2026.04 – 2026.07
THUIR Lab, Tsinghua University
Research internship on memory and skill selection for long-horizon agents.
Haotian Ren 任浩天
Incoming Master’s Student in Computer Technology
Tsinghua University
Advisor: Prof. Jingtao Zhan · FAITH Lab
About me
I am an incoming master’s student in Computer Technology at Tsinghua University, where I will join FAITH Lab under the supervision of Prof. Jingtao Zhan. I am currently an undergraduate in Artificial Intelligence at Beijing University of Posts and Telecommunications (BUPT).
I study how LLM agents can learn from experience. My interests center on agent memory and self-evolving systems: how to turn past interactions into useful knowledge, when to retrieve it, and how to improve the system that manages it. I also enjoy building tools that connect research ideas with everyday use.
Selected research
First author
S2Forge: Co-Evolving Skill and Selection Spaces for Long-Horizon Agents
S²Forge studies the mismatch between what an agent remembers and how it chooses to use that knowledge. It jointly evolves a skill space of reusable experience and a selection space that determines when to activate, suppress or exclude a skill. State-transition evidence updates both spaces as the agent interacts with its environment. I led problem formulation, method design and evaluation during my research internship at THUIR.
ICML 2026 Co-first author
MemEvolve: Meta-Evolution of Agent Memory Systems
MemEvolve improves both an agent’s stored experience and its memory architecture. EvolveLab provides a shared implementation of twelve memory systems, organized around encoding, storage, retrieval and management. I worked on experimental design and implementation, and developed much of the open-source codebase.
Research experience
2026.02 – 2026.03
ValueByte Lab, Peking University
Extended Meta Context Engineering to long-horizon dialogue scenarios. Built an AppWorld-based evaluation pipeline and adapted the original setup to new tasks.
Related project2025.07 – 2025.12
Collaboration with NUS and OPPO AI Lab
Developed MemEvolve and EvolveLab, studying how agent memory architectures can improve through experience.
Education
Incoming
Tsinghua University
Master’s program in Computer Technology
Undergraduate
Beijing University of Posts and Telecommunications
Artificial Intelligence
Graduate recommendation ranking: 1/184
Selected honors
- National Scholarship, 2023–2024 and 2024–2025 (Top 1%)
- First Prize, China Undergraduate Mathematical Contest in Modeling (Top 0.5%)
- Gold Award, China International College Students’ Innovation Competition, Beijing Region (2025)
- Beijing Merit Student, 2024–2025 (1/463, 0.22%)
Beyond research
I have been building personal blogs since 2017. This one is where I keep course notes, work through ideas, and occasionally write about life. I enjoy programming, writing, and exploring new tools.
Outside research, I enjoy playing pool and table tennis. I also play video games, including Overwatch, PUBG, and Elden Ring.
Read my blog