About Me

I am a Research Scientist at BandAI, ByteDance, working on cutting-edge artificial intelligence technologies. I focus on advancing the capabilities of AI systems through innovative research in deep learning and autonomous agent development. I have been recognized through prestigious talent programs including Alibaba Star and Huawei Genius Youth Program.

My research interests primarily focus on Deep Research, Agentic Reinforcement Learning, Tool Use, and Self Evolution. I am particularly interested in developing intelligent systems that can autonomously learn, adapt, and evolve their capabilities to solve complex real-world problems.

I have published several papers in top-tier conferences, including AAAI, EMNLP, KDD, and contributed to open-source projects like TableBank, DocBank, LayoutLM, and TrOCR that have gained significant impact in the community. Notably, LayoutLM won the 2024 International Congress of Basic Science Frontier Science Award, and my research has received 2,288 citations on Google Scholar. Currently, I am working on next-generation AI systems that can perform autonomous research and continuously improve themselves.

🔬 My Research Interests

• 🤖 Deep Research
• 🎯 Agentic Reinforcement Learning
• 🛠️ Tool Use
• 🌱 Self Evolution

🔥 News

• 2025.09: 🌱 Released SamplingEvolve: A test-time scaling framework that transforms model sampling from independent trajectories to experience-guided evolution, achieving 91.36% accuracy on GAIA dataset through continuous trajectory optimization.
• 2025.08: 📊 Published ReportBench: The first systematic benchmark for evaluating Deep Research agents on academic survey tasks, featuring automated construction from arXiv papers and comprehensive fact-checking mechanisms.

📖 Education

• 2019.09 - 2024.06, Ph.D. in Computer Science and Technology, Beihang University, Joint Ph.D. program with Microsoft Research Asia (MSRA)
• 2015.09 - 2019.06, Bachelor in Computer Science and Technology, Beihang University

💼 Experience

• 2025.04 - Present, Research Scientist, BandAI, ByteDance, Beijing
  • Research Focus: Deep Research, Agentic Reinforcement Learning, Tool Use, Self Evolution
  • Working on next-generation AI systems capable of autonomous research and self-improvement
  • Developing advanced agentic frameworks for complex problem solving
• 2024.03 - 2025.04, Research Scientist, Seed, ByteDance, Beijing
  • Research Focus: User Preference Optimization, Instruction Following, Model Alignment
  • User Flywheel - Semantic Signal Optimization: Developed methods to extract user preference signals from conversation patterns where users modify prompts or add critiques when unsatisfied with initial responses, constructing RM training data to improve model alignment with user preferences
  • User Flywheel - Behavioral Signal Optimization: Built PointRM, a binary classification model to identify user copy behavior, using it to create preference pairs from equal-rated responses, achieving 3.5% improvement in human evaluation
  • Instruction Following Optimization: Designed verifiable atomic instructions across multiple languages and user needs, combined with nested logical relationships (AND, OR, NOT, IF) for instruction composition, achieving ~10 point improvement on public benchmarks and 10% advantage in human evaluation GSB
• 2023.02 - 2024.02, Research Intern, Conversational AI, DAMO Academy, Alibaba, Beijing
  • Research Focus: Tool-Augmented LLMs, Human Alignment
  • Developed API-Bank benchmark and PRO (Preference Ranking Optimization) method
  • Contributed to evaluation standards for tool-calling capabilities in LLMs
• 2018.07 - 2023.02, Research Intern, Natural Language Computing Group, Microsoft Research Asia (MSRA), Beijing
  • Mentor: Dr. Lei Cui
  • Research Focus: Document Intelligence, Multi-modal Pre-training Models
  • Long-term internship as part of joint Ph.D. program

🎖 Honors and Awards

• 2024 International Congress of Basic Science Frontier Science Award (for LayoutLM) - Awarded at the International Congress of Basic Science founded by S.T. Yau, hosted by Beijing Municipal Government, Ministry of Science and Technology, China Association for Science and Technology, and International Congress of Chinese Mathematicians
• 2024 Alibaba Star & Huawei Genius Youth Program

📝 Publications

LLMs Agents

• arXiv ReportBench: Evaluating Deep Research Agents via Academic Survey Tasks, Minghao Li, Ying Zeng, Zhihao Cheng, Cong Ma, Kai Jia [Paper] [Code]
• Technical Report SamplingEvolve: Enhancing LLM Performance through Strategic Sampling and Self-Evolution, Minghao Li, Ying Zeng, Cong Ma, Siyao Song, Kai Jia [Project Page] [Code]
• EMNLP 2024 API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs, Minghao Li, Yingxiu Zhao, Bowen Yu, Feifan Song, Hangyu Li, Haiyang Yu, Zhoujun Li, Fei Huang, Yongbin Li [Paper] [Code]

Document AI

• AAAI 2024 TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models, Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei [Project Page] [Paper]
• KDD 2020 LayoutLM: Pre-training of Text and Layout for Document Image Understanding, Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou [Project Page] [Paper]
• COLING 2020 DocBank: A Benchmark Dataset for Document Layout Analysis, Minghao Li, Yiheng Xu, Lei Cui, Shaohan Huang, Furu Wei, Zhoujun Li, Ming Zhou [Project Page] [Paper]
• LREC 2020 TableBank: Table Benchmark for Image-based Table Detection and Recognition, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou, Zhoujun Li [Project Page] [Paper]