Bernhard Kainz

Achieving robust ultra-long-term time series forecasting, Diffusion-LLM outperforms traditional LLMs by leveraging distribution-aware regularization for enhanced generalization.

VLMs can achieve state-of-the-art segmentation in histopathology without any manual annotations, outperforming traditional models on challenging stains.

Counterintuitively, moderately similar reference images are the key to unlocking accurate VLM-based anomaly localization in medical imaging.

Ditch fixed-size 3D blocks: SigVLP uses rotary embeddings to let vision-language models handle CT volumes with variable slice counts, unlocking better pre-training.