Jana Zeller

Hi! I'm an ELLIS PhD student at the ETH Max Planck Center for Learning Systems, where I'm fortunate to be supervised by Wieland Brendel and Ryan Cotterell.

Previously, I earned my M.Sc. in Advanced Computer Science from Oxford, supported by a Google DeepMind Scholarship. During my time in Oxford, I worked on visual prompting with Suny Shtedritski and Chrisitan Rupprecht. Before that, I completed my Bachelor at Karlsruhe Institute of Technology. I also spent an exciting summer at Carnegie Mellon University as part of the Robotics Institute Summer Scholars program, exploring efficient architectures for videos with Jonathon Luiten and Deva Ramanan.

Besides research, I enjoy theatre, board games, and baking 🧁
(I am always happy to trade baking recipes and on the hunt for somebody to play set with :D )

Email / CV / LinkedIn / Github / Scholar

Research

I'm interested in computer vision, natural language processing, and how we can combine both to build more intelligent systems, that show similar data efficiency to humans.

	LittleLearner: Language Models Under Pedagogically-Controlled Knowledge Exposure Fanfei Li, Jana Zeller, Manuel de Prada Corral, Thaddäus Wiedemer, Prasanna Mayilvahanan, Ryan Cotterell, Wieland Brendel NeurIPS 2026, Oral* (top 0.4% of submissions)* project page / arXiv / models Trained an LLM with the knowlede and capabities of a 5th grader. Studied how post-training, in-context learning, and scaling effect performance on out-of-scope tasks. Invited Talks: Mrinmaya Sachan (ETH) reading group, Flourish AI
	MentisOculi: Revealing the Limits of Reasoning with Mental Imagery Jana Zeller, Thaddäus Wiedemer, Fanfei Li, Thomas Klein, Prasanna Mayilvahanan, Matthias Bethge, Felix Wichmann, Ryan Cotterell, Wieland Brendel ICML 2026 project page / arXiv / code Can models reason better with intermediate visualizations, akin to human mental imagery? MentisOculi evaluates this across frontier models and finds that visual thoughts do not yet improve reasoning. Invited Talk: Cohere Multimodal Reading Group
	Highlight: Learning Visual Prompts for Vision-Language Models Jana Zeller, Aleksandar (Suny) Shtedritski, Christian Rupprecht, CVPR (Emergent Visual Abilities and Limits of Foundation Models), 2025 project page / code We automatically discover visual prompts for CLIP. Interestingly, they look like red circles.
	Treating Video as an Image Jana Zeller, Deva Ramanan, Jonathon Luiten Preprint, 2022 preprint / poster / video We speedup video processing by merging frames together and passing them through the same vision model only once.

This website template is borrowed from Jon Barron. Thanks!