Nassim User

At BioMaps, CEA, and Université Paris-Saclay, I'm developing a memory-optimized 3D CycleGAN for thoracic CT-to-UTE-MRI translation and a reproducible medical-imaging pipeline using MONAI, PyTorch, TensorBoard, and MLFlow. My Hist-CycleGAN loss improved MRI realism from FID 225.96 to 217.89 and KID 0.1158 to 0.0969 while preserving anatomical consistency.

I've also improved constrained neural networks for phylogenetics at the Collège de France, reducing constraint violations from 15% to under 1%, and built projects in neural-network verification, speech classification, and YOLOv1 benchmarking. I bring strong mathematical foundations to deep learning, computer vision, generative modeling, and LLM work, with an emphasis on rigor, reproducibility, and measurable impact.