Hi, I am Sitao Cheng, a Ph.D. student at the University of Waterloo, fortunate to be advised by Prof. Victor Zhong. I am currently a Student Researcher at Samaya AI, working with Yuhao Zhang. Before that, I was a research scholar in the UCSB NLP Group with Prof. William Wang, and a research intern at Microsoft Research Asia. I received my Master’s degree from Nanjing University.

My research asks how reasoning in large language models can generalize efficiently: how a model can autonomously compose the skills it already has (parametric) with newly acquired ones (contextual), and recursively improve itself. I work across language agents, reinforcement learning, retrieval-augmented generation (RAG), and neural-symbolic reasoning. Three questions drive my current work:

1. Language agents. What makes reasoning transfer to open, real-world environments rather than to curated benchmarks?
2. Automatic reward modeling. Can reward functions be discovered rather than hand-designed, through differentiable evolutionary meta-rewards?
3. RL and compositional generalization. Which training strategies produce models that compose skills they were never explicitly taught, and what does that reveal about how RL works? I study this concretely by building robust GUI agents.

I am always glad to talk about research, so please feel free to reach out. You can also read my CV.

Recent News

Preprints

• From Atomic to Composite: Reinforcement Learning Enables Generalization in Complementary Reasoning
  Sitao Cheng, Xunjian Yin, Ruiwen Zhou, Yuxuan Li, Xinyi Wang, Liangming Pan, William Yang Wang, Victor Zhong
  paper code data

• Differentiable Evolutionary Reinforcement Learning
  Sitao Cheng*, Tianle Li*, Xuhan Huang*, Xunjian Yin, Difan Zou
  paper code model

• Epistemic Context Learning: Building Trust the Right Way in LLM-Based Multi-Agent Systems
  Ruiwen Zhou*, Maojia Song*, Xiaobao Wu, Sitao Cheng, Xunjian Yin, Yuxi Xie, Zoey Hao, Wenyue Hua, Liangming Pan, Soujanya Poria, Min-Yen Kan
  paper code

Selected Publications

• [KnowFM @ ACL’25 Oral] Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models
  Sitao Cheng, Liangming Pan, Xunjian Yin, Xinyi Wang, William Yang Wang
  paper code

• [ACL’24 Findings] Call Me When Necessary: LLMs can Efficiently and Faithfully Reason over Structured Environments
  Sitao Cheng, Ziyuan Zhuang, Yong Xu, Fangkai Yang, Chaoyun Zhang, Xiaoting Qin, Xiang Huang, Ling Chen, Qingwei Lin, Dongmei Zhang, Saravan Rajmohan, Qi Zhang
  paper code

• [ACL’26 Oral] LEDOM: Reverse Language Model
  Xunjian Yin, Sitao Cheng, Yuxi Xie, Xinyu Hu, Li Lin, Xinyi Wang, Liangming Pan, William Yang Wang, Xiaojun Wan
  paper model

• [ACL’24 Oral] QueryAgent: A Reliable and Efficient Reasoning Framework with Environmental Feedback based Self-Correction
  Xiang Huang*, Sitao Cheng*, Shanshan Huang, Jiayu Shen, Yong Xu, Chaoyun Zhang, Yuzhong Qu
  paper code

• [ACL’25] Disentangling Memory and Reasoning Ability in Large Language Models
  Mingyu Jin, Weidi Luo, Sitao Cheng, Xinyi Wang, Wenyue Hua, Ruixiang Tang, William Yang Wang, Yongfeng Zhang
  paper code

• [EMNLP’24] EfficientRAG: Efficient Retriever for Multi-Hop Question Answering
  Ziyuan Zhuang, Zhiyang Zhang, Sitao Cheng, Fangkai Yang, Jia Liu, Shujian Huang, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Qi Zhang
  paper code

• [ACL’25] RuleArena: A Benchmark for Rule-Guided Reasoning with LLMs in Real-World Scenarios
  Ruiwen Zhou, Wenyue Hua, Liangming Pan, Sitao Cheng, Xiaobao Wu, En Yu, William Yang Wang
  paper code

Teaching

• CS240 Data Structures and Data Management, Teaching Assistant, Winter & Spring 2026, UWaterloo
• CS115 Introduction to Computer Science 1, Teaching Assistant, Spring 2025, UWaterloo

Services

• Reviewer: ARR, ICLR, ICML, NeurIPS
• Volunteer: ACL 2024