Jonathan Crabbé

Jonathan Crabbé

PhD Researcher

University of Cambridge

Biography

I am currently working towards my PhD thesis in the Applied Mathematics Department at the University of Cambridge. In this stimulating environment, I am learning to become a well-rounded Machine Learning researcher.

We are gradually entering in a phase where humans will increasingly interact with generative and predictive AIs, hence forming human-AI teams. I see an immense potential in these teams to approach cutting-edge scientific and medical problems. My research focuses on making these teams more efficient by improving the information flow between complex ML models and human users. This touches upon various subjects of the AI literature, including ML Interpretability, Robust ML and Data-Centric AI. In some sense, my goal is to build this microscope that would allow human beings to look inside a machine learning model. Through the interface of this microscope, human beings can rigorously validate ML models, extract knowledge from them and learn to use these models more efficiently.

Download my resumé.

Interests
  • Interpretability
  • Robust ML
  • Generative AI
  • ML for Science and Healthcare
  • Representation Learning
  • Data-Centric AI
Education

  • PhD in Applied Mathematics, 2020-2024

    University of Cambridge

  • MASt in Applied Mathematics, 2018-2019

    University of Cambridge

  • M1 in Physics, 2017-2018

    Ecole Normale Supérieure Paris

  • Bachelor in Engineering, 2014-2017

    Université Libre de Bruxelles

Skills

Coding

Experience in big-tech companies, implementation of SOTA ML

Mathematical Modelling

Strong mathematical background, my publications have strong theoretical components

Presenting

Presentation of my research at many prestigious venues (NeurIPS, ICML, ICLR)

Team Working

~50% of my publications are the result of collaborative work

Autonomy

~50% of my publications are the result of autonomous work

Supervision

Supervision of several MPhil and PhD students, creation of pedagogical YouTube videos

Experience

 
 
 
 
 
Apple Research
Research Intern
Apple Research
May 2023 – Oct 2023 Cambridge, UK
 
 
 
 
 
Microsoft Research
Research Intern
Microsoft Research
Feb 2023 – May 2023 Cambridge, UK
  • Conducted research in the AI4Science team.
  • Contributed to the development of generative ML techniques for material discovery.
  • Presented findings in front of AI4Science team, results will be integrated in a big paper.
 
 
 
 
 
University of Cambridge
PhD Researcher
University of Cambridge
Oct 2020 – Present Cambridge, UK
  • Conducted research in various sub-fields of machine learning.
  • Published several papers in top-tier conferences (NeurIPS, ICML).
  • Supervised the research of several MPhil/PhD students.
 
 
 
 
 
G-Research
Quantitative Research Intern
G-Research
Jun 2022 – Sep 2022 London, UK
  • Conducted research in machine learning applied to quantitative finance.
  • Learned to turn raw financial data into predictive features.
  • Presented findings in front of quant managers.
 
 
 
 
 
Université Libre de Bruxelles
Research Intern
Université Libre de Bruxelles
Oct 2019 – Oct 2020 Bruxelles, BE
  • Conducted research in black holes physics.
  • Created several pedagogical videos to help young students with maths and physics.
  • Responsible of physics example classes for first year pharma students.
 
 
 
 
 
Imperial College London
Research Intern
Imperial College London
Feb 2018 – Jul 2018 London, UK
  • Conducted research in quantum field theory and cosmology.
  • Implemented numerical solver for simulating the evolution of a toy cosmological model.
  • Demonstrated the emergence of a new type of singularity called caustics.

Accomplish­ments

PhD Fellowship
Aviva Sep 2020 – Sep 2024
Full PhD funding (tuition and maintenance).
Research Assistant Fellowship
Université Libre de Bruxelles Sep 2019 – Sep 2020
Funding for a year of research.
Jennings Price
Wolfson College, University of Cambridge Sep 2019
Awarded based on outstanding results for my MASt.
Scholarship
Labex-ICFP Sep 2017 – Sep 2018
Awarded based on academic excellence.

Recent Posts

2 Papers Accepted at ICML 2022
Label-Free Explainability and Data-SUITE have been accepted at ICML 2022. These are two new methods that machine learning interpretability and robustness. I provide a quick summary of the methods along with the relevant links.
Jonathan Crabbé
Jun 8, 2022 1 min read Conferences
Project Project Project
2 Papers Accepted at ICML 2022

Research Projects

Representation Learning
Learning realistic representations of the world.
Representation Learning
Robust Machine Learning
Making sure that machines are doing what they are supposed to do.
Robust Machine Learning
Interpretable Machine Learning
Creating an interface between machine learning models and human beings.
Interpretable Machine Learning

Recent & Upcoming Talks

2 Spotlight Presentations at ICML 2022
Presentation of two accepted papers at the main conference.
Jul 17, 2022 — Jul 23, 2022 Baltimore Convention Center
Project Project LFXAI Data-SUITE
2 Spotlight Presentations at ICML 2022

Featured Publications

Jonathan Crabbé, Pau Rodríguez, Vaishaal Shankar, Luca Zappella, Arno Blaas
October, 2023 In arXiv
Robust multimodal models have outlier features and encode more concepts
Robust multimodal models have outlier features and encode more concepts

We investigate what makes multimodal models that show good robustness with respect to natural distribution shifts (e.g., zero-shot CLIP) different from models with lower robustness using interpretability.
PDF Cite Project Project Project
Jonathan Crabbé, Mihaela van der Schaar
April, 2023 In NeurIPS 2023
Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance
Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance

We assess the robustness of various interpretability methods by measuring how their explanations change when applying symmetries of the model to the input features.
PDF Cite Code Project Project Poster
Jonathan Crabbé, Mihaela van der Schaar
September, 2022 In NeurIPS 2022
Concept Activation Regions: A Generalized Framework For Concept-Based Explanations
Concept Activation Regions: A Generalized Framework For Concept-Based Explanations

We extend existing feature and example importance methods to unsupervised learning.
PDF Cite Code Project Project Video
See all publications

Recent Publications

Quickly discover relevant content by filtering publications.
Jonathan Crabbé, Pau Rodríguez, Vaishaal Shankar, Luca Zappella, Arno Blaas (2023). Robust multimodal models have outlier features and encode more concepts. In arXiv.

PDF Cite Project Project Project

Arthur Vandenhoeke, Lennert Antson, Guillem Ballesteros, Jonathan Crabbé, Michal Shimoni (2023). Explaining the Absorption Features of Deep Learning Hyperspectral Classification Models. In IGARSS 2023.

PDF Cite Project

Nabeel Seedat, Jonathan Crabbé, Zhaozhi Qian, Mihaela van der Schaar (2023). TRIAGE: Characterizing and auditing training data for improved regression. In NeurIPS 2023.

PDF Cite Project Poster

Jonathan Crabbé, Mihaela van der Schaar (2023). Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance. In NeurIPS 2023.

PDF Cite Code Project Project Poster

Alan Jeffares, Tennison Liu, Jonathan Crabbé, Fergus Imrie, Mihaela van der Schaar (2023). TANGOS: Regularizing Tabular Neural Networks through Gradient Orthogonalization and Specialization. In ICLR 2023.

PDF Cite Code Project

Alan Jeffares, Tennison Liu, Jonathan Crabbé, Mihaela van der Schaar (2023). Joint Training of Deep Ensembles Fails Due to Learner Collusion. In NeurIPS 2023.

PDF Cite Project Project Poster

Jonathan Crabbé, Mihaela van der Schaar (2022). Concept Activation Regions: A Generalized Framework For Concept-Based Explanations. In NeurIPS 2022.

PDF Cite Code Project Project Video

Nabeel Seedat, Jonathan Crabbé, Ioana Bica, Mihaela van der Schaar (2022). Data-IQ: Characterizing subgroups with heterogeneous outcomes in tabular data. In NeurIPS 2022.

PDF Cite Code Project

Jonathan Crabbé, Alicia Curth, Ioana Bica, Mihaela van der Schaar (2022). Benchmarking Heterogeneous Treatment Effect Models through the Lens of Interpretability. NeurIPS 2022 Datasets and Benchmarks Track.

PDF Cite Code Project Project

Jonathan Crabbé, Mihaela van der Schaar (2022). Label-Free Explainability for Unsupervised Models. In ICML 2022.

PDF Cite Code Project Project Video

Nabeel Seedat, Jonathan Crabbé, Mihaela van der Schaar (2022). Data-SUITE: Data-centric identification of in-distribution incongruous examples. In ICML 2022.

PDF Cite Project

Hao Sun, Boris van Breugel, Jonathan Crabbé, Nabeel Seedat, Mihaela van der Schaar (2022). Latent Density Models for Uncertainty Categorization. In NeurIPS 2023.

PDF Cite Project Project Poster

Jonathan Crabbé, Zhaozhi Qian, Fergus Imrie, Mihaela van der Schaar (2021). Explaining Latent Representations with a Corpus of Examples. In NeurIPS 2021.

PDF Cite Code Project Video

Jonathan Crabbé, Mihaela van der Schaar (2021). Explaining Time Series Predictions with Dynamic Masks. In ICML 2021.

PDF Cite Code Project Video

Jonathan Crabbé, Yao Zhang, William R. Zame, Mihaela van der Schaar (2020). Learning outside the black-box: the pursuit of interpretable models. In NeurIPS 2020.

PDF Cite Code Project Video

Popular Topics

Academic Aleatoric Uncertainty Autoencoder Case-Based Reasoning Conformal Uncertainty Data-Centric AI Deep learning Example Importance Explainability Feature Importance Generative AI Influential Examples Interpretability Latent Representation Machine learning Robust Machine Learning Robustness Self-Supervised Learning Uncertainty Unsupervised Learning

Contact