Swin-VIB

Swin-VIB is a knowledge-conflict control framework for retrieval-augmented LLMs that adaptively regulates external context injection with an information-bottleneck sliding-window mechanism to improve reliability and consistency in the wild.

Citation

If you find our work useful in your research, please consider citing:

@article{wang2025accommodate,
  title={Accommodate Knowledge Conflicts in Retrieval-augmented LLMs: Towards Reliable Response Generation in the Wild},
  author={Wang, Jiatai and Xu, Zhiwei and Jin, Di and Yang, Xuewen and Li, Tao},
  journal={arXiv preprint arXiv:2504.12982},
  year={2025}
}