Naicheng Yu
I am a master's student in Computer Science and Engineering at UC San Diego. Before UCSD, I received my B.E. in Software Engineering from Zhejiang University.
My research focuses on LLM agents and benchmarking and evaluation.
I am applying to PhD programs for Fall 2027 and am always happy to discuss research ideas or potential collaborations.
Research Interests
Benchmarking and Evaluation
Understanding what benchmarks actually measure, and designing evaluations whose results are reliable and interpretable.
Self-Evolving Agents
Building agents that accumulate knowledge from their past attempts and use it to keep improving, rather than starting from scratch on every task.
Chain-of-Thought Monitoring
Studying whether a model's reasoning faithfully reflects its behavior, and how reasoning traces can help oversee increasingly capable models.
Agent Memory Systems
Designing how agents store, organize, and share memory over long horizons and across multiple agents.
Publications
Meta-Delta: Evolving Knowledge through Global Review for LLM-Driven Optimization
NeurIPS 2026 Workshop on Meta-Agents
Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead
Architecture 2.0 Workshop, ASPLOS 2026 · [Paper]
* Equal contribution.
Education
University of California, San Diego
Master of Science in Computer Science and Engineering
University of California, Berkeley
Visiting Student
Zhejiang University
Bachelor of Engineering in Software Engineering
Teaching Experience
Principles in Computer Architecture (CSE 240A)
Teaching Assistant · UC San Diego
Fundamentals of Software Engineering
Teaching Assistant · Zhejiang University
Academic Service
- Reviewer · ICLR 2027
- Reviewer · NeurIPS 2026 TAE Workshop