PeopleofSuperintelligence

Researcher

Jonathan Crabbé

Jonathan Crabbé is a machine learning researcher and member of technical staff at Latent Labs in London, and he coauthored the Latent-X protein design model released in July 2025. As a PhD student in Mihaela van der Schaar's group at Cambridge's Department of Applied Mathematics and Theoretical Physics, he developed interpretability methods such as Dynamask, presented at ICML in 2021, and SimplEx, which explains a model's latent representations through a chosen corpus of examples. During an internship at Microsoft Research AI4Science, he contributed to MatterGen, a diffusion model for inorganic materials that Nature published in January 2025. In 2024, he presented a paper on time series diffusion in the frequency domain at ICML, and his study of robust CLIP vision encoders appeared as an Apple Machine Learning Research publication in January 2025.

Badges

Timeline

  1. 2020

    PhD research at Cambridge's DAMTP

    He pursues a PhD in machine learning at Cambridge in the Department of Applied Mathematics and Theoretical Physics (DAMTP), in Mihaela van der Schaar's machine learning and AI for medicine group, with a thesis on new interpretability methods. In 2020, he coauthors "Learning outside the Black-Box", presented at NeurIPS 2020, an algorithm that builds global interpretations of black-box functions from Meijer G-functions.

  2. 2021

    Dynamask and SimplEx

    At ICML in 2021, he and van der Schaar present Dynamask, which scores the importance of each feature at each time step of a time series by fitting a perturbation mask to the input sequence. SimplEx, presented at NeurIPS 2021 with Zhaozhi Qian, Fergus Imrie and van der Schaar, explains a model's latent representations through a freely chosen corpus of examples.

  3. 2022

    Label-free explainability at ICML

    In February 2022, he presents SimplEx at the van der Schaar Lab's Inspiration Exchange session on interpretable machine learning. With van der Schaar, he presents "Label-Free Explainability for Unsupervised Models" at ICML 2022, which extends post-hoc feature-importance and example-importance methods to unsupervised models trained without labels.

  4. October 2023–January 2025

    Interpreting CLIP's robustness

    In October 2023, he and Pau Rodríguez, Vaishaal Shankar, Luca Zappella and Arno Blaas release "Interpreting CLIP", a study of 16 robust zero-shot CLIP vision encoders that finds outlier features only in the robust models. The paper, an Apple Machine Learning Research publication, appears in Transactions on Machine Learning Research in January 2025.

  5. December 2023–January 2025

    MatterGen at Microsoft Research

    As an intern at Microsoft Research AI4Science, he contributes to MatterGen, a diffusion model for generating novel, stable materials with desired properties, which Microsoft announces in December 2023. Nature publishes the MatterGen paper, "A generative model for inorganic materials design", with Crabbé among its coauthors, in January 2025.

  6. 2024

    Diffusion in the frequency domain

    In 2024, he presents "Time Series Diffusion in the Frequency Domain" at ICML with Nicolas Huynh, Jan Pawel Stanczuk and van der Schaar, showing that diffusion in the time domain has a dual process in the frequency domain. On real-world healthcare and finance datasets, its frequency diffusion models capture the training distribution better than time diffusion models.

  7. July 2025

    Latent-X for protein binders

    In July 2025, he is one of the Latent Labs team authors of Latent-X, an all-atom protein design model that jointly generates the structure and sequence of a protein binder and its target. For macrocyclic peptides, its designs reach experimental hit rates above 90% on all evaluated benchmark targets.

  8. As of October 2026

    Technical staff at Latent Labs

    Following the 2025 release of Latent-X, he works as a member of technical staff at Latent Labs in London. He holds a PhD in machine learning from Cambridge and is a former research scientist intern at Apple and Microsoft.

Career

  1. NowLatent LabsMember of technical staff
  2. MicrosoftIntern
  3. AppleResearch scientist intern

Education

  1. From 2020University of CambridgePhD

From the biography.

  1. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Latent Labs. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  2. 5 of the 25 people with the most-cited recent papers also lead on career citationsIncludes Latent Labs. We compare citations to 2023–2025 papers with career citations and h-index, after correcting reviewed publication-date errors: who leads, when they started and where they work.
  3. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  4. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. 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.

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 Jonathan Crabbé yet. Send us a free-licensed photo.

  1. Jonathan Crabbé | DeepAI deepai.org. Checked 20 October 2026.
  2. [2011.08596] Learning outside the Black-Box: The pursuit of interpretable models arxiv.org. Checked 20 October 2026.
  3. JonathanCrabbe (Jonathan Crabbé) · GitHub github.com. Checked 20 October 2026.
  4. Explaining Time Series Predictions with Dynamic Masks proceedings.mlr.press. Checked 20 October 2026.
  5. [2110.15355] Explaining Latent Representations with a Corpus of Examples arxiv.org. Checked 20 October 2026.
  6. Inspiration Exchange: next frontiers in interpretable ML // van der Schaar Lab vanderschaar-lab.com. Checked 20 October 2026.
  7. [2203.01928] Label-Free Explainability for Unsupervised Models arxiv.org. Checked 20 October 2026.
  8. [2310.13040] Interpreting CLIP: Insights on the Robustness to ImageNet Distribution Shifts arxiv.org. Checked 20 October 2026.
  9. Interpreting CLIP: Insights on the Robustness to ImageNet Distribution Shifts - Apple Machine Learning Research machinelearning.apple.com. Checked 20 October 2026.
  10. MatterGen: Property-guided materials design - Microsoft Research microsoft.com. Checked 20 October 2026.
  11. A generative model for inorganic materials design - Microsoft Research microsoft.com. Checked 20 October 2026.
  12. Time Series Diffusion in the Frequency Domain proceedings.mlr.press. Checked 20 October 2026.
  13. [2507.19375] Latent-X: An Atom-level Frontier Model for De Novo Protein Binder Design arxiv.org. Checked 20 October 2026.