Portrait of Maxime Ronceray

Maxime RONCERAY

Astrophysics - Data Science

I build machine-learning tools to find unusual galaxies and infer their physical properties from large astronomical surveys.

View CVGitHub
  1. Nov 2026 – 2029 — PhD in astrophysics at CEA: Probabilistic machine learning for galaxy surveys and next-generation cosmology.
  2. Mar 2026 – Sep 2026 — Galaxy population inference at CEA: Learning galaxy-population priors and fast physical posteriors with differentiable models and variational inference.
  3. Sep 2025 – Apr 2026 — Foundation model for anomaly detection: Benchmarking AstroPT, AstroCLIP and AION to discover rare objects in matched Euclid images and DESI spectra.
  4. Sep 2024 – Sep 2026 — MSc Data Science & AI: MSc in Data Science and AI, focused on machine learning, statistics and scientific computing.

Loading the universe

Redshifting galaxies...

Publications

Papers

Euclid galaxy cutouts paired with DESI spectra for four anomaly candidatesEuclid × DESI · anomaly candidates

Featured paper · 2026

Benchmarking Foundation Models for Unsupervised Discovery in Large Multimodal Astrophysical Datasets

M. Ronceray, M. Huertas-Company, A. Chanson, M. Siudek, A. Preto, M. J. Smith, J. Zoubian, C. Bonini · ICLR 2026 Workshop on Foundation Models for Science (FM4Science)

We compare AstroPT, AstroCLIP and AION representations for unsupervised discovery in matched Euclid imaging and DESI spectroscopy. A scalable density-estimation pipeline combines per-modality rarity with cross-modal misalignment to reveal model-dependent rankings, instrumental artefacts and physically coherent candidates including AGN and strong gravitational lenses.

    Conferences

    Posters

    • Poster presenting the architecture and preliminary results for data-driven galaxy population priors01

      Poster · 2026

      Variational Inference for Data-Driven Galaxy Population Priors

      CEA · Université Paris-Saclay

      A differentiable inference pipeline that jointly learns a galaxy-population prior and fast per-galaxy posteriors from catalogue photometry using DSPS and reweighted wake-sleep.

      Open full poster
    • Poster about anomaly detection in Euclid Q1 with foundation-model embeddings02

      Poster · 2025

      Anomaly Detection in the Euclid Q1 Data Release using Foundation Model Embeddings

      Data Science Day · University of Tours

      An anomaly-ranking study across AstroPT, AION and AstroCLIP embeddings, combining Euclid images with DESI spectra to surface rare systems and instrumental artefacts.

      Open full poster

    Talks

    Slides

    • Latent-space map and nearest-neighbour examples from the foundation-model benchmark

      Benchmarking Foundation Models for Unsupervised Discovery in Large Multimodal Astrophysical Datasets

      Research talk · Blois · 12 Feb 2026

      View slides
    • Opening slide of the Euclid foundation-model anomaly-detection presentation

      Anomaly Detection on Euclid Q1 using Foundation Model Embeddings

      University of Tours · 18 Nov 2025

      View slides
    • Opening slide of the PLASMAG space magnetic-sensor simulation presentation

      Development of a Simulation Environment for Space Magnetic Sensors: PLASMAG

      Café Spatial · LPP / CNRS · 5 Jul 2024

      View slides

    Documents

    Reports

    • PLASMAG logo: Python library for accurate search-coil magnetometer adjustments with a graphical interface

      Technical report · 2024

      Development of a Simulation Environment for Space Magnetic Sensors

      LPP · CNRS / École Polytechnique internship

      Design and implementation of PLASMAG, a modular simulation environment for search-coil magnetometers, including a dependency-driven simulation engine, SPICE integration and sensor-parameter optimization.

      Read full report

    Research focus

    About

    Incoming PhD student at CEA · Astrophysics & machine learning

    I am completing an MSc in Data Science and Artificial Intelligence at the University of Tours. My research focuses on reusable machine-learning representations and probabilistic methods for scientific discovery in large astronomical surveys.

    I have worked on multimodal anomaly detection with Euclid and DESI data, variational inference for galaxy population priors at CEA, and scientific simulation software for space instrumentation at LPP. My foundation-model benchmarking work was accepted at the ICLR 2026 FM4Science workshop.

    Research interests

    • Scientific foundation models
    • Multimodal anomaly detection
    • Simulation-based inference
    • Galaxy populations and SEDs
    • Large astronomical surveys

    Tools & methods

    • Python · JAX · PyTorch
    • Probabilistic inference · Normalizing flows
    • Deep learning · Multimodal transformers
    • HPC · Scientific computing
    • C/C++ · Data engineering

    Elsewhere

    CV & links

    GitHubCode, pipelines & experimentsView interactive CVDownload PDF