Dual PhD Researcher in Machine Learning, Bayesian Imaging and Uncertainty Quantification
Mathematically grounded machine-learning researcher developing uncertainty-aware methods for reconstructing images from noisy, incomplete and indirect measurements. Research spans Bayesian inverse problems, self-supervised conformal prediction, posterior sampling and generative priors, with peer-reviewed work in Gaussian and Poisson imaging. Combines statistical modelling and optimisation with PyTorch/JAX implementation and GPU/HPC experimentation for scientific and medical imaging.
Bernardin works across applied mathematics, statistical machine learning and research engineering. His current work asks how imaging systems can produce both useful reconstructions and defensible uncertainty estimates when measurements are noisy, incomplete or indirect.
Core research areas
Bayesian inverse problems
Computational and quantitative imaging
Uncertainty quantification
Conformal prediction
Posterior inference and sampling
Self-supervised learning
Diffusion, flow and variational generative models
Learned and data-driven priors
Medical and scientific imaging
Optimisation, stochastic algorithms and kernel methods
Mathematical Modelling in Economics and Finance; Computational Finance · African Centre of Excellence in Information and Communication Technologies (CETIC), National Advanced School of Engineering, University of Yaoundé I
Yaoundé, Cameroon
Best student in Mathematical Modelling in Economics and Finance (2019)
Dissertation: Analysis of a Brownian Motion Functional.
Research in machine learning and Bayesian computation for uncertainty-aware imaging inverse problems.
Developing self-supervised uncertainty-quantification methods for ill-posed imaging problems where reliable ground truth is scarce or unavailable.
Co-developed SURE- and PURE-based conformal methods and currently studies symmetry-aware latent-variable models for scalable posterior sampling.
Implements and evaluates methods in Python with PyTorch/JAX, GPU acceleration and SLURM-based HPC workflows across denoising, deblurring, CT and MRI settings.
Research supervised by Andrés Almansa, Julie Delon and Marcelo Pereyra.
Image restoration problems are often ill-posed, leading to significant uncertainty in reconstructed images. Accurately quantifying this uncertainty is essential for the reliable interpretation of reconstructed images. This work develops a self-supervised conformal prediction method for Poisson imaging problems, leveraging a Poisson unbiased risk estimator to calibrate uncertainty without ground-truth images. The resulting self-calibrating conformal prediction approach is applicable to any Poisson linear imaging problem that is ill-conditioned, and is particularly effective when combined with modern self-supervised image restoration techniques trained directly on measurement data.
Self-supervised Conformal Prediction for Uncertainty Quantification in Imaging Problems
Most image restoration problems are ill-conditioned or ill-posed and hence involve significant uncertainty. This work proposes a self-supervised conformal prediction approach for uncertainty quantification in imaging inverse problems, using Stein’s unbiased risk estimator (SURE) to self-calibrate directly from noisy measurements, bypassing the need for ground truth data. The method is suitable for any linear imaging inverse problem that is ill-conditioned, and delivers results comparable to supervised conformal prediction with ground truth data.
Taught Statistical Models B (2025-2026) and Introduction to University Mathematics (2024-2025).
Led tutorials and laboratories across Bayesian inference, stochastic processes, probability and statistics, data science, numerical analysis, optimisation and scientific programming.
Supported marking, assessment delivery and student guidance across undergraduate and postgraduate courses.
