I joined Computer Science & Engineering (CSE) Department at Washington University in St. Louis (WashU) in 2024 as an Assistant Professor. I received my Ph.D. degree in Computer Science Department, UIUC, advised by Prof. Jiawei Han. After that, I visited UW as a researcher and worked with Prof. Hanna Hajishirzi. Prior to UIUC, I received my Bachelor Degree in Electronic Engineering in Tsinghua University in 2018. My research interest broadly lies in the intersection of natural language processing and machine learning, and I am especially interested in understanding the properties of language models as well as improving their trustworthiness and efficiency.

News

Recent Research Interests

My group studies how to make language models self-improving, efficient, and trustworthy. Selected recent work:

Self-Evolving Models: Learning without Human-Curated Data

Training models that generate their own tasks, data, and environments.

• ICLR 2026 Chengsong Huang et al. R-Zero: Self-Evolving Reasoning LLM from Zero Data
• CVPR 2026 Yicheng He et al. VisPlay: Self-Evolving Vision-Language Models from Images
• Preprint 2026 Chengsong Huang et al. G-Zero: Self-Play for Open-Ended Generation from Zero Data
• Preprint 2026 Chengsong Huang et al. EnvHarness: Awakening Static Worlds for Agent Learning
• EMNLP 2023 Jiaxin Huang et al. Large Language Models Can Self-Improve

Reinforcement Learning and Exploration for Reasoning

How reasoning improves under RL, and how to keep exploration broad.

• Preprint 2026 Zhepei Wei et al. You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories
• EMNLP 2026 Langlin Huang et al. Nonsense Helps: Prompt Space Perturbation Broadens Reasoning Exploration
• COLM 2025 Jixuan Leng et al. CrossWordBench: Evaluating the Reasoning Capabilities of LLMs with Controllable Puzzle Generation

Efficient Reasoning and Inference

Reducing the cost of long reasoning chains at training and inference time.

• EMNLP 2026 Chengsong Huang et al. RelayLLM: Efficient Reasoning via Collaborative Decoding
• COLM 2026 Langlin Huang et al. SPEED: Position Specialist Generates Better Draft for Speculative Decoding
• ICML 2026 Tong Zheng et al. Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing
• EACL 2026 Chengsong Huang et al. Divide, Reweight, and Conquer: A Logit Arithmetic Approach for In-Context Learning

Reward Modeling and Calibration

Making reward signals and model confidence reliable enough to train on.

• ICLR 2026 Chengsong Huang et al. Efficient Test-Time Scaling via Self-Calibration
• ICML 2026 Jinyuan Li et al. Training Data Efficiency in Multimodal Process Reward Models
• ICLR 2025 Jixuan Leng et al. Taming Overconfidence in LLMs: Reward Calibration in RLHF
• Preprint 2026 Jinyuan Li et al. Process Rewards with Learned Reliability

Honors and Awards

NSF CAREER Award 2026
AAAI New Faculty Highlight Program 2026
Microsoft Research PhD Fellowship 2021-2023
C.W. Gear Outstanding Graduate Award
Chirag Foundation Graduate Fellowship
Outstanding Graduates, Tsinghua University 2018
Academic Excellence Scholarship, Tsinghua University 2015-2017
China National Scholarship (Top 1%) 2016
Samsung Scholarship 2015