Fangru Lin
Researcher
Now Research Scientist, Gemini pretraining · Google DeepMind
@FangruLin99fangru-lin.github.io ↗GitHub ↗LinkedIn ↗Latest move
University of Oxford → Google DeepMind · Research Scientist, Gemini pretraining
Computational linguist. DPhil student at Oxford supervised by Janet Pierrehumbert and Daniel Altshuler, working on how language models reason and plan. Interned at Microsoft Research and on the Gemini Deep Research team before joining Gemini pretraining in London.
Path
Google DeepMindResearch Scientist, Gemini pretrainingNEW- ATAlan Turing InstituteEnrichment Student2024
Microsoft ResearchResearch Intern2024–2024- UOUniversity of OxfordDPhil, Linguistics, Philology and Phonetics, advised by Janet Pierrehumbert and Daniel Altshuler2023
- UOUniversity of OxfordMPhil, Linguistics, Philology and Phonetics2021–2023
- MAMicrosoft (Azure Cloud+AI)Software Engineer Intern2022–2022
- SIShanghai International Studies UniversityBA, Korean2017–2021
Papers
- Can Large Language Models Generalize Procedures Across Representations?ICML 2026 2026Tests whether language models trained on procedures written as code or graphs can carry them over to the same tasks written in natural language, and proposes a training order that helps.
- Assessing Dialect Fairness and Robustness of Large Language Models in Reasoning TasksACL 2025 2024Builds ReDial, reasoning benchmarks rewritten in African American Vernacular English, and measures how language models answer them compared with Standard English.
- Graph-enhanced Large Language Models in Asynchronous Plan ReasoningICML 2024 2024Studies how language models plan tasks with steps that can run in parallel, and adds a graph representation of the plan to the prompt.
SourcesLatest move: Xfangru-lin.github.ioEducation and jobs are read from their own homepage or employer's page; papers checked on arXiv. From LinkedIn only the person's own public posts are read; GitHub is a link only.