bridging intent and execution in agentic systems
Amazon Science · 8 iun. 2026