Bethge Lab

Bethge Lab is an AI research group based at the University of Tübingen in Germany. The lab focuses on neuro AI, aiming to understand how brains learn and to build machines that can learn in similarly open-ended, adaptive ways. Its mission is to develop agentic systems that acquire knowledge over time, mirroring human learning and scientific discovery. The group's work spans topics like open-ended model evaluation, language model agents, lifelong compositional learning, modeling brain representations, and attention in humans and machines.

The lab approaches these challenges through data-centric machine learning, building multi-modal foundation models that support rapid retrieval and compositional integration of knowledge. Researchers create digital twins of brain areas like the retina and visual cortex to understand neural computation. They also develop benchmarks and tools for infinite model evaluation, tackling issues like domain contamination and safety. A distinctive part of their work involves studying attention mechanisms in humans to improve machine vision and behavior models.

Bethge Lab's research benefits AI scientists, graduate students, and practitioners interested in advancing machine learning toward human-like adaptability. Their findings have practical applications in building reliable language agents, creating scalable lifelong learning systems, and developing transparent evaluation methods. The lab also contributes to society through initiatives like the federal AI competition BWKI and the IT4Kids programming education nonprofit.