BrainWideBench Finds Mixed Cross-Animal Transfer in Large-Scale Neural Data, with Pretraining Gains Vary by Task
A preprint benchmark reveals that learning representations across many mice and brain regions helps some tasks but not all, highlighting limits of current methods.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the abstract
- Authors
- Alexandre Andre, Shivashriganesh P. Mahato, Vinam Arora, Keshav Balaji, Divyansha Lachi, Nanda H. Krishna, Jingyun Xiao, Yizi Zhang, Ximeng Mao, Wenrui Ma, Han Yu, International Brain Laboratory, Daniel Birman, Niccolò Bonacchi, Gaelle A. Chapuis, Joana A. Catarino, Felicia Davatolhagh, Mayo Faulkner, Laura Freitas-Silva, Fei Hu, Julia M. Huntenburg, Anup Khanal, Inês Laranjeira, Petrina Lau, Guido T. Meijer, Nathaniel J. Miska, Jean-Paul Noel, Alejandro Pan-Vazquez, Georg Raiser, Cyrille Rossant, Karolina Z. Socha, Anne E. Urai, Miles J. Wells, Steven J. West, Olivier Winter, Blake Richards, Guillaume Lajoie, Cole Hurwitz, Mehdi Azabou, Matthew R. Whiteway, Liam Paninski, Eva L. Dyer
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
What the paper reports
Researchers introduced BrainWideBench to evaluate across-animal transfer using the International Brain Laboratory dataset (276 brain regions, 139 mice) across three task suites. They found pretraining generally improves performance versus single-session baselines, but transfer gains are uneven and depend on how well pretraining objectives align with downstream tasks.
Why it matters
The study cautions against assuming universal transfer from large-scale neural data. It shows mixed gains, underscoring the need to tailor pretraining and evaluation to specific biological and behavioral goals.
BrainWideBench assembles a unified, reproducible framework to test whether learned neural representations can transfer across animals and tasks. It uses multi-region neural recordings and behavioral data to probe decoding, predicts neural dynamics, and recovers anatomical organization, offering a lens on generalization limits.
A key takeaway is that no single pretraining approach dominates all scenarios; performance depends on task alignment, geometry of neural signals, and the downstream objective, pointing to a nuanced path for general-purpose models of brain activity.
What this does not tell us
Abstract-only preprint; abstract-only scope applies. No final peer-reviewed validation reported.
Original sources · 1
- BrainWideBench: Benchmarking large-scale pretraining and across-animal transfer in multi-region neural recordings ↗arXiv · 2026-09-18
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.