Focus-LIME: Surgical Interpretation of Long-Context Large Language Models via Proxy-Based Neighborhood Selection

Focus-LIME is a coarse-to-fine framework for surgical feature-level interpretation of long-context large language models. It uses a proxy model to curate the perturbation neighborhood, allowing the target model to perform fine-grained attribution only within the optimized context.

The paper was accepted at IJCAI-ECAI 2026 and is also available on arXiv.

Recommended citation: Junhao Liu, Haonan Yu, Zhenyu Yan, and Xin Zhang. (2026). "Focus-LIME: Surgical Interpretation of Long-Context Large Language Models via Proxy-Based Neighborhood Selection." IJCAI-ECAI 2026.
Download Paper