Elamur Research⌕
☰
home / authors / Zheng Yuan

Zheng Yuan

Zheng Yuan works in natural language processing. Their papers here centre on how large language models are trained, aligned with human preferences and taught to solve problems, alongside an earlier strand on detecting errors in learner writing. A recurring question is how training data shapes what a model can do: how much of it there is, where it comes from, and how different kinds are combined. The usual method is to fine-tune open LLaMA-family models and test them on public benchmarks. GSM8K, a benchmark for mathematical reasoning, appears often.

The most cited of these papers is Scaling Relationship on Learning Mathematical Reasoning with Large Language Models. It studies how improvements from fine-tuning on math word problems vary with the underlying model and the size of the training set. It also introduces rejection sampling fine-tuning, where a model generates many candidate solutions per question and keeps the distinct correct ones as additional examples. RRHF offers a lighter route to preference alignment than the PPO stage of the standard RLHF pipeline. It gathers candidate answers produced by the model, by other systems or by human writers, scores them, and trains with a ranking loss. Earlier, Artificial Error Generation with Machine Translation and Syntactic Patterns, written with Marek Rei and Mariano Felice, created synthetic faulty sentences from correct text to compensate for the shortage of annotated learner writing.

The newest work moves from single skills to combinations of them. How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition (2024) blends fine-tuning data for mathematics, code generation and general alignment. It then looks at how the volume of each kind of data, the proportions in which they are blended, the size of the model and the order of training influence each skill.

Yuan is currently at the University of Sheffield, after earlier periods at King's College London and the University of Cambridge, where the work on learner errors was done. The language-model papers grew out of collaborations with Alibaba Group and Tsinghua University. Frequent co-authors include Hongyi Yuan, Guanting Dong, Keming Lu and Chengpeng Li.

Zheng Yuan has 6 papers in our library, in 2017–2025, most of them on Language models & pretraining, Machine translation, Summarization; often with Hongyi Yuan, Chang Zhou, Chengpeng Li.

6 papers in our library · 2017–2025

OpenAlex · ORCID · h-index 19 · 88 works · 1,356 citations OpenAlex, 2026-10-14

📄 6 works in our library · 🔗 Cited 983 times in our library · 7 more in our library without a page yet · ⭐ Paper Stars ranking: #2495

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

 Affiliations

🎓University of SheffieldSheffield · United Kingdom2025–2026 · 9 papersEmployment since 2025 · Senior Lecturer, School of Computer Sciencecurrent🎓University of CambridgeCambridge · United Kingdom2013–2024 · 21 papersEmployment 2016–2022 · Research Associate, Department of Computer Science and Technology🎓King's College LondonLondon · United Kingdom2022–2025 · 15 papersEmployment until 2025Employment 2022–2025 · Lecturer, Department of Informatics🎓Tsinghua UniversityBeijing · China2021–2023 · 5 papers🎓Beijing Jiaotong UniversityBeijing · China2021–2024 · 4 papers🎓Southwest Jiaotong UniversityChengdu · China2007–2020 · 3 papers🎓Shanghai Jiao Tong UniversityShanghai · China2024–2025 · 2 papers🏥Shanghai Ninth People's HospitalShanghai · China2024–2025 · 2 papers🏢Alibaba GroupHangzhou · China2023 · 2 papers🎓Northwestern UniversityEvanston · United States2018–2020 · 2 papers

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 14 more institutions on single bylines are not listed.

 Fields, topics and concepts

Of its works in our library, with the number of works on each.

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 6 worksComputer Science · 6 worksArtificial Intelligence · 6 worksLanguage and Linguistics · 6 worksComputational Theory and Mathematics · 1 workComputer Vision and Pattern Recognition · 1 workInformation Systems · 1 work

🧭 Topics 6Research subjects and languages the work is about.Language models & pretraining · 4 worksReasoning · 1 workMachine translation · 1 workInformation retrieval & search · 1 workEvaluation & benchmarks · 1 workSummarization · 1 work

 Works by year

2020s

20251 paperZheng Yuan's papers of 202520241 paperZheng Yuan's papers of 202420233 papersZheng Yuan's papers of 2023

2010s

20171 paperZheng Yuan's papers of 2017

 Papers

The most cited first (OpenAlex and our library).

  1. Scaling Relationship on Learning Mathematical Reasoning with Large Language Models (2023 · 252 citations)
    Solving math word problems is hard for LLMs, and it was unclear how fine-tuning gains depend on the base model and on how much data is available.
  2. RRHF: Rank Responses to Align Language Models with Human Feedback without tears (2023 · 194 citations)
    Large language models are usually aligned with human preferences using the RLHF pipeline from InstructGPT.
  3. How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition (2024 · 145 citations)
    Open-source models are usually fine-tuned for one skill at a time, while proprietary LLMs handle many skills at once.
  4. Evaluation Metrics in the Era of GPT-4: Reliably Evaluating Large Language Models on Sequence to Sequence Tasks (2023 · 89 citations)
    The authors examine whether standard automatic metrics still judge modern generative LLMs fairly on sequence-to-sequence tasks.
  5. Artificial Error Generation with Machine Translation and Syntactic Patterns (2017 · 82 citations)
    Automated detection of writing errors is limited by the small amount of manually annotated learner data. Rei, Felice, Yuan and Briscoe (Cambridge, ALTA Institute) address this by synthesising erroneous text from correct sentences. They compare two supervised approaches, both trained on the public FCE training set…
  6. A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models (2025 · 64 citations)
    General-purpose LLMs often fall short in specialized fields. Their domain knowledge is shallow, they struggle with long chains of rule-bound reasoning, and they misread terms whose meaning depends on context. Fine-tuning on domain data can introduce hallucinations and catastrophic forgetting. Classic RAG avoids…

All works

13 works of theirs in our library, the most cited first; 6 with a page here, the others link their original until we summarise them.

  1. Grammatical error correction using neural machine translation ↗Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · no summary yetZheng Yuan, Ted Briscoe2016 · 339 citations
  2. Scaling Relationship on Learning Mathematical Reasoning with Large Language ModelsarXiv:2308.01825Zheng Yuan, Hongyi Yuan, Chengpeng Li, et al.2023 · 252 citations
  3. RRHF: Rank Responses to Align Language Models with Human Feedback without tearsarXiv:2304.05302Zheng Yuan, Hongyi Yuan, Chuanqi Tan, et al.2023 · 194 citations
  4. How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data CompositionAnnual Meeting of the Association for Computational Linguistics (ACL)Guanting Dong, Hongyi Yuan, Keming Lu, et al.2024 · 145 citations
  5. Constrained Grammatical Error Correction using Statistical Machine Translation ↗Proceedings of the Seventeenth Conference on Computational Natural Language Learning: Shared Task · no summary yetZheng Yuan, Mariano Felice2013 · 98 citations
  6. How well do Large Language Models perform in Arithmetic tasks? ↗arXiv:2304.02015 · no summary yetZheng Yuan et al.2023 · 91 citations
  7. Evaluation Metrics in the Era of GPT-4: Reliably Evaluating Large Language Models on Sequence to Sequence TasksConference on Empirical Methods in Natural Language Processing (EMNLP)Andrea Sottana, Bin Liang, Kai Zou, et al.2023 · 89 citations
  8. Artificial Error Generation with Machine Translation and Syntactic PatternsWorkshop on Innovative Use of NLP for Building Educational Applications (BEA)Marek Rei, Mariano Felice, Zheng Yuan, et al.2017 · 82 citations
  9. A Survey of Graph Retrieval-Augmented Generation for Customized Large Language ModelsarXiv:2501.13958Qinggang Zhang, Shengyuan Chen, Yuanchen Bei, et al.2025 · 64 citations
  10. On the application of Large Language Models for language teaching and assessment technology ↗arXiv:2307.08393 · no summary yetAndrew Caines et al.2023 · 51 citations
  11. The BEA 2023 Shared Task on Generating AI Teacher Responses in Educational Dialogues ↗Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023) · no summary yetAnaïs Tack et al.2023 · 51 citations
  12. Prompting open-source and commercial language models for grammatical error correction of English learner text ↗Findings of the Association for Computational Linguistics: ACL 2024 · no summary yetChristopher Davis et al.2024 · 42 citations
  13. How effective is machine translation on low-resource code-switching? A case study comparing human and automatic metrics ↗Findings of the Association for Computational Linguistics: ACL 2023 · no summary yetLi Nguyen et al.2023 · 5 citations

 Co-authors here

12 of 19 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

 Cited by

Papers of our library cite their works 983 times (of the 141,840 papers whose bibliographies we read); 74 citing papers with a page here →.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/authors/zheng-yuan/