One paper accepted at NeurIPS 2026 Main Conference
Our lab will be presenting one paper at the The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS) 2026 as a Spotlight:
- Entropy Minimization without Model Collapse: Mitigating Prediction Bias for Medical Imaging
Tim Nielen, Sameer Ambekar, Johannes Kiechle, Daniel M. Lang, Julia A. Schnabel
(https://arxiv.org/pdf/2606.02339)
TL;DR: We identify the root cause of model collapse: Entropy Minimization, which amplifies prediction bias and ultimately drives models into collapse. Building on this insight, we propose Distribution Shift Bias Reduction (DSBR), a method that effectively mitigates model collapse across both natural vision and medical imaging settings.
Julia A. Schnabel
Professor for Computational Imaging and AI in Medicine, Director of the Institute of Machine Learning in Biomedical Imaging
My research interests include machine/deep learning, nonlinear motion modeling, as well as multimodal and quantitative imaging, for cancer-, cardiac-, neuro- and perinatal imaging.
Sameer Ambekar
Doctoral Researcher
Current focus is to Adapt Vision-Language Models (VLMs) with reasoning abilities using GRPO (Group Refinement Policy Optimization) and test-time strategies for medical imaging.
Daniel M. Lang
Research Scientist
My current research focuses on the development of deep generative models for dynamic settings in cancer imaging.