Zihao Tang

AI4GC Lab

Zihao Tang

Now @ MicrosoftLLM AgentsAI MemoryAgentic RL
HomepageGoogle ScholarLinkedIn

About Me

Hi! I am Zihao Tang (唐子豪), a graduated Master student from AI4GC Lab at Zhejiang University.

My research focuses on LLM Agents, AI Memory, and Agentic RL. I am especially interested in agents that can accumulate experience, form reusable procedures, and improve through interaction rather than solving each task from scratch.

I am currently affiliated with Microsoft.

During AI4GC

During my time at AI4GC Lab, I worked on efficient and adaptable AI systems, moving from knowledge transfer under distribution shift to LLM-assisted model generation. In AuG-KD, I studied data-free knowledge distillation when teacher-domain knowledge cannot be directly transferred to real-world student domains, using uncertainty-guided anchors and mixup generation to selectively transfer useful knowledge. This line of work gave me a practical lens on adaptation: useful AI systems should not only perform well in controlled settings, but also adjust to new users, domains, and deployment constraints.

I then explored this direction through ModelGPT, which uses LLMs to generate tailored models from user data or task descriptions, making model construction faster and more accessible across NLP, CV, and tabular tasks. During my internship at MSRA, I also worked on Sigma, an efficient system-domain LLM built around DiffQKV attention. Together, these projects shaped my interest in systems that connect model capability with real deployment needs, from model generation to efficient long-context inference.

Selected Papers

ACL2026

Mnemis: Dual-Route Retrieval on Hierarchical Graphs for Long-Term LLM Memory

Zihao Tang, Xin Yu, Ziyu Xiao, Zengxuan Wen, Zelin Li, Jiaxi Zhou, Hualei Wang, Haohua Wang, Haizhen Huang, Weiwei Deng, Feng Sun, Qi Zhang

PaperProject96

ACL2026

TL; DR: Too Long, Do Re-weighting for Efficient LLM Reasoning Compression

Zhong-Zhi Li, Xiao Liang, Zihao Tang, Lei Ji, Peijie Wang, Haotian Xu, Xing W, Haizhen Huang, Weiwei Deng, Yeyun Gong, Zhijiang Guo, Xiao Liu, Fei Yin, Cheng-Lin Liu

PaperProject25

arXiv2025

Sigma: Differential Rescaling of Query, Key and Value for Efficient Language Models

Zhenghao Lin, Zihao Tang, Xiao Liu, Yeyun Gong, Yi Cheng, Qi Chen, Hang Li, Ying Xin, Ziyue Yang, Kailai Yang, Yu Yan, Xiao Liang, Shuai Lu, Yiming Huang, Zheheng Luo, Lei Qu, Xuan Feng, Yaoxiang Wang, Yuqing Xia, Feiyang Chen, Yuting Jiang, Yasen Hu, Hao Ni, Binyang Li, Guoshuai Zhao, Jui-Hao Chiang, Zhongxin Guo, Chen Lin, Kun Kuang, Wenjie Li, Yelong Shen, Jian Jiao, Peng Cheng, Mao Yang

Paper

Now

Now: Affiliated with Microsoft; continuing research on LLM Agents, AI Memory, and Agentic RL.

At Microsoft, our team focuses on LLM Agents, AI Memory, and Agentic RL. We are interested in how agents can remember, reason, search, code, and improve across long-horizon interactions, while keeping both memory retrieval and reasoning efficient. In Mnemis, I drive the technical direction for long-term memory retrieval beyond similarity-only RAG, combining System-1 similarity search with System-2 Global Selection over base and hierarchical memory graphs. With GPT-4.1-mini, Mnemis achieves state-of-the-art results: 93.9 on LoCoMo and 91.6 on LongMemEval-S.

Currently, we are exploring search agents, code agents, and procedural memory, especially how agents can turn experience into reusable action knowledge. The broader goal is to move from agents that complete isolated tasks toward agents that build up skills, habits, and memory over time.