王曦立 (Xili Wang)

About me

I received my Ph.D. in Computational Mathematics from the School of Mathematical Sciences, Peking University. Before that, I received my bachelor degree from the School of Mathematical Sciences, Fudan University, majoring in Information and Computational Science and ranking 1/52.

Education

Collaborators

• Kejun Tang
• Xiaoliang Wan

Work Experience

Meituan LongCat Team, Mar 2025 - Sep 2025
Research intern and core member for agentic tool use, working on tool-use data synthesis, supervised fine-tuning, and reinforcement learning for multi-turn LLM agents.

• Built automated multi-turn tool-use data synthesis pipelines based on Multi-LLMs and MCP. The pipelines cover both simulated tool outputs with task-specific virtual databases and real MCP tool execution through user-agent-tool interactions.
• Used the synthesized multi-turn tool-use data for SFT of LongCat-Flash-Chat, improving its agentic tool-use capability. LongCat-Flash-Chat achieved TAU2 Retail 71, TAU2 Airline 58, and TAU2 Telecom 68 on TAU2-Bench.
• Worked on a multi-turn user-interacting reinforcement learning framework for agentic tool use. The work introduces LLM-simulated users into RL rollout so that agents learn to clarify dynamic user requirements, maintain task state, and coordinate tool calls over long interactions.

Research

My research interests include

• Computational statistics and adaptive sampling for scientific computing
• Density estimation and deep generative models
• Surrogate modeling and rare-event sampling
• Numerical methods for PDE-constrained optimization problems

Publications and preprints

• Zhitao Zhu, Xili Wang^†, Shizhe Wu, Jiawei Fu, and Xiaoqing Liu. How Should LLMs Consume High-Quality Data? Optimal Data Scheduling via Quality-Aware Functional Scaling Laws. arXiv:2605.25698, 2026.
• Yueyang Wang, Jiawei Fu, Baolong Bi, Xili Wang, and Xiaoqing Liu. HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench. International Conference on Machine Learning (ICML), 2026. arXiv:2601.20255.
• Yueyang Wang, Kejun Tang, Xili Wang, Xiaoliang Wan, Weiqing Ren, and Chao Yang. Estimating committor functions via deep adaptive sampling on rare transition paths. Journal of Computational Physics, 544:114439, 2026.
• Meituan LongCat Team, including Xili Wang. LongCat-Flash Technical Report. arXiv:2509.01322, 2025.
• Xili Wang, Kejun Tang, Jiayu Zhai, Xiaoliang Wan, and Chao Yang. Deep Adaptive Sampling for Surrogate Modeling Without Labeled Data. Journal of Scientific Computing, 101:77, 2024.
• Xili Wang, Pengfei Yin, Bo Zhang, and Chao Yang. AONN-2: An adjoint-oriented neural network method for PDE-constrained shape optimization. Journal of Computational Physics, 513:113160, 2024.

^† Corresponding author.

Patents

• 杨超, 王曦立, 尹鹏飞, 张博. 形状优化方法及计算机存储介质和终端设备, 中国专利, CN116680763B[P], 2024-05-17.

Invited talks

• "AONN-2: An adjoint-oriented neural network method for PDE-constrained shape optimization", SIAM Conference on Mathematics of Data Science (MDS24), Atlanta, Georgia, USA, Oct 2024.
• "DL for PDEs: towards parametric, high-dimensional and PDE-constrained optimization", Louisiana State University, Baton Rouge, Louisiana, USA, Oct 2024.

Honors and Awards

• 2023-2024 学年北京大学优秀科研奖
• 2023-2024 学年北京大学兴业银行奖学金
• 2022-2023 学年北京大学“硕博杯”羽毛球比赛第八名
• 复旦大学“拔尖计划”荣誉学生，2021 年
• 2020-2021 学年复旦大学本科优秀学生奖学金（毕业生奖学金）
• 2019-2020 学年复旦大学本科优秀学生奖学金一等奖（董氏奖学金）
• 2019-2020 学年复旦大学本科生专业奖学金
• 2018-2019 学年复旦大学本科优秀学生奖学金二等奖
• 2018-2019 学年复旦大学本科生专业奖学金