PhD Position: Robust Few-Shot Learning for Foundation Models in CT Imaging

This three-year doctoral position is offered within the OPIS team at the Inria Saclay–Île-de-France Research Centre, located on the CentraleSupélec campus. The project is part of the ANR FM2AI initiative, which leverages 50,000 real-world clinical 3D CT scans from the Greater Paris University Hospitals data warehouse to build a foundation model for abdominal-pelvic CT imaging. The PhD candidate will design novel few-shot learning techniques enabling on-site adaptation of this foundation model to specific diagnostic tasks such as classification and tumor segmentation, with a strong emphasis on robustness and generalization assessment. The work draws on the OPIS team's established expertise in few-shot learning and mathematical optimization. The ideal candidate will combine a solid theoretical foundation in optimization and statistics with practical deep-learning programming experience.

• Develop novel few-shot learning techniques for CT image classification
• Design and implement models for few-shot tumor segmentation
• Analyze robustness and generalization capabilities of the proposed models
• Validate approaches on both public benchmark datasets and EDS AP-HP clinical datasets
• Conduct bibliographical studies to stay current with relevant literature
• Design, train, and test deep learning architectures in a Python/PyTorch environment
• Perform mathematical optimization and convergence analysis of proposed methods
• Write scientific reports and contribute to research publications
• Collaborate regularly with members of the ANR FM2AI consortium