Focus-LIME: Surgical Interpretation of Long-Context Large Language Models via Proxy-Based Neighborhood Selection
Published in arXiv preprint arXiv:2602.04607, 2026
Long-context LLM explanations suffer from attribution dilution when local model-agnostic methods perturb too many features.
Focus-LIME uses a proxy model to select a compact perturbation neighborhood, then performs fine-grained attribution on the target model, making surgical long-context explanations more practical and faithful.
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. CoRR abs/2602.04607 (2026). PDF