Roy Xie

Ph.D. Student in Computer Science
Duke University
Duke Natural Language Processing Group

ruoyu.xie@duke.edu

I am a final-year Ph.D. student at Duke University, advised by Bhuwan Dhingra. My research focuses on post-training methods for foundation models. I am particularly interested in improving models’ agentic capabilities and alignment on complex long-horizon tasks.

My research is partially supported by the National Science Foundation Graduate Research Fellowship and the Apple Scholars in AI/ML Ph.D. Fellowship. I am currently interning at Google Research. Previously, I interned with the Apple Foundation Models team.

Recent Publications

LensVLM: Selective Context Expansion for Compressed Visual Representation of Text

Roy Xie, Dan Friedman, Donghan Yu, Bowen Pan, Christopher Fifty, Jang-Hyun Kim, Xianzhi Du, Zhe Gan, Vivek Rathod, Bhuwan Dhingra

NeurIPS 2026

Paper Code Models

@misc{xie2026lensvlmselectivecontextexpansion,
      title={LensVLM: Selective Context Expansion for Compressed Visual Representation of Text},
      author={Roy Xie and Dan Friedman and Donghan Yu and Bowen Pan and Christopher Fifty and Jang-Hyun Kim and Xianzhi Du and Zhe Gan and Vivek Rathod and Bhuwan Dhingra},
      year={2026},
      eprint={2605.07019},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2605.07019},
}

Over-Searching in Search-Augmented Large Language Models

Roy Xie, Deepak Gopinath, David Qiu, Dong Lin, Haitian Sun, Saloni Potdar, Bhuwan Dhingra

EACL 2026 (Oral)

Paper Code

@misc{xie2026oversearchingsearchaugmentedlargelanguage,
      title={Over-Searching in Search-Augmented Large Language Models},
      author={Roy Xie and Deepak Gopinath and David Qiu and Dong Lin and Haitian Sun and Saloni Potdar and Bhuwan Dhingra},
      year={2026},
      eprint={2601.05503},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2601.05503},
}

When Greedy Wins: Emergent Exploitation Bias in Meta-Bandit LLM Training

Sanxing Chen, Xiaoyin Chen, Yukun Huang, Roy Xie, Bhuwan Dhingra

ICLR 2026

Paper Code

@misc{chen2025greedywinsemergentexploitation,
      title={When Greedy Wins: Emergent Exploitation Bias in Meta-Bandit LLM Training},
      author={Sanxing Chen and Xiaoyin Chen and Yukun Huang and Roy Xie and Bhuwan Dhingra},
      year={2025},
      eprint={2509.24923},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2509.24923},
}

Knowing When to Stop: Efficient Context Processing via Latent Sufficiency Signals

Roy Xie, Junlin Wang, Paul Rosu, Chunyuan Deng, Bolun Sun, Zihao Lin, Bhuwan Dhingra

NeurIPS 2025

Paper Code Webcite

@inproceedings{NEURIPS2025_3afa7f9b,
      title={Knowing When to Stop: Efficient Context Processing via Latent Sufficiency Signals},
      author={Xie, Roy and Wang, Junlin and Rosu, Paul and Deng, Chunyuan and Sun, Bolun and Lin, Zihao and Dhingra, Bhuwan},
      booktitle={Advances in Neural Information Processing Systems},
      volume={38, Main Conference},
      pages={41213--41240},
      publisher={Curran Associates, Inc.},
      year={2025},
      doi={10.52202/085713-1376},
      url={https://proceedings.neurips.cc/paper_files/paper/2025/hash/3afa7f9b7ac8da474b3d915570d58291-Abstract-Conference.html},
}

Improving Model Alignment Through Collective Intelligence of Open-Source Models

Junlin Wang, Roy Xie, Shang Zhu, Jue Wang, Ben Athiwaratkun, Bhuwan Dhingra, Shuaiwen Leon Song, Ce Zhang, James Zou

ICML 2025

Paper Webcite

@inproceedings{pmlr-v267-wang25dr,
      title={Improving Model Alignment Through Collective Intelligence of Open-Source Models},
      author={Wang, Junlin and Xie, Roy and Zhu, Shang and Wang, Jue and Athiwaratkun, Ben and Dhingra, Bhuwan and Song, Shuaiwen Leon and Zhang, Ce and Zou, James},
      booktitle={Proceedings of the 42nd International Conference on Machine Learning},
      series={Proceedings of Machine Learning Research},
      volume={267},
      pages={64976--64997},
      publisher={PMLR},
      year={2025},
      url={https://proceedings.mlr.press/v267/wang25dr.html},
}