I am an Assistant Professor in the Computer Science Department at UC San Diego. My research interests broadly span natural language processing, machine learning, and artificial intelligence. My group’s research is generously supported by grants from Google, Amazon, Apple, the AI for Science program, and other sponsors. I received my PhD from University of Washington (UW, NLP) working with Yejin Choi.
Research
Building AI agents that can interact, reason, and generalize in complex environments.
- Controllable, constrained, and creative reasoning (e.g., Uni-LaDiR; Diffusion-RL; Latent Reasoning; COLD-Decoding)
- Self-evolving, any-time learning — agents that continue to learn during deployment, not only during training (e.g., SimWorld Studio, DeliveryGym, Self-Evolving Agent for ARC)
- Open-ended world and agent modeling (e.g., Code4Scene, RT-SAFE, SimWorld)
- AI reasoning for science and real-world systems (e.g., Coding-Agent for Scientific Simulators, Chemistry Reasoning)
News
- [2026-10] Our team is part of MAESTRO, one of 12 new Phase II projects nationwide selected under the DOE Genesis Mission ($159M in total awards). I lead the project’s agentic AI framework, developing physics-informed coding agents that orchestrate simulations, autonomously conduct scientific experiments, and continuously improve through feedback. (DOE announcement | UCSD feature)
- [Upcoming] More talks at NeurIPS 2026 workshops. Stay tuned!
- [2026-09] Four new releases: Code4Scene, DeliveryGym, and RT-SAFE from SimWorld, and Uni-LaDiR for unified multimodal latent reasoning (blog). See the demos.
- [2026-09] Talk at the Agent in World workshop at ECCV 2026 on “Scaling Embodied AI through Automatic Environment Generation with Evolving Coding Agents.”
- [2026-08] Talk at the Apple Workshop on Language Interaction, hosted by Apple, on “Evolving Coding Agents for Automatic Environment Generation and Scalable Embodied AI.”
- [2026-07] Two talks at ACL workshops: ALVR: Automatic Environment Generation for Scaling Embodied Learning; and KnowFM : Agents Reason, Learn, and Act in Complex Embodied and Scientific Environments .
- [2026-05] We’re thrilled to release SimWorld Studio, an open-source platform for automatic 3D environment generation with coding agents..
- [2026-01] We're honored to receive the ARC Prize 2025 Paper Track Runner-Up Award for ArcMemo.
- [2025-11] Our lab is grateful to receive research support from Apple, Amazon, and Google.
- [2025-10] We’re partnering with Microsoft Research on Multi-agent collaberation and LLM memory.
- [2025-09] We’re excited to have 5 papers accepted to the NeurIPS 2025 including 1 spotlight.
- [2024-09] My CSE291: Large Model Reasoning lectures start on September 27th!
- [2024-07] I'm super excited to join UCSD CSE!
- [2024-06] I finished visiting AI2 mosaic as an Young Investigator.
Q-Lab
Ph.D. Students
- Haoqiang (Murray) Kang
- Matthew Ho
- Jixuan Chen
- Yixing Chen
- Diptanshu Sikdar (NSF GR Fellow)
- Mehmet Hamza Erol (co-advised with Yuxiang Wang)
- Xiaokang (Koe) Ye (co-advised with Zhiting Hu)