BrainSymphony Introduces Lightweight Multimodal Neuroscience Foundation Model

An arXiv preprint (v2 posted Feb 12, 2026) introduces BrainSymphony, a lightweight multimodal foundation model that integrates fMRI time series and diffusion-derived structural connectivity for unimodal or multimodal training and deployment. The architecture combines parallel spatial and temporal transformers, a Perceiver bottleneck, and a signed graph transformer with adaptive fusion, outperforming larger models on prediction, classification, and unsupervised network discovery benchmarks.