Yuchen Zhu 朱雨宸

Machine Learning PhD @ Georgia Tech 🍀

prof_pic.jpg

Photo credit to Sichen. Grand Canyon.

Hi, I am Yuchen Zhu, a final-year Machine Learning PhD at Georgia Tech, advised by Molei Tao and Yongxin Chen.

I work on generative AI. My research centers on advancing agentic LLMs from data-centric and system perspectives, with current focus on coding agents. In parallel, I push the speed–intelligence frontier of LLM through diffusion language models (dLLMs), from both the algorithmic and system perspectives. I am also broadly interested in diffusion models for probabilistic inference and applications across images, video, and the sciences.

During Summer 2026, I am a Research Scientist Intern at NVIDIA, exploring frontier agentic LLMs.

During Spring 2026, I was fortunate to work with Jiuxiang Gu and Jing Shi as a Research Scientist Intern at Adobe Research, building efficient & capable dLLMs at scale.

I graduated with BS in Mathematics (Honors) from NYU Shanghai and MA in Statistics from Yale University. My research started in applied mathematics, optimal control and RL theory, and has since evolved toward generative AI — a path that still informs how I think about modeling and inference today.

You can find more details in my CV here.

📧 Feel free to reach out: yzhu738@gatech.edu / yuchenzhu0226@gmail.com

Updates

  • 06/2026 I start as a Research Scientist Intern at NVIDIA, working on LLM coding agents.
  • 06/2026 A2D2 is online! The follow-up to TR2-D2 — a unified framework for reward-guided fine-tuning of any-length masked discrete diffusion via joint insertion-and-unmasking policies with adaptive decoding!
  • 06/2026 Check out ALE, by far the most comprehensive benchmark of frontier models' capability in real professional work. Happy to be part of the effort and contributing my expertise.
  • 06/2026 FLARE is online! Check out our new study on how to build SOTA dLLMs from LLMs with hybrid-attention!
  • 05/2026 DMPO got Spotlight (< 2.2%) at ICML 2026! Code is available here. See you in Seoul!
  • 05/2026 5 papers @ ICML 2026: LaViDa-R1, Discrete ASBS, DMPO, PRISM, and Rethinking Diffusion RL.
  • 02/2026 Rethinking Diffusion RL is online! Importance of likelihood estimation in diffusion model RL is more than you think.
  • 01/2026 I start as a Research Scientist Intern at Adobe Research, working on post-training of diffusion LLMs.
  • 10/2025 DMPO is online! Check out a completely new RL paradigm of diffusion LLMs for effective post-training! Code is available here.
  • 10/2025 TR2-D2 is online! Discover the SOTA method for finetuning MDM for bio seqs design with tree search + off-policy RL!
  • 10/2025 PDNS is online! See our new upgrade of MDNS with additional proximal gradient steps!
  • 09/2025 MDNS and Discrete Fast Solvers got accepted to NeurIPS 2025, see you in San Diego!
  • 08/2025 MDNS is online! Come find out our new work on ways to doing RL with masked discrete diffusion!
  • 06/2025 Mimicking or Reasoning is public! Check out our new work on evaluating MM-ICL for VLM reasoners!
  • 06/2025 Diffuse Everything is online! Check out our new work on multimodal diffusion models on native state spaces!
  • 05/2025 Learning to Stop and Diffuse Everything got accepted to ICML 2025, see you in Vancouver!
  • 04/2025 I wrote a new blog on how group structures aid generative modeling of manifold data.
  • 02/2025 Check out my new work on fast high-order samplers for discrete diffusion models!
  • 01/2025 TDM and STEM got accepted to ICLR 2025, see you in Singapore!
  • 10/2024 New work Plug-and-Play Controllable Generation for Discrete Masked Models is online!
  • 05/2024 New work Trivialized Momentum Facilitates Diffusion Generative Modeling on Lie Groups is online!
  • 04/2024 New work Quantum state generation with Structure Preserving Diffusion Model is online!

Selected Publications

  1. agents-last-exam.png
    Agents' Last Exam
    Agents' Last Exam Team
    Preprint, 2026
    llm · agent
  2. flare.png
    FLARE: Diffusion for Hybrid Language Model
    Yuchen Zhu, Jing Shi, Chongjian Ge, Hao Tan, Yiran Xu, Wanrong Zhu, Jason Kuen, Koustava Goswami, Rajiv Jain, Yongxin Chen, Molei Tao, and Jiuxiang Gu
    Preprint, 2026
    dllm · llm
  3. rethinking-rl-diffusion.png
    Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design
    Jaemoo Choi*, Yuchen Zhu*, Wei Guo, Petr Molodyk, Bo Yuan, Jinbin Bai, Yi Xin, Molei Tao, and Yongxin Chen
    ICML 2026
    rl · diffusion
  4. dmpo.png
    Enhancing Reasoning for Diffusion LLMs via Distribution Matching Policy Optimization
    Yuchen Zhu*, Wei Guo*, Jaemoo Choi, Petr Molodyk, Bo Yuan, Molei Tao, and Yongxin Chen
    ICML 2026 ⭐ Spotlight (< 2.2%)
    rl · dllm
  5. tr2d2.png
    TR2-D2: Tree Search Guided Trajectory-Aware Fine-Tuning for Discrete Diffusion
    Sophia Tang*, Yuchen Zhu*, Molei Tao, and Pranam Chatterjee
    Preprint, 2025
    rl · dllm · bio
  6. fast-solver.png
    Fast Solvers for Discrete Diffusion Models: Theory and Applications of High-Order Algorithms
    Yinuo Ren*, Haoxuan Chen*, Yuchen Zhu*, Wei Guo*, Yongxin Chen, Grant M Rotskoff, Molei Tao, and Lexing Ying
    NeurIPS 2025
    dllm · inference
  7. diffusion-gene-expression.png
    Diffusion Generative Modeling for Spatially Resolved Gene Expression Inference from Histology Images
    Sichen Zhu*, Yuchen Zhu*, Molei Tao, and Peng Qiu
    ICLR 2025
    diffusion · bio

Talks

  • 03/2026 INFORMS Optimization Society Conference 2026
  • 09/2025 GT ML Student Conference
  • 08/2025 MolSS Reading Group
  • 11/2024 GT ML Student Seminar
  • 10/2024 SIAM MDS 2024
  • 04/2024 Southeast ACM Student Workshop 2024