Overview
Mice navigate complex visual scenes with remarkable robustness, often outperforming artificial agents when visibility is degraded. The Mouse vs AI: Robust Foraging competition and workshop challenge the AI community to close this gap. Participants train agents in the same Unity-based VR task used for mouse experiments, testing both behavioral robustness under perturbations and neural alignment to recorded mouse visual cortex activity.
This NeurIPS workshop brings together researchers from reinforcement learning, computer vision, and systems neuroscience to explore how robust perception emerges and how artificial and biological systems can inform one another.
Logistics
Due to visa issues and travel constraints, the workshop will be held entirely on Zoom. A meeting link will be shared with all registered NeurIPS attendees shortly before the event.
For updates and discussion, join our Discord or email us at robustforaging@gmail.com.
Please note: this is an official NeurIPS 2025 event, so conference registration is required to attend.
Keynote Speakers
University of Tübingen
Foundation models for mouse vision
UC Santa Barbara
Geometric approaches to neural activity prediction
University of Oregon
Visual processing in freely moving mice
Schedule
| Time (PST) | Session | Speaker / Topic |
|---|---|---|
| 11:00 - 11:05 | Opening Remarks | Michael Beyeler |
| 11:05 - 11:30 | Mouse vs AI Challenge | The Competition: Goal, Tracks, Evaluation (Marius Schneider) The Mouse: Task, setup performance (Joe Canzano) The AI: leaderboard, model analysis (Yuchen Hou) |
| 11:30 - 12:00 | Keynote Talk 1 | Nina Miolane: Geometric approaches to neural activity prediction |
| 12:00 - 12:30 | Keynote Talk 2 | Fabian Sinz: Foundation models for mouse vision |
| 12:30 - 12:45 | Competition Talk | 3rd Place |
| 12:45 - 1:00 | Competition Talk | 2nd Place |
| 1:00 - 1:30 | Keynote Talk 3 | Cris Niell: Freely moving mice |
| 1:30 - 1:45 | Competition Talk | 1st Place |
| 1:45 - 1:50 | Closing Remarks | Michael Beyeler |