Physics-informed generative models for Mueller polarimetric imaging
At CMAP, École Polytechnique, I work on deep learning for Mueller polarimetric imaging with applications to medical diagnosis. The research studies how generative architectures can model complex imaging data while respecting mathematical and physical structure.
Core directions
- Deep generative architectures for Mueller matrix modeling.
- Synthetic data generation for low-data clinical imaging.
- Physical constraints such as positivity and Mueller matrix realizability integrated into neural architectures.
- Uncertainty-aware learning with geometry-aware label smoothing.
- Interpretable and physically consistent representation learning for biomedical imaging.
Generative AI
I am particularly interested in variational autoencoders (VAEs), diffusion models and representation learning. A recurring question in my work is how latent spaces can encode meaningful scientific structure rather than only optimize a reconstruction or generation objective.
Scientific machine learning
My broader interests include Physics-Informed Machine Learning, Bayesian Deep Learning, uncertainty quantification, computer vision, explainable AI and probabilistic modeling. I enjoy approaches that connect theoretical machine learning with real scientific constraints and measurable physical quantities.
Explore the code
Selected public research repositories are available on the projects page and on GitHub @ayouboa30. The home page also contains Latent Quest, a playable pixel-art research game about latent spaces, diffusion, gradient descent and classification.