Haoyu Dong

소개

Haoyu Dong is a Senior Researcher at Microsoft Research. His work focuses on LLM post-training and reinforcement learning environments for agents that perform real-world knowledge work. He develops reinforcement learning, on-policy distillation, and supervised fine-tuning methods, alongside training environments that integrate Office applications, Bash tools, and search over web and enterprise data. His work connects model training, agent evaluation, and production deployment to improve long-horizon reasoning, tool use, and reliable execution across spreadsheets, documents, slides, and email.

His research has contributed to features deployed in Microsoft Copilot and Excel Agent. His paper Neural Formatting for Spreadsheet Tables was nominated for Best Full Research Paper at CIKM 2020, and his more recent project, SpreadsheetLLM, was featured as Hugging Face Paper of the Day (ranked (1st for the day) (opens in new tab).

Haoyu co-founded and co-organized the NeurIPS Table Representation Learning Workshop (opens in new tab) from 2022 to 2024 and served on the steering committee for the VLDB 2023 Tabular Data Analysis Workshop. He co-presented the “Large Language Models for Tabular Data” tutorial (opens in new tab) at SIGIR 2024 and delivered the “Spreadsheet Intelligence and Tabular Data Analytics” keynote (opens in new tab) at the VLDB 2024 TaDA Workshop.