AI Research Scientist · EURECOM
Mustapha
Bounoua
Research scientist building multimodal generative models that learn structure across complex, heterogeneous data. My work spans vision, language and scientific data.
Research interests
Generative modelsMultimodal learningRepresentation learningInformation theoryAlignmentAI for science
News
16 updates-
DRIFT: Disentangled Responsive-Invariant Flow Transport for Single-Cell Perturbation Prediction accepted at the GDDL workshop, NeurIPS 2026.
-
Improved Sampling Schedules for Discrete Diffusion Models accepted at the PRIGM workshop, NeurIPS 2026.
-
New preprint: DRIFT: Disentangled Responsive-Invariant Flow Transport for Single-Cell Perturbation Prediction.
-
New preprint: DIPHINE: Diffusion-based Φ-ID Neural Estimator.
-
New preprint: Improved Sampling Schedules for Discrete Diffusion Models.
-
TENDE: Transfer Entropy Neural Diffusion Estimation accepted at AISTATS 2026. paper
-
PhD thesis online: Harnessing Multimodality: Diffusion-Based Generative Modeling and Information Estimation.
-
Learning to Match Unpaired Data with Minimum Entropy Coupling accepted at ICML 2025. paper
-
New preprint: Learning to Match Unpaired Data with Minimum Entropy Coupling.
-
SΩI accepted at ICML 2024 as an oral presentation (top 1.5%). paper
-
Multi-modal Latent Diffusion published in Entropy, special issue on Deep Generative Modeling: Theory and Applications. paper
-
New preprint: SΩI: Score-based O-Information Estimation.
-
MINDE accepted at ICLR 2024. paper
-
Masked Multi-time Diffusion for Multi-modal Generative Modeling accepted at the NeurIPS 2023 Workshop on Diffusion Models. paper, poster
-
New preprint: MINDE: Mutual Information Neural Diffusion Estimation.
-
New preprint: Multi-modal Latent Diffusion.
Selected publications
* equal contribution
NeurIPS 2026 · GDDL workshop
DRIFT: Disentangled Responsive-Invariant Flow Transport for Single-Cell Perturbation Prediction
M. Bounoua, G. Franzese, P. Michiardi
AISTATS 2026
TENDE: Transfer Entropy Neural Diffusion Estimation
S. P. G. Munoz, M. Bounoua, G. Franzese, P. Michiardi, M. Filippone
NeurIPS 2026 · PRIGM workshop
Improved Sampling Schedules for Discrete Diffusion Models
A. Foresti*, M. Bounoua*, G. Franzese, L. Ambrogioni, P. Michiardi
ICLR 2024
MINDE: Mutual Information Neural Diffusion Estimation
G. Franzese, M. Bounoua, P. Michiardi
SIA-Vision 2024
Enhancing Sensor Robustness in Automotive Systems: A Multimodal Generative Approach
M. Bounoua, C. Beaugeant, G. Franzese, P. Michiardi
PhD thesis · Sorbonne University · 2025
Harnessing Multimodality: Diffusion-Based Generative Modeling and Information Estimation
Supervised by Prof. Pietro Michiardi, EURECOM, in partnership with Renault Group (CIFRE). Read the manuscript
Experience
-
2025–nowEURECOM · AI Research Scientist
-
2022–25Renault Group · AI Research Scientist
-
2021–22Orange · Data Scientist
-
2020Huawei · AI Research Scientist
Education
-
2022–25Sorbonne University · PhD in Artificial Intelligence and Machine Learning
-
2019–20Paris Descartes University · MSc in Computer Science, Data Science
-
2014–19ESI · Engineering degree in Computer Science
-
TalksICML 2024 oral, Vienna · SIA Vision, Paris, invited talk, 2024
-
ReviewingICLR, ICML, NeurIPS, AAAI