Lukas Kuhn

I'm a research scientist at the MLO Lab lead by Prof. Dr. Florian Buettner in Frankfurt and a Masters student at Goethe University Frankfurt in AI & Computational Neuroscience.

Email  /  Scholar  /  Github

Research

I'm passionate about the intersection of AI and Neuroscience, trying to bridge the gap between the sample efficient & robust primate brain and sample hungry & brittle AI. Some papers are highlighted.

Small vision models in high-stakes domains often learn spurious correlations that don't align with human expertise, leading to unreliable real-world performance. LVLM-VA addresses this by using large vision-language models as a bridge between domain experts and task-specific models, translating human knowledge into actionable feedback that reduces reliance on spurious features without requiring fine-grained annotations.

By leveraging MLLMs and the representational structure of the Transformer architecture we are able to detect and mitigate shortcuts completely unsupervised.

Multi-agent architecture that generates interpretable reports explaining how much computer vision model performance likely degrades during deployment, discussing specific likely failure modes and identifying root causes and mitigation strategies.

Empirical analysis of vision foundation models that shows that these models tend to be underconfident in in-distribution predictions, resulting in higher calibration errors, while demonstrating improved calibration under distribution shifts.

We demonstrate that optimizing convolutional network architectures for brain-alignment via evolutionary neural architecture search results in models with clear representational hierarchies. The identified models achieve brain-alignment scores surpassing even those of pretrained classification models.

Website Design copied with permission from https://jonbarron.github.io/