Biosketch
Inés Rosellón-Inclán is a doctoral researcher at the Chair for Mathematical Foundations of Artificial Intelligence working at the intersection of applied mathematics, deep learning, and imaging science. Her research focuses on developing mathematically grounded methods for explainability and hallucination detection in deep learning–based image processing, with applications in medicine and astronomy.
Inés holds a Bachelor’s degree in Mathematics and a Master’s degree in Economathematics from Karlsruhe Institute of Technology (KIT). During her studies, she gained valuable experience as a research assistant at Fraunhofer ISI and Fraunhofer IOSB.
relAI Research
Reliable Deep Learning for Image Processing
Deep learning has transformed image processing, enabling powerful tools for reconstruction, segmentation, and classification across domains. However, these models can introduce realistic-looking artifacts in reconstructed images, and their predictions often lack interpretable justification, limiting their trustworthiness in high-stakes applications. We develop mathematically grounded frameworks to address these challenges, spanning hallucination estimation and explainability methods for deep learning–based image processing. Our hallucination metric (CHEM) is a distribution-free, model-agnostic metric that identifies high-risk hallucination regions in images reconstructed by a given model (arXiv:2512.09806). Furthermore, we explore adaptable explanation masks for segmentation and classification models to support transparent and reliable AI-assisted decision-making in clinical settings.