About
Hey there!
I'm Yujie Zhao, a third-year Ph.D. student advised by Prof. Jishen Zhao in Computer Science and Engineering Department of UC San Diego, starting in 2024. Before coming to UCSD, I received my B.S. degree in Applied Physics at University of Science and Technology of China.
I used to be an undergraduate research intern in EIC Lab at GaTech under the supervision of Professor Yingyan (Celine) Lin and Yang (Katie) Zhao. Previously I have also done research about RL theory under the supervision of Professor Huazheng Wang.
My research focuses on building agents from language to physics: self-evolving and efficient agent systems that reason, learn, and adapt across both digital and physical environments. Feel free to contact me if you share these interests!
Recent Research Interests
- Self-Evolving Agent Systems
- Physical AI and Robotic Agents
News
- Our paper HiddenRAG was accepted to ICCAD 2026.
- Our collaborative paper VideoScience-Bench was accepted to ECCV 2026.
- Released our paper MetaAgent-X on end-to-end reinforcement learning for automatic multi-agent systems.
- One first-author paper AMA-Bench accepted to ICML 2026.
- Selected as a 63rd DAC Young Fellow.
- Joined Google as a Student Research Intern.
- One first-author paper Pro-V-R1 accepted to DAC 2026.
- One first-author paper Stronger-MAS accepted to ICLR 2026.
- Excited to release our PettingLLMs for multi-agent reinforcement learning training framework — check it out and give it a try!
- Completed three-month research internship at Intel AI Team.
- One first-author paper accepted to DAC 2025.
- One first-author paper accepted to NeurIPS 2024.
- One first-author paper accepted to DAC 2024.
First Author Publications
Full list available on Google Scholar
-
AMA-Bench: Evaluating Long-Horizon Memory for Agentic ApplicationsICML 2026
-
MetaAgent-X: Breaking the Ceiling of Automatic Multi-Agent Systems via End-to-End Reinforcement LearningUnder Review, 2026
-
PRO-V-R1: Reasoning Enhanced Programming Agent for RTL VerificationDAC 2026
-
Stronger-MAS: Multi-Agent Reinforcement Learning for Collaborative LLMsICLR 2026
-
MAGE: A Multi-Agent Engine for Automated RTL Code GenerationDAC 2025
-
RA-PbRL: Provably Efficient Risk-Aware Preference-Based Reinforcement LearningNeurIPS 2024
-
3D-Carbon: An Analytical Carbon Modeling Tool for 3D and 2.5D Integrated CircuitsDAC 2024
Selected Collaborated Publications and Manuscripts
-
HiddenRAG: Minimizing Retrieval-Augmented Generation Latency via In-Storage ProcessingIEEE/ACM ICCAD 2026
-
Benchmarking Scientific Understanding and Reasoning for Video Generation using VideoScience-BenchECCV 2026
-
OrcaLoca: An LLM Agent Framework for Software Issue LocalizationICML 2025