Bowen Fang

Ph.D. Student at Columbia University.

bwf.JPG

I am a Ph.D. student at Columbia University , where I work with the Data Science Institute (Smart Cities Center). My research interests are:

  1. How learning systems discover useful structure, search meaningfully, and build competence in huge spaces with sparse signal and minimal useful prior structure.
  2. Understanding the behaviors behind post-training algorithms for large models.
  3. Evaluating and improving system resilience under uncertainty, stress, and distribution shift.

This summer, I am returning as an Applied Scientist Intern at Amazon AGI Foundations, working on AWS Bedrock Model Customization , where my work centers on post-training for large models.

I hold an M.S. in Operations Research from Columbia University and a Bachelor in Big Data Management and Applications, minor in Economics from Peking University .

This website serves as a central hub for my publications, projects, and professional activities.

news

Nov 05, 2026 Our paper D²-OPD: Separating Direction and Distance in Extrapolative On-Policy Distillation is now on arXiv! This work from my Amazon AGI internship improves language-model distillation by separating where to extrapolate from how far to move.
Oct 13, 2026 I am honored to receive a 2026 Google TPU Builders Research award to support my research.
Apr 26, 2026 I will return to Amazon this summer as an Applied Scientist Intern with Amazon AGI Foundations, working on AWS Bedrock Model Customization.
May 27, 2025 I will be starting my Applied Scientist Internship at AWS AI Labs this summer!
May 20, 2025 I am honored to be a winner of the CS3 VALIDATE Accelerator program, which will provide continued funding for our work SINA.

latest posts

Oct 29, 2023 Some Recent Advancement Around MuZero
Jun 20, 2023 MPC with a Differentiable Forward Model: An Implementation with Jax
May 12, 2023 Adding MuZero into RL Toolkits at Ease

selected publications

  1. D²-OPD: Separating Direction and Distance in Extrapolative On-Policy Distillation
    Bowen Fang, Dingmin Wang, and Hengzhi Pei
    arXiv preprint arXiv:2611.07241, Nov 2026
    arXiv Bib PDF
  2. mr_overview_v2.png
    Do Math Reasoning LLMs Help Predict the Impact of Public Transit Events?
    Bowen Fang, Ruijian Zha, and Xuan Di
    Transportation Research Part C: Emerging Technologies (Special Issue: Foundation Models and Large Language Models in Urban Mobility), Nov 2025
    PDF
  3. algo_efficient.png
    Efficient Consistency Model Training for Policy Distillation in Reinforcement Learning
    Bowen Fang and Xuan Di
    In ICLR 2025 Workshop on Deep Generative Model in Machine Learning: Theory, Principle and Efficacy, Nov 2025
    PDF
  4. nts_b.png
    SLAMuZero: Plan and learn to Map for Joint SLAM and Navigation
    Bowen Fang, Xu Chen, Zhengkun Pan, and 1 more author
    In Proceedings of the International Conference on Automated Planning and Scheduling, Nov 2024
    PDF Code