Lang Cao

Hi! đź‘‹ Here is Lang Cao, a PhD student in Computer Science at the University of Illinois Urbana-Champaign (UIUC), happily advised by Prof. Guo Yue. Before starting my PhD, I interned at Microsoft Research for about one year, where I had a wonderful experience and was fortunate to collaborate with Haoyu Dong and Mengyu Zhou. Previously, I received my Master's degree in Computer Science from UIUC and earned my Bachelor's degree from Wuhan University of Technology.

I am an open researcher. My research primarily focuses on artificial intelligence (AI), particularly large language models (LLMs) and their applications. Beyond AI research, I'm also an active practitioner in Web3 and quantitative trading.

Email  /  CV  /  Bio  /  Scholar  /  Github  /  Linkedin

profile photo

Research

My academic research focuses on machine learning, machine reasoning, and their applications on health. Recently, I have been interested in improving the reasoning capabilities of LLMs and exploring their practical applications. My previous research experience spans various areas, including LLM Reasoning, LLM Generation, Table LLMs, AI for Health, LLM Agents, NLP Applications, as well as other topics related to LLMs and machine learning (ML).

Selected research projects are listed below.

Fortune Fortune: Formula-Driven Reinforcement Learning for Symbolic Table Reasoning in Language Models
Lang Cao, Jingxian Xu, Hanbing Liu, Jinyu Wang, Mengyu Zhou, Haoyu Dong, Shi Han, Dongmei Zhang
Under Review, 2025
bibtex / paper
[Table LLMs] Formula Tuning (Fortune) is a reinforcement learning approach that enables language models to perform symbolic table reasoning by deriving executable spreadsheet formulas.
Bingo Bingo: Boosting Efficient Reasoning of LLMs via Dynamic and Significance-based Reinforcement Learning
Hanbing Liu, Lang Cao, Yuanyi Ren, Mengyu Zhou, Haoyu Dong, Xiaojun Ma, Shi Han, Dongmei Zhang
Under Review, 2025
bibtex / paper
[LLM Reasoning] Bingo is a reinforcement learning framework that trains LLMs for efficient reasoning by combining significance-aware and dynamic length rewards, improving both accuracy and efficiency across reasoning benchmarks.
SuperRL SuperRL: Reinforcement Learning with Supervision to Boost Language Model Reasoning
Yihao Liu, Shuocheng Li, Lang Cao, Yuhang Xie, Mengyu Zhou, Haoyu Dong, Xiaojun Ma, Shi Han, Dongmei Zhang
Under Review, 2025
bibtex / paper
[LLM Reasoning] SuperRL is a unified training framework that adaptively combines supervised and reinforcement learning to boost language model reasoning, achieving greater stability and generalization, especially under sparse rewards.
DeepRetrieval DeepRetrieval: Hacking Real Search Engines and Retrievers with Large Language Models via Reinforcement Learning
Pengcheng Jiang, Jiacheng Lin, Lang Cao, Runchu Tian, SeongKu Kang, Zifeng Wang, Jimeng Sun, Jiawei Han
Under Review, 2025
bibtex / page / paper / code
[LLM Others] DeepRetrieval is a framework that leverages reinforcement learning to hack real-world search engines and retrievers by training LLMs to perform effective retrieval tasks.
tablepilot TablePilot: Recommending Human-Preferred Tabular Data Analysis with Large Language Models
Deyin Yi, Yihao Liu, Lang Cao, Mengyu Zhou, Haoyu Dong, Shi Han, Dongmei Zhang
ACL 2025 Industry Track (Oral), 2025
bibtex / paper / poster
[Table LLMs] TablePilot is a framework based on LLMs; when given an input table, it outputs recommended data analysis queries along with their corresponding code and results.
tablemaster TableMaster: A Recipe to Advance Table Understanding with Language Models
Lang Cao, Hanbing Liu
Under Review, 2025
bibtex / paper / code
[Table LLMs] We analyze the challenges of table understanding with language models and propose a recipe and comprehensive framework to address them.
leads A foundation model for human-AI collaboration in medical literature mining
Zifeng Wang, Lang Cao, Qiao Jin, et al. (23 authors total), Jimeng Sun
Nature Communications, 2025
bibtex / paper
[AI for Health] LEADS is a specialized foundation model for medical literature mining that outperforms generic LLMs across multiple tasks and significantly improves accuracy and efficiency in expert workflows for evidence-based medicine.
leads Accelerating clinical evidence synthesis with large language models
Zifeng Wang, Lang Cao, Benjamin Danek, Qiao Jin, Zhiyong Lu, Jimeng Sun
npj Digital Medicine, 2025
bibtex / paper
[AI for Health] TrialMind is a specialized generative AI system for clinical evidence synthesis that surpasses general LLMs in search, screening, and data extraction, significantly enhancing accuracy and efficiency in expert-driven systematic reviews.
KG-FIT KG-FIT: Knowledge Graph Fine-Tuning Upon Open-World Knowledge
Pengcheng Jiang, Lang Cao, Cao Xiao, Parminder Bhatia, Jimeng Sun, Jiawei Han
NeurIPS 2024, 2024
bibtex / paper / code
[LLM Others] KG-FIT enhances knowledge graph embeddings by integrating LLM-guided hierarchical semantics with KG structure, achieving significant performance gains in link prediction across multiple benchmarks.
Learn to Refuse Learn to Refuse: Making Large Language Models More Controllable and Reliable through Knowledge Scope Limitation and Refusal Mechanism
Lang Cao
EMNLP 2024 Main Conference, 2024
bibtex / paper / code
[LLM Generation] Learn to Refuse (L2R) is a method empowering large language models to refuse answering difficult questions, thereby improving accuracy and reliability by utilizing a separate, expandable knowledge base.
GraphReason GraphReason: Enhancing Reasoning Capabilities of Large Language Models through A Graph-Based Verification Approach
Lang Cao
ACL 2024 Natural Language Reasoning and Structured Explanations Workshop, 2024
bibtex / paper / code
[LLM Reasoning] GraphReason is a graph-based verification approach that enhances the reasoning capabilities of large language models by verifying the reasoning process through a graph-based framework.
diaggpt DiagGPT: An LLM-based and Multi-agent Dialogue System with Automatic Topic Management for Flexible Task-Oriented Dialogue
Lang Cao
Technical Report, 2023
bibtex / paper / code
[LLM Agents] DiagGPT extends large language models for task-oriented dialogue by enabling proactive question-asking and topic management, achieving strong performance in complex diagnostic interactions across domains like medicine and law.
PILOT PILOT: Legal Case Outcome Prediction with Case Law
Lang Cao, Zifeng Wang, Cao Xiao, Jimeng Sun
NAACL 2024 Main Conference, 2024
bibtex / paper / code
[NLP Applications] PILOT is a legal case outcome prediction model for case law systems that retrieves relevant precedents and accounts for temporal legal shifts, significantly outperforming prior civil-law-focused approaches.

For more research, please visit my full research page or check out my Google Scholar .


Education and Experiences

blind-date Ubiquant
May 2025 - Sept. 2025 | Shanghai, China
AI and Quant Intern
Research Focus: AI for Quantitative Trading.
Mentor: Ziwei Yang
blind-date Microsoft Research
Aug. 2024 - June 2025 | Beijing, China
Research Intern at Data Knowledge and Intelligence Group
Research Focus: Spreadsheet Intelligence, Table LLMs, General Machine Reasoning.
Mentor: Haoyu Dong, Mengyu Zhou
blind-date Tsinghua University
Nov. 2024 - Feb. 2025 | Beijing, China
Research Assistant at THU-NLP Lab
Research Focus: Multi-modal Learning and Reasoning.
blind-date iFLYTEK
June 2021 - Aug. 2021 | Hefei, China
AI Algorithm Intern at Smart Car Technology R&D Division
Focus: AI Applications on Smart Cars.
Mentor: Shenan Li
blind-date University of Illinois Urbana-Champaign (UIUC)
Doctor of Philosophy in Computer Science
Aug. 2025 - Present | Urbana, US
Research Area: Machine Learning; Machine Reasoning; AI for Health
Advisor: Yue Guo

Master of Science in Computer Science
Aug. 2023 - May 2024 | Urbana, US
Research Area: AI for Healthcare; Natural Language Processing

Resaerch Assistant at Sunlab
Jan. 2023 - May 2024 | Urbana, US
Research Focus: NLP / LLMs Applications for Healthcare and Legal.
blind-date Wuhan University of Technology (WUT)
Bachelor of Engineering in Software Engineering
Sept. 2018 - June 2022 | Wuhan, China
Rank 1st/79, GPA 93.51/100 (3.94/4.0), National Scholarship


Miscellanea

Selected Rewards
  • Silver Medal, top 5% in Kaggle Common Lit Readability Prize (2021.8)
  • Top 2% in Alibaba Tianchi NLP Chinese Pre-training Model Generalization Ability Challenge (2021.1)
  • National Scholarship (1%), WUT (2020); Merit Student Model Honor (5‰), WUT (2020)
  • Outstanding Student Leader of the School (2%), WUT (2019)
  • Outstanding Graduate (2%), WUT (2022.6); Outstanding Thesis (1%), WUT (2022.6)
  • The National Champion of the 2014 FIRST LEGO League Challenge in China (core) (2014.6)
  • Gold Award at the 2016 Asia-Pacific Championship of the FIRST LEGO League Challenge (leader) (2%) (2016.7)

Invited Talks
  • “Thoughts of DeepSeek R1 & Reasoning in Language Models”, invited by Haoyu Dong at Microsoft Research, (2025.02).
  • “Language Models' Table Understanding with Formula”, invited by Mengyu Zhou at Microsoft Research, (2025.04).

Academic Service
  • Paper Review: ACL Rolling Review (ACL, EMNLP, NAACL), ICLR, ICML, NeurIPS, KDD


© Copyright Lang Cao (last updated May 26, 2025).