
Luca M. Schulze Buschoff
I am a fourth year PhD student in cognitive science and machine learning under the supervision of Eric Schulz in the Institute for Human-Centered AI at Helmholtz AI, Munich Germany.
My research asks whether machines perceive and reason about the world the way people do — and how to close the gap where they do not. I use vision and vision–language models both as instruments for understanding human visual cognition and as systems to be measured against it, working toward a single model that captures human visual perception and cognition across multiple levels.
- Post-training makes large language models less human-like↗
M. Binz, ..., L. M. Schulze Buschoff, ..., E. Schulz. Preprint, 2026. - Can vision language models learn intuitive physics from interaction?↗
L. M. Schulze Buschoff, K. Voudouris, C. Demircan, E. Schulz. ICML, 2026. - A foundation model to predict and capture human cognition↗
M. Binz, ..., L. M. Schulze Buschoff, ..., E. Schulz. Nature, 2025. - Testing the Limits of Fine-Tuning for Improving Visual Cognition in Vision Language Models↗
L. M. Schulze Buschoff*, K. Voudouris*, E. Akata, M. Bethge, J. B. Tenenbaum, E. Schulz. ICML, 2025. - Visual cognition in multimodal large language models↗
L. M. Schulze Buschoff*, E. Akata*, M. Bethge, E. Schulz. Nature Machine Intelligence, 2025. - metabench -- A Sparse Benchmark to Measure General Ability in Large Language Models↗
A. Kipnis, K. Voudouris, L. M. Schulze Buschoff, E. Schulz. ICLR, 2025. - Next state prediction gives rise to entangled, yet compositional representations of objects↗
T. Saanum, L. M. Schulze Buschoff, P. Dayan, E. Schulz. Preprint, 2024. - The Acquisition of Physical Knowledge in Generative Neural Networks↗
L. M. Schulze Buschoff, E. Schulz, M. Binz. ICML, 2023. - Trivial or Impossible—dichotomous data difficulty masks model differences (on ImageNet and beyond)↗
K. Meding*, L. M. Schulze Buschoff*, R. Geirhos, F. A. Wichmann. ICLR, 2022. - ImageNet suffers from dichotomous data difficulty↗
K. Meding*, L. M. Schulze Buschoff*, R. Geirhos, F. A. Wichmann. NeurIPS Workshop, 2021.