TreasureX Research

New Benchmark Tests Whether AI Models Can Transfer Brain Knowledge Across Animals

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 · September 18, 2026

Neuroscience has entered a big-data era: researchers can now record activity from hundreds of thousands of neurons across many brain regions and many animals at once. The hope is that, as in language models, scale alone could produce general-purpose representations of brain activity that transfer to new tasks and new subjects. A new study introduces BrainWideBench, a benchmark built on the International Brain Laboratory's Brainwide Map — recordings from 276 brain regions in 139 mice performing a sensory decision task.

The benchmark tests learned representations on three fronts: decoding behavior, predicting masked or future neural activity, and recovering anatomical structure. The team evaluated many pretraining methods under both finetuning and zero-shot transfer to unseen animals.

Pretraining clearly beat single-session baselines, but no method won everywhere. Gains depended heavily on how well a pretraining objective matched the downstream task, and most methods were built for only one of the three suites. Jointly generalizing across behavior, dynamics, and anatomy remains unsolved.

Why it matters: this is infrastructure for AI-driven brain research, similar to benchmarks that accelerated computer vision and language models. Better brain models could eventually inform neurotechnology, drug development, and clinical diagnostics — though no listed company is named here.

Source: arXiv paper https://arxiv.org/abs/2609.22064

Educational content only, generated with AI assistance from public research abstracts. Not investment advice. Thank you to arXiv for use of its open access interoperability.