publications

2026

1. arXiv

   Why On-Policy Distillation Sometimes Fails: Vanishing Learning Signals

   Lei Zhao, Qichao Zhao, Bowen Zuo, and 1 more author

   2026

   Abs arXiv Bib HTML

   Studies why on-policy distillation (OPD) fails to transfer capability between language models, finding early loss plateaus with larger-scale teachers on code generation and math reasoning.

   @misc{zhao2026opd,
     title = {Why On-Policy Distillation Sometimes Fails: Vanishing Learning Signals},
     author = {Zhao, Lei and Zhao, Qichao and Zuo, Bowen and Zhan, Qishi},
     howpublished = {arXiv preprint arXiv:2610.11247},
     year = {2026},
   }

2. arXiv

   UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement

   Fang Wu, Da Xing, Yanjie Huang, and 16 more authors

   2026

   Abs arXiv Bib HTML

   Introduces UniEvo-VL, a self-evolving framework for unified multimodal models that learn from their own self-correction feedback.

   @misc{wu2026unievo,
     title = {UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement},
     author = {Wu, Fang and Xing, Da and Huang, Yanjie and Wang, Junxi and Wang, Ji and Geng, Hejia and Wan, Guancheng and Zuo, Bowen and Li, Xiaomin and Tang, Shixiang and Xiang, Xinyu and Wang, Zehong and Du, Shiyi and Xia, Peng and Zheng, Shuangjia and Hong, Yining and Li, Li Erran and Leskovec, Jure and Choi, Yejin},
     howpublished = {arXiv preprint arXiv:2609.38721},
     year = {2026},
   }

3. NAR Workshop

   Exploring More, Reasoning Better: Stepwise Risk-Sensitive GRPO for Diffusion Language Models

   Y. Yu, Bowen Zuo, D. J. Crandall, and 2 more authors

   Workshop on Non-Autoregressive Language Models for Fast and Flexible Text Generation, 2026

   Abs Bib

   Studies risk-sensitive policy optimization for diffusion language models, applying stepwise GRPO to improve exploration and reasoning performance.

   @article{zuo2026stepwise,
     title = {Exploring More, Reasoning Better: Stepwise Risk-Sensitive GRPO for Diffusion Language Models},
     author = {Yu, Y. and Zuo, Bowen and Crandall, D. J. and Zhu, Yinglun and Zhou, D.},
     journal = {Workshop on Non-Autoregressive Language Models for Fast and Flexible Text Generation},
     year = {2026}
   }

4. ACL Findings

   Adaptive Test-Time Compute Allocation with Evolving In-Context Demonstrations

   Bowen Zuo, D. Zhou, and Yinglun Zhu

   Findings of the Association for Computational Linguistics (ACL), 2026

   Abs arXiv Bib HTML

   Introduces a test-time compute allocation framework that jointly adapts where computation is spent and how generation is performed, conditioning generation on successful responses from semantically related queries.

   @article{zuo2026adaptive,
     title = {Adaptive Test-Time Compute Allocation with Evolving In-Context Demonstrations},
     author = {Zuo, Bowen and Zhou, D. and Zhu, Yinglun},
     journal = {Findings of the Association for Computational Linguistics (ACL)},
     year = {2026},
   }

5. ICLR

   Strategic Scaling of Test-Time Compute: A Bandit Learning Approach

   Bowen Zuo and Yinglun Zhu

   In International Conference on Learning Representations (ICLR), 2026

   Abs arXiv Bib HTML

   Formulates test-time compute allocation as a bandit learning problem, proposing adaptive algorithms that estimate query difficulty on the fly and allocate more compute to harder queries.

   @inproceedings{zuo2026strategic,
     title = {Strategic Scaling of Test-Time Compute: A Bandit Learning Approach},
     author = {Zuo, Bowen and Zhu, Yinglun},
     booktitle = {International Conference on Learning Representations (ICLR)},
     year = {2026},
   }

2024

1. JINST

   Machine Learning Evaluation in the Global Event Processor FPGA for the ATLAS Trigger Upgrade

   Z. Jiang, B. Carlson, Bowen Zuo, and 1 more author

   Journal of Instrumentation (JINST), 2024

   Abs arXiv Bib HTML

   Presents methods to automatically create machine learning based algorithms for the ATLAS Global Event Processor FPGA, achieving microsecond-scale latencies with low resource utilization.

   @article{jiang2024machine,
     title = {Machine Learning Evaluation in the Global Event Processor FPGA for the ATLAS Trigger Upgrade},
     author = {Jiang, Z. and Carlson, B. and Zuo, Bowen and others},
     journal = {Journal of Instrumentation (JINST)},
     volume = {19},
     pages = {P05031},
     year = {2024},
   }