Yite Wang

Bellevue, WA

I’m a Research Scientist at Snowflake AI Research, where I train reasoning models and LLM agents, including data science agents and SWE agents. Before that, I was a Research Scientist on ByteDance’s Seed-Foundation-Code team.

I earned my Ph.D. from the University of Illinois at Urbana-Champaign under Prof. Ruoyu Sun, with co-advising from Prof. Naira Hovakimyan. In my first year I collaborated with Prof. Justin Sirignano on deep-learning for computational fluid dynamics (CFD), and during my master’s studies I worked with Prof. Kyle Smith on numerical simulation for energy storage systems.

Research interests: Reasoning Models; SWE Agents; Data Science Agents; Efficient Deep Learning; Natural Language Processing; Computer Vision; Numerical Methods.

news

Jul 09, 2026	We have three papers accepted to COLM 2026. Papers coming soon!
Apr 30, 2026	Our paper Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning is accepted to ICML 2026. [arXiv] [code]
Feb 27, 2026	Our paper DARE-bench: Evaluating Modeling and Instruction Fidelity of LLMs in Data Science is accepted to ICLR 2026. [arXiv] [code]
Apr 01, 2025	I joined Snowflake AI Research, working on reasoning models and LLM agents.
May 01, 2024	I joined ByteDance Seed-Foundation-Code as a Research Scientist working on LLM for code.

blogs

Arctic RL: Open-Source Backend Open-source infrastructure for training and evaluating reinforcement-learning agents.
Hybrid Deep Research Benchmark A benchmark for evaluating long-horizon research agents across web and enterprise data.
ArcticSwarm: Hybrid Deep Research A hybrid agent system for deep research over private and public information sources.
ArcticSwarm: Multi-Agent System Architecture Architecture notes for coordinating specialized agents in enterprise research workflows.
Enterprise Text-to-SQL with Arctic R2 Reasoning models for production-grade text-to-SQL over enterprise data.
DARE-bench: LLM Data Science Workflows A benchmark for measuring modeling and instruction fidelity in data science workflows.
Agent World Model for Agentic Reinforcement Learning Synthetic environments for training agentic reinforcement-learning systems.

selected publications

* denotes equal contribution. ^† denotes project lead.

Bridging Databases and Documents: Data-Algorithm Co-Design for Hybrid Question Answering

Ruofan Wu, Boyi Liu, Fan Shu, Yite Wang, Zhewei Yao, Yuxiong He, and Feng Yan

In COLM, 2026

RL DataAgent RL TrainingBenchmark
MidTool: Mid-training Data Synthesis for Agentic Tool Use

Fengqing Jiang, Yite Wang^†, Boyi Liu, Zhaoyang Wang, Canwen Xu, Zhewei Yao, Radha Poovendran, and Yuxiong He

In COLM, 2026

Midtrain DataTool Use AgentAgent Training
The Optimizer Is the Agent: Reasoning-Driven Search across Prompts, Programs, and ML Workflows

Junbo Li, Boyi Liu, Canwen Xu, Yite Wang, Zhewei Yao, Zhangyang Wang, Qiang Liu, and Yuxiong He

In COLM, 2026

Auto Research
Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning

Zhaoyang Wang, Canwen Xu, Boyi Liu, Yite Wang, Siwei Han, Zhewei Yao, Huaxiu Yao, and Yuxiong He

In ICML, 2026

RL DataAgent RL Training

HTML Code Blog
DARE-bench: Evaluating Modeling and Instruction Fidelity of LLMs in Data Science

Fan Shu^*, Yite Wang^*^†, Ruofan Wu, Boyi Liu, Zhewei Yao, Yuxiong He, and Feng Yan

In ICLR, 2026

RL DataAgent RL TrainingBenchmark

HTML Code Blog
FullStack Bench: Evaluating LLMs as Full-Stack Coders

ByteDance Seed-Foundation-Code Team

Preprint, 2025

Evaluation

HTML
LEMON: Lossless Model Expansion

Yite Wang, Jiahao Su, Hanlin Lu, Cong Xie, Tianyi Liu, Jianbo Yuan, Haibin Lin, Ruoyu Sun, and Hongxia Yang

In ICLR, 2024

Model Expansion

HTML Code
Balanced Training for Sparse GANs

Yite Wang^*, Jing Wu^*, Naira Hovakimyan, and Ruoyu Sun

In NeurIPS, 2023

Pruning

HTML Code
NTK-SAP: Improving Neural Network Pruning by Aligning Training Dynamics

Yite Wang, Dawei Li, and Ruoyu Sun

In ICLR, 2023

Pruning

HTML Code
Global Convergence of MAML and Theory-Inspired Neural Architecture Search for Few-Shot Learning

Haoxiang Wang^*, Yite Wang^*, Ruoyu Sun, and Bo Li

In CVPR, 2022

Neural Architecture SearchNTK

HTML Code