Reliable ML and OOD detection
Detection and calibration methods for models deployed under distribution shift, low SNR, heterogeneous clutter, and adversarial or operational stress.
PhD Researcher · Machine Learning · Radar AI
Reliable machine learning for robust detection, generative models, and uncertainty-aware AI.
I am a PhD researcher at CentraleSupélec / Université Paris-Saclay, conducting research within ONERA / SONDRA, a France–Singapore research alliance involving CentraleSupélec, ONERA, NUS, and DSO National Laboratories.
My work focuses on robust radar target detection, out-of-distribution detection, generative modeling, diffusion models, complex-valued learning, and uncertainty-aware evaluation. I am especially interested in AI systems that need to remain reliable under noise, clutter, distribution shift, and operational constraints.
OOD Detection
Diffusion Models
Radar Signal Processing
Complex-Valued Learning
AI Safety
Scientific Software
PhD researcher in robust radar target detection and reliable machine learning.
About
My research combines statistical signal processing and modern deep learning. I work on methods that connect classical radar detection principles—such as CFAR calibration, whitening, robust covariance estimation, and detector fusion—with learned generative representations such as VAEs, complex-valued VAEs, diffusion models, and flow-based models.
Before my PhD, I worked at PHIMECA on uncertainty quantification, reliability studies, Bayesian correction, and scientific software engineering. That experience shaped how I write research code: reproducible experiments, documentation, testing, packaging, CI/CD, and deployment-oriented workflows.
I am currently interested in roles involving AI research, ML engineering, defense AI, AI assurance, and quantitative ML, especially when rigorous experimentation and mathematically grounded modeling are central to the team culture.
Research interests
Generative modeling
VAEs, complex-valued VAEs, diffusion models, rectified flow matching, latent-space scoring, and interpretable anomaly localization.
Radar signal processing
Radar target detection in compound clutter and thermal noise, robust whitening, CFAR calibration, off-grid detection, and fusion with classical detectors.
AI safety and assurance
Operational-domain definition, uncertainty calibration, failure-mode analysis, robustness evaluation, and causal experimental design for high-stakes AI systems.
Scientific software
Reproducible pipelines, ablations, benchmark design, documentation, testing, CI/CD, packaging, and usable research code.
Quantitative ML
Signal extraction, robust validation, market microstructure, short-horizon alpha experimentation, transaction costs, and drift-aware evaluation.
Projects
Selected research-engineering projects that complement my publication record and make the profile readable for quant research, AI research, deep-tech, and industrial ML teams.
Diffusion OOD and backbone-equated evaluation
Research line around diffusion backbones, sparse internal features, equivariance-breaking signals, and fair OOD evaluation protocols under matched backbone and compute budgets.
MicroAlpha
A quantitative ML research platform for market microstructure, execution benchmarking, transaction-cost-aware evaluation, and short-horizon alpha experiments.
Radar generative detection
VAE, complex-valued VAE, SVDD, flow matching, and diffusion-inspired methods for radar detection under compound clutter and thermal noise.
Scientific software and reliability
Reusable research code, testing, documentation, CI/CD, packaging, and reproducible experimental workflows shaped by PHIMECA experience in uncertainty quantification and scientific computing.
Publications and preprints
Selected papers in robust radar detection, generative modeling, diffusion-based OOD detection, and signal processing. Public PDF and arXiv links are listed when available.
2026
Backbone-Equated Diffusion OOD via Sparse Internal Snapshots
Y. A. Rouzoumka et al. — submitted to NeurIPS 2026
Submitted
Preprint coming soon
GEPC: Group-Equivariant Posterior Consistency for Out-of-Distribution Detection in Diffusion Models
Y. A. Rouzoumka, J. Pinsolle, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — ICML 2026
Radar Detection through Rectified Flow Matching
P. Meena, Y. A. Rouzoumka, J. Pinsolle, C. Ren, M. N. El Korso, J.-P. Ovarlez — submitted to EUSIPCO 2026 · arXiv preprint
Out-of-Distribution Radar Detection with Complex VAEs: Theory, Whitening, and ANMF Fusion
Y. A. Rouzoumka, J. Pinsolle, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — submitted to IEEE Transactions on Signal Processing · arXiv preprint
DopplerGLRTNet for Radar Off-Grid Detection
Y. A. Rouzoumka, J. Pinsolle, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — submitted to EUSIPCO 2026 · arXiv preprint
Support Vector Data Description for Radar Target Detection
J. Pinsolle, Y. A. Rouzoumka, C. Ren, C. Morisseau, J.-P. Ovarlez — ICASSP 2026
Latent-Space Metrics for Complex-Valued VAE Out-of-Distribution Detection under Radar Clutter
Y. A. Rouzoumka, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — ICASSP 2026
Autoencodeur variationnel complexe pour la détection de cibles radar
Y. A. Rouzoumka, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — JDS 2026
Public PDF coming soon
2025
Out-of-Distribution Radar Detection in Compound Clutter and Thermal Noise through Variational Autoencoders
Y. A. Rouzoumka, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — ICASSP 2025
Complex-Valued Variational Autoencoders for Radar Detection in Joint Compound Gaussian Clutter and Thermal Noise
Y. A. Rouzoumka, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — EUSIPCO 2025
Détection de cibles radar par auto-encodeur variationnel
Y. A. Rouzoumka, E. Terreaux, C. Morisseau, J.-P. Ovarlez, C. Ren — GRETSI 2025
Experience
PhD Researcher — Machine Learning for Robust Radar Target Detection
2023–presentCentraleSupélec · Université Paris-Saclay · ONERA / SONDRA
- • Research on robust radar detection using VAEs, complex-valued deep learning, diffusion-based methods, and OOD scoring.
- • Evaluation protocols under compound clutter, thermal noise, nuisance mismatch, and operational false-alarm constraints.
- • Reproducible Python / PyTorch pipelines for benchmarking, ablations, and statistical analysis.
R&D Engineer / Scientific Software — PHIMECA
2022–2023Uncertainty quantification, reliability, Bayesian correction, and scientific software engineering.
- • Developed reusable Python modules and workflows for scientific computing and measurement-error correction.
- • Contributed to testing, packaging, Sphinx documentation, GitLab CI, Docker, and executable delivery.
- • Built habits directly relevant to AI assurance: reproducibility, QA, uncertainty-aware evaluation, and robust engineering practices.
Teaching Assistant — IUT Computer Science
2024–presentUniversité Paris-Saclay. Teaching mathematics for signal processing, Java software development, and SQL/database fundamentals.
Contact
I am open to research and engineering conversations in Alpha/Quant research, reliable ML, AI safety, defense AI, radar/signal processing, and quantitative ML.