There is a growing interest in developing foundation models for neural data, i.e. models that are pretrained on large-scale, diverse datasets and which can then be fine-tuned for a variety of downstream tasks with relatively little data. This approach has been extremely powerful in AI research more broadly, and will likely help unlock many capabilities for neuro-technology, particularly for brain-machine interface applications and the development of digital twins as powerful discovery platforms for neuroscience research. Prior work on neuro-foundation models from our groups (Paninski, Richards, & Tolias) has provided demonstrations of self-supervised pretraining for spiking data (Azabou et al 2023; Zhang et al 2025a&b), encoding model pretraining for visual response spiking data (Wang et al 2024), and multi-region and multi-cell type pretraining for spiking and calcium imaging data (Yu et al 2025 & Azabou et al 2025a). This work has provided the proof-of-principle that pretraining on diverse data can indeed provide a foundation for efficient finetuning on a variety of downstream neural encoding and decoding tasks.

However, we are lacking well-organized frameworks for training and benchmarks for assessing neuro-foundation models, and relatively little work has been done on leveraging combined sensory and behavioural data to improve pretraining at scale. The goal of this project is to address this gap by providing the community with: (1) An integrated code base (TorchBrain; Azabou et al., 2025a) and baseline models for comparison within this framework; (2) Novel benchmark datasets to push forward performance on both sensory and behavioral data; and (3) A demonstration that pretraining on both sensory and behavioural data leads to better downstream performance on a variety of encoding and
decoding tasks. This will help the neuro-AI community to develop well-grounded, sophisticated neuro-foundation models, which will be a critical component of ARNI’s mission to pursue use-inspired work accelerating progress in both neuroscience and AI and to broaden their transformative impact on society.