Timothy
Schaumlöffel
PhD student at Goethe University Frankfurt working on mechanistic interpretability and evaluation of multimodal foundation models.
About
I am a 3rd-year PhD student at the CVAI Lab, Goethe University Frankfurt, working under Prof. Gemma Roig. My research is centred on the mechanistic interpretability and evaluation of multimodal foundation models: analysing internal representations and emergent behaviours in vision-language systems through causal probing, representation analysis, and controlled ablations.
Alongside this, I develop cognitively inspired self-supervised learning methods that draw on developmental science to study how meaningful structure emerges from temporal experience. I'm interested in the dialogue between how brains and models build representations of the visual world.
My work has appeared at CVPR, NeurIPS, and ICLR. I am also a core contributor to Net2Brain, an open-source toolbox for comparing artificial vision models with human brain responses.
Selected papers, grouped by .
Loading publications…
Tools, code, and open science.
An open-source analysis framework for comparing vision-model representations with neural data. Adopted by many research institutions and published in Frontiers in Neuroinformatics.
cvai-roig-lab/Net2BrainTraining and validation code, and analyses behind our work on how temporal slowness in central vision drives semantic object learning.
t9s9/central-vision-sslCode and analyses behind our work on the internal mechanisms of object localization in vision-language models such as CLIP, LLaVA, and InternVL.
t9s9/vlm-object-localization