Maxime Ronceray

Variational inference for data-driven galaxy population priors with differentiable stellar-population synthesis.

• Develop amortized posteriors for 15 physical galaxy parameters.
• Combine DSPS, normalizing flows and reweighted wake-sleep training.
• Study scalable population-level inference for next-generation imaging surveys.

Benchmarking foundation models for unsupervised discovery in matched Euclid imaging and DESI spectroscopy.

• Compare AstroPT, AstroCLIP and AION multimodal representations.
• Build scalable anomaly scores with normalizing flows and cross-modal alignment.
• Paper accepted at the ICLR 2026 FM4Science workshop.

Remote sensing of dissolved organic carbon in Arctic rivers with scalable machine-learning workflows.

• Evaluated mixture-of-experts approaches for remote-sensing data.
• Parallelized computation on HPC resources and transferred the workflow to domain scientists.

Developed PLASMAG, a simulation environment for search-coil magnetic sensors used in space instrumentation.

• Designed a dependency-driven simulation engine and scientific interface.
• Integrated SPICE simulations and sensor-parameter optimization.
• Benchmarked the calculation engine and presented the project to researchers.

Built an interface to automate space-instrument testing and measurement acquisition.

• Controlled oscilloscope hardware through an SDK.
• Automated measurements, signal processing and analysis.

Machine learning, deep learning, statistics, probability, data engineering and advanced data structures.

• Research track focused on scientific machine learning and astronomical applications.
• Student representative for the University of Tours.

Software development, project management and an artificial-intelligence specialization.

• Elected student representative to the institute board of directors.

Software engineering, physics, linear algebra, calculus and scientific simulation.

• Developed Python and MATLAB simulations and a team Qt application.