avatar

Antoine de Mathelin

Ph.D. Student
Centre Borelli, ENS Paris-Saclay
antoine.de_mathelin (at) ens-paris-saclay.fr


Short Bio

I am a last year Ph.D. student at the Centre Borelli of the ENS Paris-Saclay in France. My thesis is sponsored by the Michelin tire company and conducted under the supervison of Pr. Mathilde Mougeot, Pr. Nicolas Vayatis and François Deheeger.

My research focuses on developing reliable machine learning models under the intrinsic constraints of engineering design, such as domain shift and costly labeling. I am particularly interested in transfer learning, domain adaptation, active learning, uncertainty quantification, and out-of-distribution detection.

Publications [Google Scholar]

  1. ICTAI
    Personalized One-Shot Collaborative Learning
    Garin, M.*, de Mathelin, A.*, Mougeot, M. and Vayatis, N.
    (* Equal contribution)
    IEEE 35th International Conference on Tools with Artificial Intelligence (ICTAI), 2023.
    PDF Code BibTex

  2. FTL
    From theoretical to practical transfer learning, The ADAPT library
    de Mathelin, A., Deheeger, F., Mougeot, M. and Vayatis, N.
    Federated and Transfer Learning (FTL), 2022.
    PDF Code BibTex

  3. ECML-PKDD
    Fast and Accurate Importance Weighting for Correcting Sample Bias
    de Mathelin, A., Deheeger, F., Mougeot, M. and Vayatis, N.
    European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), 2022.
    PDF Code BibTex

  4. ICLR
    Discrepancy-Based Active Learning for Domain Adaptation
    de Mathelin, A., Deheeger, F., Mougeot, M. and Vayatis, N.
    International Conference on Learning Representations (ICLR), 2022.
    PDF Code BibTex

  5. AI2ASE
    Unsupervised domain adaptation for constraining star formation histories
    Gilda, S.*, de Mathelin, A.*, Bellstedt, S. and Richard, G.
    (* Equal contribution)
    3rd Annual AAAI Workshop on AI to Accelerate Science and Engineering (AAAI-AI2ASE), 2022.
    PDF Code BibTex

  6. DistShift
    Handling distribution shift in tire design
    de Mathelin, A., Deheeger, F., Mougeot, M. and Vayatis, N.
    NeurIPS Workshop on Distribution Shifts, Connecting Methods and Applications (NeurIPS-DistShift), 2021.
    PDF BibTex

  7. DGM
    A Binded VAE for Inorganic Material Generation
    Oubari, F., de Mathelin, A., Décatoire, R. and Mougeot, M.
    NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications (NeurIPS-DGM), 2021.
    PDF Code BibTex

  8. ICTAI
    Adversarial weighting for domain adaptation in regression
    de Mathelin, A., Richard, G., Deheeger, F., Mougeot, M. and Vayatis, N.
    IEEE 33rd International Conference on Tools with Artificial Intelligence (ICTAI), 2021.
    PDF Code BibTex

  9. ECML-PKDD
    Unsupervised multi-source domain adaptation for regression
    Richard, G., de Mathelin, A., Hébrail, G., Mougeot, M. and Vayatis, N.
    European Conference in Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), 2020.
    PDF Code BibTex

Softwares

  1. ADAPT Awesome Domain Adaptation Python Toolbox
    de Mathelin, A., Atiq, M., Richard, G., de la Concha, A., Yachouti, M., Deheeger, F., Mougeot, M. and Vayatis, N. (2021)
    ADAPT is an open source library providing numerous tools to perform Transfer Learning and Domain Adaptation. The library is specifically designed for Scikit-learn and Tensorflow users with a "user-friendly" approach.
    PDF Code BibTex
    GitHub stars GitHub forks GitHub contributors Programming language PyPI version PyPI downloads

  2. SKADA Domain Adaptation with scikit-learn and PyTorch
    Gnassounou T., Kachaiev O., Flamary R., Collas A., Lalou Y., de Mathelin A., Gramfort A., Bueno R., Michel F., Mellot A., Loison V., Odonnat A., Moreau T. (2024).
    SKADA is a library for domain adaptation (DA) with a scikit-learn and PyTorch/skorch compatible API.
    Code BibTex
    GitHub stars GitHub forks GitHub contributors Programming language PyPI version PyPI downloads

  3. SKADA Bench Benchmarking Unsupervised Domain Adaptation Methods with Realistic Validation
    Lalou, Y., Gnassounou, T., Collas, A., de Mathelin, A., Kachaiev, O., Odonnat, A., Gramfort, A., Moreau, T. and Flamary, R. (2024).
    SKADA Bench proposes a framework to evaluate DA methods based on a fair evaluation of existing shallow algorithms, including reweighting, mapping, and subspace alignment. Realistic hyperparameter selection is performed with nested crossvalidation and various unsupervised model selection scores.
    PDF Code BibTex
    GitHub stars GitHub forks GitHub contributors Programming language

Invited Talks

Services

Conference Reviewers

Journal Reviewers