Yazid Janati
I am currently a postdoctoral research at CMAP, Ecole polytechnique working with Eric Moulines and Alain Durmus. Previously I was a PhD candidate in Statistics at Télécom SudParis - Institut Polytechnique de Paris, advised by Sylvain Le Corff (LPSM, Sorbonne Université) and Yohan Petetin (CITI, Télécom SudParis). My PhD work focused on building new algorithms related to Monte Carlo methods and studying their theoretical properties. I was and still am particularly interested in the interplay between MC and deep learning methods. At the moment I work on solving inverse problems with Denoising Diffusion models.
Research
A Mixture-Based Framework for Guiding Diffusion Models Y. Janati, B. Moufad, M. Abou El Qassime, A. Durmus, E. Moulines, J. Olsson. Under review.
Variational Diffusion Posterior Sampling with Midpoint Guidance B. Moufad, Y. Janati, L. Bedin, A. Durmus, R. Douc, E. Moulines, J. Olsson. International Conference on Learning Representations (ICLR). 2025. Oral, top 1.8%.
Divide-and-Conquer posterior sampling with Denoising Diffusion priors Y. Janati, B. Moufad, A. Durmus, E. Moulines, J. Olsson. Advances in Neural Information Processing Systems (NeurIPS). 2024.
Entropic Mirror Monte Carlo Y. Janati, A. Durmus, S. Le Corff, Y. Petetin, J. Stoehr. Under review.
Monte Carlo guided Denoising Diffusion models for Bayesian linear inverse problems G. Cardoso, Y. Janati, S. Le Corff, E. Moulines. International Conference on Learning Representations (ICLR). 2024. Oral, top 1.2%.
State and parameter learning with PaRISian Particle Gibbs G. Cardoso, Y. Janati, S. Le Corff, E. Moulines and J. Olsson. International Conference on Machine Learning (ICML). 2023.
Variance estimation for Sequential Monte Carlo Algorithms: a backward sampling approach Y. Janati, S. Le Corff and Y. Petetin. Bernoulli. 2024.
NEO: Non Equilibrium Sampling on the Orbit of a Deterministic Transform A. Thin, Y. Janati, S. Le Corff, C. Ollion, A. Doucet, A. Durmus, E. Moulines and C. Robert. Advances in Neural Information Processing Systems (NeurIPS). 2021.
Structured variational Bayesian inference for Gaussian state-space models with regime switching. Y. Petetin, Y. Janati and F. Desbouvries. IEEE Signal Processing Letters 28. 2021.