PeopleofSuperintelligence

Researcher

Lei Yu

Lei Yu is a researcher at Google DeepMind whose work spans neural models of language and AI systems that prove olympiad-level mathematics. After undergraduate study at Nottingham and an MPhil at Cambridge, they did doctoral work at Oxford with Phil Blunsom and Stephen Pulman, writing papers on answer sentence selection, neural variational inference for text and the neural noisy channel. They joined DeepMind as a research scientist in January 2017 and in 2020 coauthored IsarStep, a benchmark for high-level mathematical reasoning. Yu was a key contributor to AlphaProof, which with AlphaGeometry 2 reached silver-medal standard at the 2024 International Mathematical Olympiad, and was among the key contributors of the IMO 2025 team whose Gemini Deep Think model earned a gold-medal score. In 2026, they coauthored the paper on AlphaProof Nexus, a Lean proof search framework whose most capable agent resolved 9 of 353 open Erdős problems.

Works
260
Badges

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. By 2014

    Oxford PhD and answer selection

    After undergraduate study at the University of Nottingham and an MPhil at the University of Cambridge, they become a doctoral student at Mansfield College in Oxford's Department of Computer Science, supervised by Phil Blunsom and Stephen Pulman. In December 2014, they post “Deep Learning for Answer Sentence Selection” with Karl Moritz Hermann, Blunsom and Pulman, a paper accepted at the NIPS deep learning workshop that matches state-of-the-art results on a TREC benchmark.

  2. 2015–2016

    Neural variational inference for text

    With Yishu Miao and Phil Blunsom, they coauthor “Neural Variational Inference for Text Processing” for ICML 2016, a variational inference framework for generative and conditional models of text. At EMNLP 2016, they present “Online Segment to Segment Neural Transduction” with Jan Buys and Blunsom.

  3. January 2017

    Research scientist at DeepMind

    They join DeepMind as a research scientist in January 2017. At ICLR 2017, they publish “The Neural Noisy Channel” with Blunsom, Chris Dyer, Edward Grefenstette and Tomas Kocisky, which casts sequence-to-sequence transduction as noisy channel decoding with recurrent neural networks.

  4. October 2018

    A thesis on sequence mapping

    Their Oxford PhD thesis, “Tackling Sequence to Sequence Mapping Problems with Neural Networks,” appears on arXiv in October 2018. It studies neural models for representing sequences, modelling interactions between sequences and using unpaired data to improve neural models.

  5. 2020–2021

    Document translation and IsarStep

    In 2020, they publish a TACL paper with Laurent Sartran, Wojciech Stokowiec, Wang Ling, Lingpeng Kong, Phil Blunsom and Chris Dyer on document-level machine translation with Bayes' rule, using models that can be learned from only parallel sentences and monolingual documents. With Wenda Li, Yuhuai Wu and Lawrence C. Paulson, they build IsarStep, a benchmark for high-level mathematical reasoning posted in June 2020 and published at ICLR 2021.

  6. July 2024

    AlphaProof at IMO 2024

    They are a key contributor to AlphaProof, a reinforcement-learning system for formal math reasoning. In July 2024, AlphaProof and AlphaGeometry 2 together solve four of the six problems at the 2024 International Mathematical Olympiad and score 28 of 42 points, the level of a silver medalist.

  7. 2025

    IMO gold and AlphaProof in Nature

    In July 2025, they are among the leads and key contributors of the IMO 2025 team behind an advanced Gemini Deep Think model, which solves five of the six problems perfectly and earns 35 of 42 points, a gold-medal score. In November 2025, they coauthor the Nature paper on AlphaProof, an AlphaZero-inspired agent that learns to find formal proofs through reinforcement learning, and make significant contributions to its auto-formalization research.

  8. As of October 2026

    AlphaProof Nexus at Google DeepMind

    In May 2026, they coauthor a Google DeepMind paper introducing AlphaProof Nexus, a framework for LLM-aided proof generation with feedback from the Lean compiler. Its most capable agent resolves 9 of 353 open Erdős problems and proves 44 of 492 OEIS conjectures. They are a researcher at Google DeepMind.

Career

  1. Since 2017Google DeepMindResearcher

Education

  1. University of Nottingham
  2. University of CambridgeMPhil
  3. University of OxfordPhD

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Lei Yu. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Lei Yu yet. Send us a free-licensed photo.

  1. Lei Yu | Department of Computer Science cs.ox.ac.uk. Checked 20 October 2026.
  2. [1412.1632] Deep Learning for Answer Sentence Selection arxiv.org. Checked 20 October 2026.
  3. [1511.06038] Neural Variational Inference for Text Processing arxiv.org. Checked 20 October 2026.
  4. Online Segment to Segment Neural Transduction - ACL Anthology aclanthology.org. Checked 20 October 2026.
  5. [1611.02554] The Neural Noisy Channel arxiv.org. Checked 20 October 2026.
  6. [1810.10802] Tackling Sequence to Sequence Mapping Problems with Neural Networks arxiv.org. Checked 20 October 2026.
  7. Better Document-Level Machine Translation with Bayes’ Rule - ACL Anthology aclanthology.org. Checked 20 October 2026.
  8. [2006.09265] IsarStep: a Benchmark for High-level Mathematical Reasoning arxiv.org. Checked 20 October 2026.
  9. AI achieves silver-medal standard solving International Mathematical Olympiad problems — Google DeepMind deepmind.google. Checked 20 October 2026.
  10. Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad — Google DeepMind deepmind.google. Checked 20 October 2026.
  11. Olympiad-level formal mathematical reasoning with reinforcement learning | Nature nature.com. Checked 20 October 2026.
  12. [2605.22763] Advancing Mathematics Research with AI-Driven Formal Proof Search arxiv.org. Checked 20 October 2026.
  13. Advancing Mathematics Research with AI-Driven Formal Proof Search arxiv.org. Checked 20 October 2026.